Distributed Object Storage toward Storage and Usage of Packet Data in a High-speed Network

Size: px
Start display at page:

Download "Distributed Object Storage toward Storage and Usage of Packet Data in a High-speed Network"

Transcription

1 Distributed Object Storage toward Storage and Usage of Packet Data in a High-speed Network Masahisa Tamura, Ken Iizawa, Munenori Maeda, Jun Kato, Tatsuo Kumano, Yuji Nomura, Toshihiro Ozawa Fujitsu Laboratories Ltd. Kawasaki, Japan {masahisa, iizawa.ken, maeda.munenori, jun.kato, kumano_tatsuo, nomura.yuji, t.ozawa}@jp.fujitsu.com Abstract As the amount of data flowing in the network increases and as the number of applications having complex traffic patterns increases as well, packet data analysis is becoming all the more important for improving the quality of communications and facilitating troubleshooting. In particular, demand for the storage and analysis of individual items of packet data is increasing in addition to analysis based on statistical information to enable a detailed diagnosis of phenomena that have occurred on the network and at terminals. This paper proposes distributed object storage that can achieve both highthroughput storage of packet data and high-speed session searching. Specifically, we propose and evaluate a system for storing small-size data such as packet data with high throughput while deleting old data and conducting searches, and compare the storage performance of the proposed system with that of other distributed storage systems, and present the results of evaluating search performance for typical search patterns. Keywords packet capture; distributed object storage I. INTRODUCTION Recent years have seen increasing amounts of data flowing in the network as well as the appearance of problems caused by complex traffic patterns not envisioned at the network design stage. Packet data analysis has consequently been growing in importance as a means of solving these problems. In particular, there is a strong demand for techniques that go beyond analysis based on statistical information and sampled data by enabling the storing and analyzing of all packet data to obtain a detailed diagnosis of phenomena occurring on the network. Furthermore, it has been shown that some targeted attacks are skillfully mounted by making small, step-by-step preparations over the long term, so there is also a need to store data over an extended period of time so that long-term traffic patterns can also be analyzed. The storage of large amounts of packet data and the analysis of that data in parallel is required to analyze and communications quality and security problems. Such analysis often involves the extraction of sessions or calls included in groups of packet data, so functions that search for required sessions or calls must have high-performance characteristics. However, high-throughput storage and high-speed searching have a tradeoff relationship, so achieving both requires some type of improvement to be taken with regard to data structure. To store large amounts of data, the development of distributed storage technology for forming a large-capacity storage system in combination with general-purpose servers (commodity hardware) has been accelerating in conjunction with data storage technologies on the cloud. This type of storage has the advantages on the capacity and data management. While the packet data stored can be useful at the time of a problem analysis, it is difficult to estimate just how long packet data should be stored, but distributed storage, which has excellent scale-out capabilities, can be initially implemented on a small scale and can be added the capacities if necessary. Additionally, packet data incorporates metadata such as time stamps and source addresses, which can be used to narrow down packet data to a targeted group of packets during the analysis process. As a result, object storage, which features a search function based on metadata, can reduce data management costs by enabling the uniform management of packet data and metadata. However, distributed object storages on the cloud are typically developed for a large size data and not designed for small size data while searching the sequence of data such as a session. So, some types of improvement need to be taken to utilize distributed object storage for packet data storage and session searching. In this paper, we propose distributed object storage technology for achieving both high-throughput storage of packet data and high-speed session searching. We propose, in particular, a system that can store small-size data such as packet data with high throughput while deleting old data and conducting searches. We also compare the storage performance of the proposed system with that of other distributed storage systems and present the results of evaluating search performance for typical search patterns. II. ISSUES The following issues must be addressed in the development of distributed object storage toward the storage and usage of packet data flowing in a high-speed network. A. High-throughput storage of small- and medium-size data Object storage typically targets relatively large-size data (several tens of MB and larger), small- and medium-size data (up to several MB) is generally not the focus of this type of storage technology. However, achieving both storage and analysis capabilities in a packet-data storage and usage system Copyright IEICE - Asia-Pacific Network Operation and Management Symposium (APNOMS) 2014

2 requires high-throughput storage on the level of network throughput specifically for small- and medium-size data. Here, an optimal balance must be achieved between analysis and storage needs due to the tradeoff relationship between these functions. That is, if data size is too large, data becomes difficult to use at the time of analysis, and if data size is too small, acceptable storage performance cannot be extracted. Moreover, for the same total storage capacity, a smaller object size means a greater number of objects over the entire system, which might become bottlenecks and degrade searching performance. B. High-speed data deletion In object storage, the focus is on data storage and the deletion of data is usually not a matter of importance. The storage of packet data, however, requires that new data be stored while deleting old data once storage fills up. This datadeletion operation therefore requires a level of performance the same as that of storage. module is to cache the hash-space allocation table that indicates which server an object is stored on (or which server the object should be sent to at the time of initial writing) by querying the storage servers beforehand and to update the allocation table if access should later fail by querying the servers again. Here, to save index information for search purposes, the concept of a bucket is created for objects as shown by the conceptual diagram of Figure 1. The relationship between buckets and objects is such that multiple objects exist under a bucket similar to the way that files exist under a directory. In addition, a bucket will consolidate the metadata used for searching from all of the metadata associated with its group of objects and will likewise save that metadata in index form. Buckets will also be distributed among storage servers using name-based consistent hashing the same as done for objects and will also be saved in redundant form among multiple storage servers. C. Real-time analysis maintaining storage performance In packet data analysis, it is essential that no packets be lost. This means that data needed for analysis must be searched for, retrieved, and read at high speed while maintaining the required level of storage performance. We propose the distributed object storage for highthroughput packet data as a total system for resolving issues A to C above. III. BASIC DESIGN OF DISTRIBUTED OBJECT STORAGE FOR HIGH-THROUGHPUT PACKET DATA Objects can be made redundant among storage servers by having a capture machine directly transmit the same object to multiple storage servers. This process, however, can generate transmission bottlenecks in the network. Additionally, while the storage of packet data that comes one after another generates a continuous load in the storage servers, analysis is performed in an ad hoc manner. Data access should therefore incorporate some form of load distribution at the time of analysis. Furthermore, to store a large volume of small- and medium-size objects, a very large number of objects will have to be retained across the entire system, but the centralized management of those object-storage locations could lead to bottlenecks that degrade storage and search performance. We developed new distributed object storage consisting of a proxy module and storage servers mounting hard disk drives (HDDs). Here, an item that combines the data itself and various types of metadata is called an object. It distributes objects among multiple storage servers using name-based consistent hashing and saves objects in redundant form among those servers. Each storage server in the system writes objects to its HDDs and controls interfacing with other storage servers. It also determines the area in hash space allocated to it through adjustments made with other storage servers. Access to the system is achieved via the proxy module. The role of this Figure 1: Conceptual diagram of buckets and objects ( # of replicas = 1 ) Redundant objects (called replicas ) spread across multiple servers each have a role with a ranking attached (primary, secondary, tertiary, etc.). When writing is performed by a proxy module as shown in Figure 2, an object will be transferred along a chain in the order of primary, secondary, tertiary, etc., and when the end point of the chain is reached, responses will be returned in reverse order. Furthermore, on receiving an object write request, a server will transfer that object to the next server and will transfer the associated metadata to the bucket that it belongs to. Here, the processing for adding information to indexes in a receiving bucket will be performed in order along a chain the same as object-write processing. For the sake of balancing load, reading and searching can be performed from any redundant object.

3 IV. Figure 2: Writing and index creation ( # of replicas = 3 ) HIGH-THROUGHPUT STORAGE AND HIGH-SPEED SEARCHING SYSTEM A. Data structure The type of distributed object storage that we aim for requires that small-size packet data be stored at highthroughput and that typical forms of searching be performed at high speeds. Here, if we were to aim only for high-throughput storage of small-size data, we could consider a scheme that repeatedly buffers data in memory and writes that data sequentially to HDDs. This scheme is actually used in some packet capture systems. When also performing analysis, however, this scheme would mean that the packets targeted for reading which compose a session or a call would be stored at discrete locations within this time series of stored packet data. As a result, a large volume of random reads would occur in the analysis process that would not only hinder high-speed analysis but also affect storage performance. Search keys typically used in packet data analysis are time and source address. Among these, searching on the basis of the source-address key is optimized by grouping and writing packets in appropriate ranges of source addresses (called source groups) instead of writing them in time-series order. In this way, the search range can be limited when searching for a specific source address and high-speed searching can be performed. Additionally, prior to grouping by source address, packet groups whose packets share the same values for the 5- tuple set of metadata (destination/source addresses, destination/ source port numbers, protocol number) are extracted as sessions and the metadata of those extracted sessions is saved in units of sessions. This approach can greatly reduce metadata capacity compared to the case of saving metadata in units of packets. Session metadata includes such items as session start/end times and session size in addition to the above 5-tuple items. Object structure oriented to packet data storing and searching is shown in Figure 3. Each object consists of a data section and a metadata section. The data section is of fixed length and consists of packet data and headers for analysis use. Packet data are grouped in units of sessions and ordered in time-series form within a session. Grouping and storing packet data in units of sessions in this way means that reading only has to be performed at a specific location instead of having to perform discrete reads across all items of stored packet data. This approach makes for high-speed analysis. Additionally, given that the size of the area used for buffering packets is finite, sessions that continue for a relatively long time can be spread across multiple objects. In packet-data analysis, it is generally assumed that the packet data targeted for analysis is narrowed down by a number of key-based searches. If each key-based search accesses to the HDDs, that can significantly affect packet-data storage performance. Our system, however, consolidates metadata in units of sessions thereby reducing the volume of metadata significantly. This means that the index used for searching can be placed in memory enabling searches to be executed without affecting storage performance. Once the object which includes the packets targeted is found, then it is read out for analysis. Since the bigger the object is, the more sessions are included in it and not related packets are included, smaller object size is better for reading out and reduces overhead at the time of analysis. Object metadata are consolidated and indexed in units of buckets. As a method for optimizing the search process using time and source address, the system determines these buckets in terms of the earliest start time (called object start time) from among the sessions included in an object and in terms of source group. In other words, indexes are divided along the two axes of time and source address. Here, search processing and index creation in units of buckets can be performed in parallel, which means that using a high number of buckets can lead to improved search performance. On the other hand, a very detailed division of indexes can increase the number of searches needed for analysis thereby degrading the overall performance of analysis processing. An optimal number of buckets used must therefore be set. Figure 3: Object example B. Data allocation The storage performance of an HDD is at its highest when data is written sequentially. When packet data begins to be

4 stored, maximum throughput can be achieved by writing new objects sequentially. A problem here is that new data will have to be written while deleting old data once the HDD has reached its full capacity. As a result, the HDD will have to be accessed to update HDD area management information in conjunction with the deletion of old data, and this, in turn, will impede the sequential writing of data and affect storage performance. In response to this problem, HDD area management information is stored in memory in bitmap form. Here, setting the data section of objects to a fixed length simplifies bitmap management. As for the metadata section, the metadata of multiple objects are grouped into a size close to the fixed length of the data section before recording the data on an HDD. When writing objects, HDD areas are allocated in the direction of increasing addresses regardless of the distribution of free areas, and when the end of the HDD is reached, area allocation begins again from the top of the HDD. C. Search method As described above, an index is divided into object start time and source group. This approach enables the search target to be narrowed down when search conditions includes time and source address. In conducting a search, the system begins by extracting the group of buckets targeted by the search. It then submits search queries in parallel against those extracted buckets, merges the results returned from the queries, and generates overall search results. Table 2: Storage-server/search-server specifications Model Fujitsu PRIMERGY RX100 S6 CPU Intel Xeon E Memory 16GB HDD Seagate Constellation2 SATA 1TB 4 with RAID0 RAID Card LSI SAS1064e B. Storage-server HDD performance Each server was given a RAID0 configuration using 4 SATA HDDs and a RAID card. For this RAID0 volume, we measured the performance of sequential read/write and random read/write for a range of I/O sizes. Specifically, we measured sequential performance using IOzone [6] and measured random performance by issuing random accesses across all HDD areas using asynchronous direct I/O. As shown in Figure 4, sequential write performance was steady at about 365 MB/s regardless of I/O size. Assuming that all storage servers making up distributed storage were running in an ideal state in which all writing to HDDs is sequential, then the network performance limit of 10 Gbps could be reached by 4 servers for a replica number of 1 and by 10 servers for a replica number of 3. In contrast, random write performance varied by I/O size. These results show that performance improves as I/O size increases. In the following evaluation, we used a data size of 4 MB. V. EVALUATION With the aim of evaluating the performance of our distributed object storage in the storage and use of packet data, we compared it with other distributed storage systems (Apache HDFS [7], OpenStack Swift [8]) and evaluated its search performance using typical search patterns. A. Experimental environment We performed this evaluation in an experimental system interconnecting one capture machine, 30 storage servers, and one search server in a 10 Gbps network. For the capture machine, we used a server having 2 sockets of an 8-core/16-thread Intel Xeon E processor and 256 GB of main memory (Table 1), and for each storage server, we used a server having 1 socket of a 4-core/8-thread Intel Xeon E processor, 16 GB of main memory, and 4 1TB SATA HDDs (Table 2). Additionally, for the search server, we used the same type of server as the storage server. We also used a Cisco Nexus 5548 switch to interconnect these servers in a 10 Gbps Ethernet system. Each server ran a 64-bit version of CentOS 6.4 as OS. Table 1: Capture machine specifications Model Fujitsu PRIMERGY RX300 S7 CPU Intel Xeon E Memory 256GB Figure 4: Storage-server HDD performance ( HDD 4 with RAID0 ) C. Storage performance comparison Given packet-data storage under steady-state conditions in which new data is being written while old data is being deleted, we compared storage performance with that of open-source distributed storage Apache HDFS (ver. CDH4.3.1) and OpenStack Swift (ver ). We used the capture machine described above as a load machine, and on this machine, we ran a proxy module for our system and Swift measurements and NameNode for HDFS measurements.

5 Our system creates multiple buckets according to source addresses to limit the search range as described earlier. To emulate this feature, we created 256 buckets beforehand covering all storage servers and repeatedly wrote 4 MB objects selecting buckets at random. We also measured storage performance when deleting old objects in parallel. Specifically, we used the following procedure: (1) begin object writing, (2) after 1800 s, begin deleting any objects for which a time period of 60 s or longer has elapsed since their creation, and (3) calculate average storage performance s after beginning deletion. Storage performance of 10Gbps with triple replication, for instance, consumed 225GB of storage. Storage performance is compared in Figure 5 for a replica number of 3. It can be seen from these results that our system achieves a level of storage performance near the theoretical value and reaches its performance limit of 10 Gbps using 16 storage servers. In contrast, the storage performance of HDFS and Swift for 16 storage servers is about one half and one third, respectively, that of our system. Figure 6: Comparison of storage performance ( # of replicas = 1, object size = 4MB ) HDFS achieved its limit in storage performance at about one half that of our system, but this was due to a CPU bottleneck at NameNode. When writing, HDFS determines the write destination after issuing a query to NameNode, which means that the number of queries sent to NameNode increases as object size becomes smaller. Taking that into consideration, Figure 7 shows results when making measurements for an object size of 64 MB. In this case, HDFS surpassed its performance limit under an object size of 4 MB and achieved a performance limit of 10 Gbps the same as that of our system. Figure 5: Comparison of storage performance ( # of replicas = 3, object size = 4MB ) Using a method different from that of our system and HDFS, Swift creates replicas by having the proxy module write them directly to multiple storage servers. Consequently, for a replica number of 3, the limit in storage performance that can be achieved from one load machine by Swift is one third that of others. With this in mind, storage performance is compared in Figure 6 for a replica number of 1. These results show that our system achieved a storage performance near the theoretical value and reached its performance limit of 10 Gbps using only 3 storage servers. HDFS, meanwhile, achieved the same storage performance as that for a replica number of 3 with 3 storage servers. Swift, however, with replica number set to 1, showed an improvement in its upper performance limit compared to that for a replica number of 3, but the storage performance so reached was only about half that of our system. Figure 7: Comparison of storage performance ( # of replicas = 3, object size = 64MB ) D. Search performance We next measured search performance during actual capturing and storing of packet data. The traffic targeted for capturing was created using Agilent s N2X test platform. Here, given one TCP session for each of 256 source groups (for a total of 256 sessions), traffic flowing at a total rate of 10 Gbps in the network was captured at the capture machine and stored in the system. Then, after a certain amount of time, search processing began and response time was measured. Specifically, at 600 s after the start of packet data capturing, a search was performed for sessions included in the one-second

6 interval 600 to 599 s earlier. This search was performed 10 times each for the case of specifying a source address (SA) and the case of not specifying a SA and average response time was computed. Storage performance of 10Gbps with triple replication, for instance, consumed 225GB of storage. Measurement results for 16 storage servers and a replica number of 3 while varying packet length in the range of bytes are shown in Figure 8. The left axis represents storage performance and the right axis represents search performance. Since the processing required for session extraction increases as packet length becomes smaller, a bottleneck occurred at the capture-machine CPU for a packet length of 256 bytes resulting in dropped packets during capture processing. This reduced the amount of packet data stored. However, for a packet length of 1024 bytes, it became possible to perform capture processing without dropping packets resulting in a storage load of 10 Gbps. At packet lengths of 1024 and 1280 bytes corresponding to a storage load of 10 Gbps, performing a session search while specifying SA enabled search results to be retrieved in msec but doing the same while specifying no SA resulted in a retrieval time of msec. high-speed network. Finally, Deri et al. [5] proposed a system for extracting a level of performance near 10 Gbps by making buffer processing more efficient in general-purpose servers mounting solid state drives (SSDs), but the use of SSDs here means a high-cost system for storing data over the long term. Our proposed technique represents a new approach in that it attempts to achieve distributed object storage realizing both high-throughput storage and high-speed searching of packet data using general-purpose servers mounting HDDs. VII. CONCLUSION We proposed a distributed object storage system for achieving a packet capture system that can provide a high level of performance for both storage and searching. Our system consolidates packet data in units of sessions and groups them according to source address into fixed lengths of small/medium size. It also generates objects combining packet data and the metadata of that session, creates a distribution of those objects using constant hashing and writes them to servers in a chainlike sequence for redundant storing, and creates indexes divided into the two metadata axes of object start time and source address. In comparing our system with Apache HDFS and OpenStack Swift, we showed that the proposed system achieves a level of storage performance about 2 and 4 times that of HDFS and Swift, respectively, for an object size of 4 MB and writing to 16 storage servers. Furthermore, for a system capturing and storing traffic on an order of 10 Gbps, we showed that session searching could be performed in about milliseconds when specifying a source address and in about milliseconds when not specifying a source address. Looking forward, we plan to evaluate the proposed system even further with the aim of enhancing the storage and use of packet data flowing in high-speed networks. We will also examine distributed processing on the storage-server side using the metadata attached to objects. Figure 8: Search performance during data storing VI. RELATED RESEARCH Morariu et al. [3] proposed a packet capture system using P2P-type distributed storage with a distributed hash table. They showed that this system could achieve high-speed searching by selecting a storage-destination server using, for example, packet source address and applying a data structure specialized for searching within the server. Lee et al. [2] proposed a method using Hadoop while Fusco et al. [4] proposed a method capable of high-speed packet searching using bitmap indexes, but neither method focused on storage performance. In addition to the above, Tsukahara et al. [1] proposed a method for storing packet data on the capture machine and improving packet storage performance by tuning system parameters in general-purpose servers mounting HDDs. It would be difficult to say, however, that this method achieves sufficient performance for storing packet data flowing in a REFERENCES [1] Tsukahara, Yasuhito, Takashi Tomine, and Kazunori Sugiura. The Tuning Method of Packet Capturing System to Put Analyzing the Network Traffic Data with General Computer into Reality, Information Processing Society of Japan (IPSJ) Technical Report, Computer Security Group (CSEC), 2010-CSEC-48(7), pp. 1-7, (in Japanese) [2] Lee, Yeonhee, Wonchul Kang, and Youngseok Lee. A hadoop-based packet trace processing tool. Traffic Monitoring and Analysis. Springer Berlin Heidelberg, [3] Morariu, Cristian, Thierry Kramis, and Burkhard Stiller. DIPStorage: distributed storage of IP flow records. Local and Metropolitan Area Networks, LANMAN th IEEE Workshop on. IEEE, [4] Fusco, Francesco, et al. pcapindex: an index for network packet traces with legacy compatibility. ACM SIGCOMM Computer Communication Review 42.1 (2012): [5] Deri, Luca, Alfredo Cardigliano, and Francesco Fusco. 10 Gbit line rate packet-to-disk using n2disk. Computer Communications Workshops (INFOCOM WKSHPS), 2013 IEEE Conference on. IEEE, [6] IOzone homepage: [7] Apache Hadoop homepage: [8] OpenStack homepage:

Network Intrusion Forensics System based on Collection and Preservation of Attack Evidence

Network Intrusion Forensics System based on Collection and Preservation of Attack Evidence , pp.354-359 http://dx.doi.org/10.14257/astl.2016.139.71 Network Intrusion Forensics System based on Collection and Preservation of Attack Evidence Jong-Hyun Kim, Yangseo Choi, Joo-Young Lee, Sunoh Choi,

More information

SurFS Product Description

SurFS Product Description SurFS Product Description 1. ABSTRACT SurFS An innovative technology is evolving the distributed storage ecosystem. SurFS is designed for cloud storage with extreme performance at a price that is significantly

More information

White Paper Features and Benefits of Fujitsu All-Flash Arrays for Virtualization and Consolidation ETERNUS AF S2 series

White Paper Features and Benefits of Fujitsu All-Flash Arrays for Virtualization and Consolidation ETERNUS AF S2 series White Paper Features and Benefits of Fujitsu All-Flash Arrays for Virtualization and Consolidation Fujitsu All-Flash Arrays are extremely effective tools when virtualization is used for server consolidation.

More information

Network Design Considerations for Grid Computing

Network Design Considerations for Grid Computing Network Design Considerations for Grid Computing Engineering Systems How Bandwidth, Latency, and Packet Size Impact Grid Job Performance by Erik Burrows, Engineering Systems Analyst, Principal, Broadcom

More information

Data Protection for Cisco HyperFlex with Veeam Availability Suite. Solution Overview Cisco Public

Data Protection for Cisco HyperFlex with Veeam Availability Suite. Solution Overview Cisco Public Data Protection for Cisco HyperFlex with Veeam Availability Suite 1 2017 2017 Cisco Cisco and/or and/or its affiliates. its affiliates. All rights All rights reserved. reserved. Highlights Is Cisco compatible

More information

EsgynDB Enterprise 2.0 Platform Reference Architecture

EsgynDB Enterprise 2.0 Platform Reference Architecture EsgynDB Enterprise 2.0 Platform Reference Architecture This document outlines a Platform Reference Architecture for EsgynDB Enterprise, built on Apache Trafodion (Incubating) implementation with licensed

More information

WHITEPAPER. Improve Hadoop Performance with Memblaze PBlaze SSD

WHITEPAPER. Improve Hadoop Performance with Memblaze PBlaze SSD Improve Hadoop Performance with Memblaze PBlaze SSD Improve Hadoop Performance with Memblaze PBlaze SSD Exclusive Summary We live in the data age. It s not easy to measure the total volume of data stored

More information

IBM V7000 Unified R1.4.2 Asynchronous Replication Performance Reference Guide

IBM V7000 Unified R1.4.2 Asynchronous Replication Performance Reference Guide V7 Unified Asynchronous Replication Performance Reference Guide IBM V7 Unified R1.4.2 Asynchronous Replication Performance Reference Guide Document Version 1. SONAS / V7 Unified Asynchronous Replication

More information

Accelerate Database Performance and Reduce Response Times in MongoDB Humongous Environments with the LSI Nytro MegaRAID Flash Accelerator Card

Accelerate Database Performance and Reduce Response Times in MongoDB Humongous Environments with the LSI Nytro MegaRAID Flash Accelerator Card Accelerate Database Performance and Reduce Response Times in MongoDB Humongous Environments with the LSI Nytro MegaRAID Flash Accelerator Card The Rise of MongoDB Summary One of today s growing database

More information

FEATURE. High-throughput Video Data Transfer of Striping with SSDs for 8K Super Hi-Vision

FEATURE. High-throughput Video Data Transfer of Striping with SSDs for 8K Super Hi-Vision High-throughput Video Data Transfer of Striping with s for 8K Super Hi-Vision Takeshi KAJIYAMA, Kodai KIKUCHI and Eiichi MIYASHITA We have developed a high-throughput recording and playback method for

More information

Correlation based File Prefetching Approach for Hadoop

Correlation based File Prefetching Approach for Hadoop IEEE 2nd International Conference on Cloud Computing Technology and Science Correlation based File Prefetching Approach for Hadoop Bo Dong 1, Xiao Zhong 2, Qinghua Zheng 1, Lirong Jian 2, Jian Liu 1, Jie

More information

Dell PowerEdge R720xd with PERC H710P: A Balanced Configuration for Microsoft Exchange 2010 Solutions

Dell PowerEdge R720xd with PERC H710P: A Balanced Configuration for Microsoft Exchange 2010 Solutions Dell PowerEdge R720xd with PERC H710P: A Balanced Configuration for Microsoft Exchange 2010 Solutions A comparative analysis with PowerEdge R510 and PERC H700 Global Solutions Engineering Dell Product

More information

Presented by: Nafiseh Mahmoudi Spring 2017

Presented by: Nafiseh Mahmoudi Spring 2017 Presented by: Nafiseh Mahmoudi Spring 2017 Authors: Publication: Type: ACM Transactions on Storage (TOS), 2016 Research Paper 2 High speed data processing demands high storage I/O performance. Flash memory

More information

High-Performance IP Service Node with Layer 4 to 7 Packet Processing Features

High-Performance IP Service Node with Layer 4 to 7 Packet Processing Features UDC 621.395.31:681.3 High-Performance IP Service Node with Layer 4 to 7 Packet Processing Features VTsuneo Katsuyama VAkira Hakata VMasafumi Katoh VAkira Takeyama (Manuscript received February 27, 2001)

More information

Cisco Tetration Analytics Platform: A Dive into Blazing Fast Deep Storage

Cisco Tetration Analytics Platform: A Dive into Blazing Fast Deep Storage White Paper Cisco Tetration Analytics Platform: A Dive into Blazing Fast Deep Storage What You Will Learn A Cisco Tetration Analytics appliance bundles computing, networking, and storage resources in one

More information

White Paper FUJITSU Storage ETERNUS DX S4/S3 series Extreme Cache/Extreme Cache Pool best fit for fast processing of vast amount of data

White Paper FUJITSU Storage ETERNUS DX S4/S3 series Extreme Cache/Extreme Cache Pool best fit for fast processing of vast amount of data White Paper FUJITSU Storage ETERNUS DX S4/S3 series Extreme Cache/Extreme Cache Pool best fit for fast processing of vast amount of data Extreme Cache / Extreme Cache Pool, which expands cache capacity

More information

Validating the NetApp Virtual Storage Tier in the Oracle Database Environment to Achieve Next-Generation Converged Infrastructures

Validating the NetApp Virtual Storage Tier in the Oracle Database Environment to Achieve Next-Generation Converged Infrastructures Technical Report Validating the NetApp Virtual Storage Tier in the Oracle Database Environment to Achieve Next-Generation Converged Infrastructures Tomohiro Iwamoto, Supported by Field Center of Innovation,

More information

Deploy a High-Performance Database Solution: Cisco UCS B420 M4 Blade Server with Fusion iomemory PX600 Using Oracle Database 12c

Deploy a High-Performance Database Solution: Cisco UCS B420 M4 Blade Server with Fusion iomemory PX600 Using Oracle Database 12c White Paper Deploy a High-Performance Database Solution: Cisco UCS B420 M4 Blade Server with Fusion iomemory PX600 Using Oracle Database 12c What You Will Learn This document demonstrates the benefits

More information

Low Latency Evaluation of Fibre Channel, iscsi and SAS Host Interfaces

Low Latency Evaluation of Fibre Channel, iscsi and SAS Host Interfaces Low Latency Evaluation of Fibre Channel, iscsi and SAS Host Interfaces Evaluation report prepared under contract with LSI Corporation Introduction IT professionals see Solid State Disk (SSD) products as

More information

White paper ETERNUS Extreme Cache Performance and Use

White paper ETERNUS Extreme Cache Performance and Use White paper ETERNUS Extreme Cache Performance and Use The Extreme Cache feature provides the ETERNUS DX500 S3 and DX600 S3 Storage Arrays with an effective flash based performance accelerator for regions

More information

Linux Software RAID Level 0 Technique for High Performance Computing by using PCI-Express based SSD

Linux Software RAID Level 0 Technique for High Performance Computing by using PCI-Express based SSD Linux Software RAID Level Technique for High Performance Computing by using PCI-Express based SSD Jae Gi Son, Taegyeong Kim, Kuk Jin Jang, *Hyedong Jung Department of Industrial Convergence, Korea Electronics

More information

Using Synology SSD Technology to Enhance System Performance Synology Inc.

Using Synology SSD Technology to Enhance System Performance Synology Inc. Using Synology SSD Technology to Enhance System Performance Synology Inc. Synology_WP_ 20121112 Table of Contents Chapter 1: Enterprise Challenges and SSD Cache as Solution Enterprise Challenges... 3 SSD

More information

WHITE PAPER AGILOFT SCALABILITY AND REDUNDANCY

WHITE PAPER AGILOFT SCALABILITY AND REDUNDANCY WHITE PAPER AGILOFT SCALABILITY AND REDUNDANCY Table of Contents Introduction 3 Performance on Hosted Server 3 Figure 1: Real World Performance 3 Benchmarks 3 System configuration used for benchmarks 3

More information

Intel Enterprise Edition Lustre (IEEL-2.3) [DNE-1 enabled] on Dell MD Storage

Intel Enterprise Edition Lustre (IEEL-2.3) [DNE-1 enabled] on Dell MD Storage Intel Enterprise Edition Lustre (IEEL-2.3) [DNE-1 enabled] on Dell MD Storage Evaluation of Lustre File System software enhancements for improved Metadata performance Wojciech Turek, Paul Calleja,John

More information

CA485 Ray Walshe Google File System

CA485 Ray Walshe Google File System Google File System Overview Google File System is scalable, distributed file system on inexpensive commodity hardware that provides: Fault Tolerance File system runs on hundreds or thousands of storage

More information

SMCCSE: PaaS Platform for processing large amounts of social media

SMCCSE: PaaS Platform for processing large amounts of social media KSII The first International Conference on Internet (ICONI) 2011, December 2011 1 Copyright c 2011 KSII SMCCSE: PaaS Platform for processing large amounts of social media Myoungjin Kim 1, Hanku Lee 2 and

More information

Approaches to Green Networks

Approaches to Green Networks Approaches to Green Networks Mitsuaki Kakemizu Akira Chugo (Manuscript received March 26, 2009) Communication networks show promise as a means of reducing carbon dioxide emissions. However, as networks

More information

A Fast and High Throughput SQL Query System for Big Data

A Fast and High Throughput SQL Query System for Big Data A Fast and High Throughput SQL Query System for Big Data Feng Zhu, Jie Liu, and Lijie Xu Technology Center of Software Engineering, Institute of Software, Chinese Academy of Sciences, Beijing, China 100190

More information

Solid State Storage Technologies. Jin-Soo Kim Computer Systems Laboratory Sungkyunkwan University

Solid State Storage Technologies. Jin-Soo Kim Computer Systems Laboratory Sungkyunkwan University Solid State Storage Technologies Jin-Soo Kim (jinsookim@skku.edu) Computer Systems Laboratory Sungkyunkwan University http://csl.skku.edu NVMe (1) The industry standard interface for high-performance NVM

More information

Catalogic DPX TM 4.3. ECX 2.0 Best Practices for Deployment and Cataloging

Catalogic DPX TM 4.3. ECX 2.0 Best Practices for Deployment and Cataloging Catalogic DPX TM 4.3 ECX 2.0 Best Practices for Deployment and Cataloging 1 Catalogic Software, Inc TM, 2015. All rights reserved. This publication contains proprietary and confidential material, and is

More information

Distributed Filesystem

Distributed Filesystem Distributed Filesystem 1 How do we get data to the workers? NAS Compute Nodes SAN 2 Distributing Code! Don t move data to workers move workers to the data! - Store data on the local disks of nodes in the

More information

MODERN FILESYSTEM PERFORMANCE IN LOCAL MULTI-DISK STORAGE SPACE CONFIGURATION

MODERN FILESYSTEM PERFORMANCE IN LOCAL MULTI-DISK STORAGE SPACE CONFIGURATION INFORMATION SYSTEMS IN MANAGEMENT Information Systems in Management (2014) Vol. 3 (4) 273 283 MODERN FILESYSTEM PERFORMANCE IN LOCAL MULTI-DISK STORAGE SPACE CONFIGURATION MATEUSZ SMOLIŃSKI Institute of

More information

Building a High IOPS Flash Array: A Software-Defined Approach

Building a High IOPS Flash Array: A Software-Defined Approach Building a High IOPS Flash Array: A Software-Defined Approach Weafon Tsao Ph.D. VP of R&D Division, AccelStor, Inc. Santa Clara, CA Clarification Myth 1: S High-IOPS SSDs = High-IOPS All-Flash Array SSDs

More information

IBM and HP 6-Gbps SAS RAID Controller Performance

IBM and HP 6-Gbps SAS RAID Controller Performance IBM and HP 6-Gbps SAS RAID Controller Performance Evaluation report prepared under contract with IBM Corporation Introduction With increasing demands on storage in popular application servers, the server

More information

CSE 124: Networked Services Lecture-16

CSE 124: Networked Services Lecture-16 Fall 2010 CSE 124: Networked Services Lecture-16 Instructor: B. S. Manoj, Ph.D http://cseweb.ucsd.edu/classes/fa10/cse124 11/23/2010 CSE 124 Networked Services Fall 2010 1 Updates PlanetLab experiments

More information

Spark Over RDMA: Accelerate Big Data SC Asia 2018 Ido Shamay Mellanox Technologies

Spark Over RDMA: Accelerate Big Data SC Asia 2018 Ido Shamay Mellanox Technologies Spark Over RDMA: Accelerate Big Data SC Asia 2018 Ido Shamay 1 Apache Spark - Intro Spark within the Big Data ecosystem Data Sources Data Acquisition / ETL Data Storage Data Analysis / ML Serving 3 Apache

More information

Flat Datacenter Storage. Edmund B. Nightingale, Jeremy Elson, et al. 6.S897

Flat Datacenter Storage. Edmund B. Nightingale, Jeremy Elson, et al. 6.S897 Flat Datacenter Storage Edmund B. Nightingale, Jeremy Elson, et al. 6.S897 Motivation Imagine a world with flat data storage Simple, Centralized, and easy to program Unfortunately, datacenter networks

More information

Software-defined Storage: Fast, Safe and Efficient

Software-defined Storage: Fast, Safe and Efficient Software-defined Storage: Fast, Safe and Efficient TRY NOW Thanks to Blockchain and Intel Intelligent Storage Acceleration Library Every piece of data is required to be stored somewhere. We all know about

More information

Decentralized Distributed Storage System for Big Data

Decentralized Distributed Storage System for Big Data Decentralized Distributed Storage System for Big Presenter: Wei Xie -Intensive Scalable Computing Laboratory(DISCL) Computer Science Department Texas Tech University Outline Trends in Big and Cloud Storage

More information

Analysis of Multiple Target Network Storage Access using Multi-routing VPN Connections

Analysis of Multiple Target Network Storage Access using Multi-routing VPN Connections Analysis of Multiple Network Storage Access using Multi-routing Connections Nozomi Chishima Ochanomizu University 2 1 1, Otsuka, Bunkyo-ku Tokyo 112 861, JAPAN nozomi@ogl.is.ocha.ac.jp Saneyasu Yamaguchi

More information

Cloudian Sizing and Architecture Guidelines

Cloudian Sizing and Architecture Guidelines Cloudian Sizing and Architecture Guidelines The purpose of this document is to detail the key design parameters that should be considered when designing a Cloudian HyperStore architecture. The primary

More information

IBM InfoSphere Streams v4.0 Performance Best Practices

IBM InfoSphere Streams v4.0 Performance Best Practices Henry May IBM InfoSphere Streams v4.0 Performance Best Practices Abstract Streams v4.0 introduces powerful high availability features. Leveraging these requires careful consideration of performance related

More information

Performance Analysis of iscsi Middleware Optimized for Encryption Processing in a Long-Latency Environment

Performance Analysis of iscsi Middleware Optimized for Encryption Processing in a Long-Latency Environment Performance Analysis of iscsi Middleware Optimized for Encryption Processing in a Long-Latency Environment Kikuko Kamisaka Graduate School of Humanities and Sciences Ochanomizu University -1-1, Otsuka,

More information

Exploiting the full power of modern industry standard Linux-Systems with TSM Stephan Peinkofer

Exploiting the full power of modern industry standard Linux-Systems with TSM Stephan Peinkofer TSM Performance Tuning Exploiting the full power of modern industry standard Linux-Systems with TSM Stephan Peinkofer peinkofer@lrz.de Agenda Network Performance Disk-Cache Performance Tape Performance

More information

Highly accurate simulations of big-data clusters for system planning and optimization

Highly accurate simulations of big-data clusters for system planning and optimization White Paper Highly accurate simulations of big-data clusters for system planning and optimization Intel CoFluent Technology for Big Data Intel Rack Scale Design Using Intel CoFluent Technology for Big

More information

A High-Performance Storage and Ultra- High-Speed File Transfer Solution for Collaborative Life Sciences Research

A High-Performance Storage and Ultra- High-Speed File Transfer Solution for Collaborative Life Sciences Research A High-Performance Storage and Ultra- High-Speed File Transfer Solution for Collaborative Life Sciences Research Storage Platforms with Aspera Overview A growing number of organizations with data-intensive

More information

Configuring Storage Profiles

Configuring Storage Profiles This part contains the following chapters: Storage Profiles, page 1 Disk Groups and Disk Group Configuration Policies, page 2 RAID Levels, page 3 Automatic Disk Selection, page 4 Supported LUN Modifications,

More information

Distributed Systems 16. Distributed File Systems II

Distributed Systems 16. Distributed File Systems II Distributed Systems 16. Distributed File Systems II Paul Krzyzanowski pxk@cs.rutgers.edu 1 Review NFS RPC-based access AFS Long-term caching CODA Read/write replication & disconnected operation DFS AFS

More information

The Oracle Database Appliance I/O and Performance Architecture

The Oracle Database Appliance I/O and Performance Architecture Simple Reliable Affordable The Oracle Database Appliance I/O and Performance Architecture Tammy Bednar, Sr. Principal Product Manager, ODA 1 Copyright 2012, Oracle and/or its affiliates. All rights reserved.

More information

A Comparative Study of Microsoft Exchange 2010 on Dell PowerEdge R720xd with Exchange 2007 on Dell PowerEdge R510

A Comparative Study of Microsoft Exchange 2010 on Dell PowerEdge R720xd with Exchange 2007 on Dell PowerEdge R510 A Comparative Study of Microsoft Exchange 2010 on Dell PowerEdge R720xd with Exchange 2007 on Dell PowerEdge R510 Incentives for migrating to Exchange 2010 on Dell PowerEdge R720xd Global Solutions Engineering

More information

Evaluation Report: Improving SQL Server Database Performance with Dot Hill AssuredSAN 4824 Flash Upgrades

Evaluation Report: Improving SQL Server Database Performance with Dot Hill AssuredSAN 4824 Flash Upgrades Evaluation Report: Improving SQL Server Database Performance with Dot Hill AssuredSAN 4824 Flash Upgrades Evaluation report prepared under contract with Dot Hill August 2015 Executive Summary Solid state

More information

Best Practices for Setting BIOS Parameters for Performance

Best Practices for Setting BIOS Parameters for Performance White Paper Best Practices for Setting BIOS Parameters for Performance Cisco UCS E5-based M3 Servers May 2013 2014 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public. Page

More information

Microsoft SQL Server in a VMware Environment on Dell PowerEdge R810 Servers and Dell EqualLogic Storage

Microsoft SQL Server in a VMware Environment on Dell PowerEdge R810 Servers and Dell EqualLogic Storage Microsoft SQL Server in a VMware Environment on Dell PowerEdge R810 Servers and Dell EqualLogic Storage A Dell Technical White Paper Dell Database Engineering Solutions Anthony Fernandez April 2010 THIS

More information

The Google File System

The Google File System The Google File System Sanjay Ghemawat, Howard Gobioff, and Shun-Tak Leung December 2003 ACM symposium on Operating systems principles Publisher: ACM Nov. 26, 2008 OUTLINE INTRODUCTION DESIGN OVERVIEW

More information

Performance and Optimization Issues in Multicore Computing

Performance and Optimization Issues in Multicore Computing Performance and Optimization Issues in Multicore Computing Minsoo Ryu Department of Computer Science and Engineering 2 Multicore Computing Challenges It is not easy to develop an efficient multicore program

More information

SSDs vs HDDs for DBMS by Glen Berseth York University, Toronto

SSDs vs HDDs for DBMS by Glen Berseth York University, Toronto SSDs vs HDDs for DBMS by Glen Berseth York University, Toronto So slow So cheap So heavy So fast So expensive So efficient NAND based flash memory Retains memory without power It works by trapping a small

More information

Distributed File Systems II

Distributed File Systems II Distributed File Systems II To do q Very-large scale: Google FS, Hadoop FS, BigTable q Next time: Naming things GFS A radically new environment NFS, etc. Independence Small Scale Variety of workloads Cooperation

More information

CAVA: Exploring Memory Locality for Big Data Analytics in Virtualized Clusters

CAVA: Exploring Memory Locality for Big Data Analytics in Virtualized Clusters 2018 18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing : Exploring Memory Locality for Big Data Analytics in Virtualized Clusters Eunji Hwang, Hyungoo Kim, Beomseok Nam and Young-ri

More information

Extreme Storage Performance with exflash DIMM and AMPS

Extreme Storage Performance with exflash DIMM and AMPS Extreme Storage Performance with exflash DIMM and AMPS 214 by 6East Technologies, Inc. and Lenovo Corporation All trademarks or registered trademarks mentioned here are the property of their respective

More information

Accelerating Enterprise Search with Fusion iomemory PCIe Application Accelerators

Accelerating Enterprise Search with Fusion iomemory PCIe Application Accelerators WHITE PAPER Accelerating Enterprise Search with Fusion iomemory PCIe Application Accelerators Western Digital Technologies, Inc. 951 SanDisk Drive, Milpitas, CA 95035 www.sandisk.com Table of Contents

More information

SAP High-Performance Analytic Appliance on the Cisco Unified Computing System

SAP High-Performance Analytic Appliance on the Cisco Unified Computing System Solution Overview SAP High-Performance Analytic Appliance on the Cisco Unified Computing System What You Will Learn The SAP High-Performance Analytic Appliance (HANA) is a new non-intrusive hardware and

More information

Performance and Scalability with Griddable.io

Performance and Scalability with Griddable.io Performance and Scalability with Griddable.io Executive summary Griddable.io is an industry-leading timeline-consistent synchronized data integration grid across a range of source and target data systems.

More information

Test Report: Digital Rapids Transcode Manager Application with NetApp Media Content Management Solution

Test Report: Digital Rapids Transcode Manager Application with NetApp Media Content Management Solution Technical Report Test Report: Digital Rapids Transcode Manager Application with NetApp Media Content Management Solution Jim Laing, NetApp July 2012 TR-4084 TABLE OF CONTENTS 1 Executive Summary... 3 2

More information

Condusiv s V-locity VM Accelerates Exchange 2010 over 60% on Virtual Machines without Additional Hardware

Condusiv s V-locity VM Accelerates Exchange 2010 over 60% on Virtual Machines without Additional Hardware openbench Labs Executive Briefing: March 13, 2013 Condusiv s V-locity VM Accelerates Exchange 2010 over 60% on Virtual Machines without Additional Hardware Optimizing I/O for Increased Throughput and Reduced

More information

ASPERA HIGH-SPEED TRANSFER. Moving the world s data at maximum speed

ASPERA HIGH-SPEED TRANSFER. Moving the world s data at maximum speed ASPERA HIGH-SPEED TRANSFER Moving the world s data at maximum speed ASPERA HIGH-SPEED FILE TRANSFER 80 GBIT/S OVER IP USING DPDK Performance, Code, and Architecture Charles Shiflett Developer of next-generation

More information

Design a Remote-Office or Branch-Office Data Center with Cisco UCS Mini

Design a Remote-Office or Branch-Office Data Center with Cisco UCS Mini White Paper Design a Remote-Office or Branch-Office Data Center with Cisco UCS Mini February 2015 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public. Page 1 of 9 Contents

More information

ASN Configuration Best Practices

ASN Configuration Best Practices ASN Configuration Best Practices Managed machine Generally used CPUs and RAM amounts are enough for the managed machine: CPU still allows us to read and write data faster than real IO subsystem allows.

More information

White Paper. File System Throughput Performance on RedHawk Linux

White Paper. File System Throughput Performance on RedHawk Linux White Paper File System Throughput Performance on RedHawk Linux By: Nikhil Nanal Concurrent Computer Corporation August Introduction This paper reports the throughput performance of the,, and file systems

More information

CIS 601 Graduate Seminar. Dr. Sunnie S. Chung Dhruv Patel ( ) Kalpesh Sharma ( )

CIS 601 Graduate Seminar. Dr. Sunnie S. Chung Dhruv Patel ( ) Kalpesh Sharma ( ) Guide: CIS 601 Graduate Seminar Presented By: Dr. Sunnie S. Chung Dhruv Patel (2652790) Kalpesh Sharma (2660576) Introduction Background Parallel Data Warehouse (PDW) Hive MongoDB Client-side Shared SQL

More information

The Google File System

The Google File System The Google File System Sanjay Ghemawat, Howard Gobioff and Shun Tak Leung Google* Shivesh Kumar Sharma fl4164@wayne.edu Fall 2015 004395771 Overview Google file system is a scalable distributed file system

More information

The Hadoop Distributed File System Konstantin Shvachko Hairong Kuang Sanjay Radia Robert Chansler

The Hadoop Distributed File System Konstantin Shvachko Hairong Kuang Sanjay Radia Robert Chansler The Hadoop Distributed File System Konstantin Shvachko Hairong Kuang Sanjay Radia Robert Chansler MSST 10 Hadoop in Perspective Hadoop scales computation capacity, storage capacity, and I/O bandwidth by

More information

MOHA: Many-Task Computing Framework on Hadoop

MOHA: Many-Task Computing Framework on Hadoop Apache: Big Data North America 2017 @ Miami MOHA: Many-Task Computing Framework on Hadoop Soonwook Hwang Korea Institute of Science and Technology Information May 18, 2017 Table of Contents Introduction

More information

End-to-End Adaptive Packet Aggregation for High-Throughput I/O Bus Network Using Ethernet

End-to-End Adaptive Packet Aggregation for High-Throughput I/O Bus Network Using Ethernet Hot Interconnects 2014 End-to-End Adaptive Packet Aggregation for High-Throughput I/O Bus Network Using Ethernet Green Platform Research Laboratories, NEC, Japan J. Suzuki, Y. Hayashi, M. Kan, S. Miyakawa,

More information

Extremely Fast Distributed Storage for Cloud Service Providers

Extremely Fast Distributed Storage for Cloud Service Providers Solution brief Intel Storage Builders StorPool Storage Intel SSD DC S3510 Series Intel Xeon Processor E3 and E5 Families Intel Ethernet Converged Network Adapter X710 Family Extremely Fast Distributed

More information

6. Results. This section describes the performance that was achieved using the RAMA file system.

6. Results. This section describes the performance that was achieved using the RAMA file system. 6. Results This section describes the performance that was achieved using the RAMA file system. The resulting numbers represent actual file data bytes transferred to/from server disks per second, excluding

More information

I/O CANNOT BE IGNORED

I/O CANNOT BE IGNORED LECTURE 13 I/O I/O CANNOT BE IGNORED Assume a program requires 100 seconds, 90 seconds for main memory, 10 seconds for I/O. Assume main memory access improves by ~10% per year and I/O remains the same.

More information

Oracle Database 12c: JMS Sharded Queues

Oracle Database 12c: JMS Sharded Queues Oracle Database 12c: JMS Sharded Queues For high performance, scalable Advanced Queuing ORACLE WHITE PAPER MARCH 2015 Table of Contents Introduction 2 Architecture 3 PERFORMANCE OF AQ-JMS QUEUES 4 PERFORMANCE

More information

Join Processing for Flash SSDs: Remembering Past Lessons

Join Processing for Flash SSDs: Remembering Past Lessons Join Processing for Flash SSDs: Remembering Past Lessons Jaeyoung Do, Jignesh M. Patel Department of Computer Sciences University of Wisconsin-Madison $/MB GB Flash Solid State Drives (SSDs) Benefits of

More information

Attacking Highspeed Ethernet Links

Attacking Highspeed Ethernet Links Issued by: ERNW GmbH Enno Rey Breslauer Strasse 28 D-69124 Heidelberg www.ernw.de Page: 2 of 7 Content 1 Introduction... 3 1.1 Prerequisites... 3 2 Tackling the Problem... 4 2.1 Tool pcap_extractor...

More information

Optimizing Local File Accesses for FUSE-Based Distributed Storage

Optimizing Local File Accesses for FUSE-Based Distributed Storage Optimizing Local File Accesses for FUSE-Based Distributed Storage Shun Ishiguro 1, Jun Murakami 1, Yoshihiro Oyama 1,3, Osamu Tatebe 2,3 1. The University of Electro-Communications, Japan 2. University

More information

The Google File System. Alexandru Costan

The Google File System. Alexandru Costan 1 The Google File System Alexandru Costan Actions on Big Data 2 Storage Analysis Acquisition Handling the data stream Data structured unstructured semi-structured Results Transactions Outline File systems

More information

Parallelizing Inline Data Reduction Operations for Primary Storage Systems

Parallelizing Inline Data Reduction Operations for Primary Storage Systems Parallelizing Inline Data Reduction Operations for Primary Storage Systems Jeonghyeon Ma ( ) and Chanik Park Department of Computer Science and Engineering, POSTECH, Pohang, South Korea {doitnow0415,cipark}@postech.ac.kr

More information

Identifying Performance Bottlenecks with Real- World Applications and Flash-Based Storage

Identifying Performance Bottlenecks with Real- World Applications and Flash-Based Storage Identifying Performance Bottlenecks with Real- World Applications and Flash-Based Storage TechTarget Dennis Martin 1 Agenda About Demartek Enterprise Data Center Environments Storage Performance Metrics

More information

Storage Devices for Database Systems

Storage Devices for Database Systems Storage Devices for Database Systems 5DV120 Database System Principles Umeå University Department of Computing Science Stephen J. Hegner hegner@cs.umu.se http://www.cs.umu.se/~hegner Storage Devices for

More information

Achieve Optimal Network Throughput on the Cisco UCS S3260 Storage Server

Achieve Optimal Network Throughput on the Cisco UCS S3260 Storage Server White Paper Achieve Optimal Network Throughput on the Cisco UCS S3260 Storage Server Executive Summary This document describes the network I/O performance characteristics of the Cisco UCS S3260 Storage

More information

W H I T E P A P E R. Comparison of Storage Protocol Performance in VMware vsphere 4

W H I T E P A P E R. Comparison of Storage Protocol Performance in VMware vsphere 4 W H I T E P A P E R Comparison of Storage Protocol Performance in VMware vsphere 4 Table of Contents Introduction................................................................... 3 Executive Summary............................................................

More information

White Paper. EonStor GS Family Best Practices Guide. Version: 1.1 Updated: Apr., 2018

White Paper. EonStor GS Family Best Practices Guide. Version: 1.1 Updated: Apr., 2018 EonStor GS Family Best Practices Guide White Paper Version: 1.1 Updated: Apr., 2018 Abstract: This guide provides recommendations of best practices for installation and configuration to meet customer performance

More information

Entry-level Intel RAID RS3 Controller Family

Entry-level Intel RAID RS3 Controller Family PRODUCT Brief Entry-Level Intel RAID RS3 Controller Portfolio Entry-level Intel RAID RS3 Controller Family 12Gb/s connectivity and basic data protection RAID matters. Rely on Intel RAID. Cost-effective

More information

Accelerating Big Data: Using SanDisk SSDs for Apache HBase Workloads

Accelerating Big Data: Using SanDisk SSDs for Apache HBase Workloads WHITE PAPER Accelerating Big Data: Using SanDisk SSDs for Apache HBase Workloads December 2014 Western Digital Technologies, Inc. 951 SanDisk Drive, Milpitas, CA 95035 www.sandisk.com Table of Contents

More information

Using (Suricata over) PF_RING for NIC-Independent Acceleration

Using (Suricata over) PF_RING for NIC-Independent Acceleration Using (Suricata over) PF_RING for NIC-Independent Acceleration Luca Deri Alfredo Cardigliano Outlook About ntop. Introduction to PF_RING. Integrating PF_RING with

More information

Moonstone: A Framework for Accelerating Testing of Software

Moonstone: A Framework for Accelerating Testing of Software Moonstone: A Framework for Accelerating ing of Software Atsuji Sekiguchi, Tomohiro Ohtake, Toshihiro Shimizu, Yuji Hotta, Taichi Sugiyama, Takeshi Yasuie and Toshihiro Kodaka Cloud Computing Research Center

More information

Innovative Technologies Realizing Compact PRIMERGY TX120 Server

Innovative Technologies Realizing Compact PRIMERGY TX120 Server Innovative Technologies Realizing Compact PRIMERGY TX120 Server V Akinao Tanikawa V Tadashi Mantani V Noriaki Kobayashi (Manuscript received May 31, 2007) The need for internal control in enterprises reflects

More information

Evaluation of the Chelsio T580-CR iscsi Offload adapter

Evaluation of the Chelsio T580-CR iscsi Offload adapter October 2016 Evaluation of the Chelsio T580-CR iscsi iscsi Offload makes a difference Executive Summary As application processing demands increase and the amount of data continues to grow, getting this

More information

IBM MQ Appliance HA and DR Performance Report Model: M2001 Version 3.0 September 2018

IBM MQ Appliance HA and DR Performance Report Model: M2001 Version 3.0 September 2018 IBM MQ Appliance HA and DR Performance Report Model: M2001 Version 3.0 September 2018 Sam Massey IBM MQ Performance IBM UK Laboratories Hursley Park Winchester Hampshire 1 Notices Please take Note! Before

More information

SONAS Best Practices and options for CIFS Scalability

SONAS Best Practices and options for CIFS Scalability COMMON INTERNET FILE SYSTEM (CIFS) FILE SERVING...2 MAXIMUM NUMBER OF ACTIVE CONCURRENT CIFS CONNECTIONS...2 SONAS SYSTEM CONFIGURATION...4 SONAS Best Practices and options for CIFS Scalability A guide

More information

NEC Express5800 A2040b 22TB Data Warehouse Fast Track. Reference Architecture with SW mirrored HGST FlashMAX III

NEC Express5800 A2040b 22TB Data Warehouse Fast Track. Reference Architecture with SW mirrored HGST FlashMAX III NEC Express5800 A2040b 22TB Data Warehouse Fast Track Reference Architecture with SW mirrored HGST FlashMAX III Based on Microsoft SQL Server 2014 Data Warehouse Fast Track (DWFT) Reference Architecture

More information

CLOUD-SCALE FILE SYSTEMS

CLOUD-SCALE FILE SYSTEMS Data Management in the Cloud CLOUD-SCALE FILE SYSTEMS 92 Google File System (GFS) Designing a file system for the Cloud design assumptions design choices Architecture GFS Master GFS Chunkservers GFS Clients

More information

Sun Lustre Storage System Simplifying and Accelerating Lustre Deployments

Sun Lustre Storage System Simplifying and Accelerating Lustre Deployments Sun Lustre Storage System Simplifying and Accelerating Lustre Deployments Torben Kling-Petersen, PhD Presenter s Name Principle Field Title andengineer Division HPC &Cloud LoB SunComputing Microsystems

More information

Understanding SSD overprovisioning

Understanding SSD overprovisioning Understanding SSD overprovisioning Kent Smith, LSI Corporation - January 8, 2013 The over-provisioning of NAND flash memory in solid state drives (SSDs) and flash memory-based accelerator cards (cache)

More information

Multimedia Streaming. Mike Zink

Multimedia Streaming. Mike Zink Multimedia Streaming Mike Zink Technical Challenges Servers (and proxy caches) storage continuous media streams, e.g.: 4000 movies * 90 minutes * 10 Mbps (DVD) = 27.0 TB 15 Mbps = 40.5 TB 36 Mbps (BluRay)=

More information