RDMA-like VirtIO Network Device for Palacios Virtual Machines
|
|
- Cynthia Robertson
- 5 years ago
- Views:
Transcription
1 RDMA-like VirtIO Network Device for Palacios Virtual Machines Kevin Pedretti UNM ID: CS-591 Special Topics in Virtualization May 10, 2012 Abstract This project developed an RDMA-like VirtIO network device for the Palacios virtual machine monitor that enables virtual machines running on the same host system to communicate with one another directly, without any intermediate memory copies. The project required gaining a deep understanding of the VirtIO interface, creating a Portals4-light VirtIO device in Palacios, adding VirtIO support to the Kitten operating system (which was used as the guest), and performing a simple benchmark analysis. The foundation created by this project will be used in future work to create a complete and highperformance virtual Portals4 network device for Palacios, as part of a funded DOE/ASCR X-stack project. 1 Introduction Virtualization is potentially useful for high-performance computing workloads [1], but only if virtual machines have access to near-native network performance. Direct device pass-through can provide that, but has the downside that only one virtual machine (VM) can use a pass-through device at a time, and it also greatly complicates VM checkpointing and migration. Virtualized network interfaces and networks avoid these problems, but typically use interfaces that are inappropriate for HPC (e.g., TCP/IP over Ethernet). This project takes a step in the direction of implementing an HPC-appropriate virtual network interface. The original project plan was to implement a virtual Portals network device, but time-constraints required the project scope to be scaled back. Instead, a simple RDMA-like VirtIO network device was developed for the Palacios [2] virtual machine monitor (VMM) that allows VMs running on the same host to communicate with one another directly, without any intermediate memory copies. The remainder of this report is organized as follows. Section?? describes the overall software architecture that was developed. Section 3 discusses the implementation and several interesting observations made along the way. Section 4 presents some initial performance results. Finally, Section 5 concludes the report. 2 Overall Architecture Figure 1 provides of an overview of the overall system. Two virtual machines are shown running on a Linux host. The virtual machine on the left is sending a buffer (labeled TX Buffer ) to a buffer in the virtual machine on the right (labeled RX Buffer ). To do this, the sending VM must generate a Portals Put command describing the send buffer, the target buffer, and the target Portals process ID (i.e., the Portals ID of the right VM). Once the command is constructed, the Portals driver (labeled P4 Driver ) writes the command to the virtual Portals network interface (labeled VirtIO P4 NIC ). This is accomplished via a VirtIO Ring [3]. A VirtIO ring consists of three parts: 1) a descriptor array (labeled Desc ), 2) an available descriptor ring (labeled Avail ), and 3) a used descriptor ring (labeled Used ). The Portals driver allocates a descriptor 1
2 Kitten Guest Kitten Guest P4 Put Cmd Buf P4 Driver TX Buffer RX Buffer Desc Avail Used VirtIO P4 NIC VirtIO P4 NIC NID_MAP[ ] Palacios Host Linux Kernel Figure 1: Block diagram of system. from the descriptor array, sets its address and length fields to map the Portals command buffer, adds the index of the descriptor to the available descriptor ring, kicks the virtual Portals network interface by issuing an IO port write, then polls the used descriptor ring until the index of the command descriptor appears. When the virtual Portals network interface receives the Put command, it looks up the target node via the global node ID map (labeled NID MAP ). Assuming the node is valid, it then uses the target VM s state structure to convert guest physical address of the receive buffer passed in by the sending VM to the appropriate host kernel virtual address. Similarly, the guest physical address of the transmit buffer must be converted to a host kernel virtual address using the sending VM s state structure. Once both addresses are known, the virtual Portals NIC uses memcpy() to copy the TX buffer to the RX buffer. The receiving VM can poll on the RX buffer to notice when the transfer is complete. The virtual Portals NIC takes care to ensure that the last byte of the transfer is truly written last, so polling on the last byte of the buffer is sufficient. A full Portals implementation would include event queues, which the receiver could wait on to detect completion. 3 Implementation All source code developed for this project is available on the class Wiki, at ssl/doku.php/bridges:classes:cs591:portals4. The majority of development effort was put into understanding the VirtIO queue and ring interfaces, and adding VirtIO driver support to the Kitten [4] lightweight kernel. Since the Portals VirtIO device is exposed as a PCI device, code had to be added to Kitten to enumerate the PCI bus and read each device s base address registers. For Palacios, the most important file to look at is lnx virtio portals4.c in 0001-VirtIO-Portals4.patch. This file implements the VMM-side of the VirtIO Portals device. Note, Portals4 is a misnomer here, as this isn t really a Portals4 device, but rather a simple shmem putmem() like mechanism for copying data between VMs on the same node. The APIs implemented are: 2
3 int PtlNIInit(unsigned int pid); int PtlGetId(unsigned int *nid, unsigned int *pid); int PtlPut(void *local addr, void *target addr, size t length, unsigned int target nid, unsigned int target pid); The PtlNIInit() call initializes the virtual Portals NIC. This adds the calling VM s state structure to the Portals NID MAP so other VMs can fund it. The PtlGetID() call returns a callers Portals Node ID and Portals Process ID. The PtlPut() call sends a buffer in the caller s address space to a buffer in a remote (target) address space (e.g., a process in another VM running on the same node). For Kitten, the most important file to look at is virtio portals4.c in 003-p4drv.patch. This file implements a Guest-side driver for the VirtIO Portals device. The driver initializes the device, then does a simple Ping-Pong test between two VMs running on the same node. Results are presented in Section 4. 4 Results Performance of the VirtIO Portals network interface was evaluated using a Ping-Pong benchmark. The Ping-Pong is performed between two hard-coded Portals NID/PID (Node ID / Process ID) pairs. Each guest has its own config file which defines a LNX VIRTIO PORTALS4 device with a hard-coded Portals NID. Node 4 is the pinger and Node 5 is the ponger. When each guest starts, its Portals driver calls PtlNIInit() with the Portals Process ID that it wants to be associated with this is hard-coded to be 16 for both nodes. Then, Portals NID/PID 4.16 sends a ping to is polling on memory. When it notices that the message has arrived, it sends a pong message back of the same size to 4.16, which also polls on memory to notice completion. The round-trip time is measured by The benchmark tests message sizes from 1 to 4 MB in power of two increments. Bandwidth results are shown in Figure 2. The VirtIO P4 line is the performance of the virtual Portals NIC when moving data between two virtual machines running on the same node. The v3 migrate core command was used to ensure that each VM was running on its own CPU, and the two CPUs used were selected to be on the same processor socket. Palacios s default behavior is to run both VMs on the same CPU, which was found to provide very poor performance due to the Ping-Pong benchmark s use of busy polling. The MPI line in Figure 2 is the performance of OpenMPI when sending MPI messages between two MPI native processes running on the same node. The MPI processes were bound to the same two CPUs that were used for the VirtIO Portals tests. The Intel Messaging Benchmarks(IMB) Ping-Pong benchmark was used for MPI testing. The VirtIO Portals bandwidth curve ramps up more slowly than MPI, but reaches a much higher peak bandwidth. This is likely due to MPI having an extra memory copy in the data path, where the sender first copies the message into a shared memory region and then the receiver copies the message out of the shared memory region. In contrast, the VirtIO Portals NIC copies directly from the send buffer to the receive buffer. MPI s double copy requires double the bandwidth compared to the VirtIO Portals NIC s single copy data path. For both MPI and VirtIO Portals, bandwidth drops significantly after the 1 MB message size. This is due to cache effects the Xeon processor that was used has a shared 4 MB level 3 cache. Figure 3 plots the latency for both the VirtIO Portals NIC and MPI. The higher overhead of the VirtIO Portals NIC is due to the extra overhead of entering and exiting into Palacios. Each VMExit was benchmarked to require around 5 microseconds of overhead. Assuming this is split evenly between entering and exiting Palacios, this likely accounts for the roughly 2.5 microsecond higher latency of the VirtIO Portals NIC. Error bars are shown for the VirtIO Portals NIC results in both figures. The reason for the outliers is not known at this time. The outliers are not repeatable, and move around from run to run. This suggests that host OS or VMM noise may be an issue. 3
4 VirtIO P4 MPI Bandwidth (MB/s) K 8K 32K 128K 512K 2M Message Size (Bytes) Figure 2: Bandwidth between two VMs/Processes on the same node. 20 VirtIO P4 MPI Latency (microseconds) K 2K 4K 8K 16K 32K Message Size (Bytes) Figure 3: Latency between two VMs/Processes on the same node. 4
5 5 Conclusion This project developed an RDMA-like VirtIO network device for the Palacios virtual machine monitor that enables virtual machines running on the same host system to communicate with one another directly, without any intermediate memory copies. The project required gaining a deep understanding of the VirtIO interface, creating a VirtIO Portals network interface device in Palacios, adding VirtIO support to the Kitten operating system (which was used as the guest), and performing a benchmark analysis using a Ping-Pong benchmark. The foundation created by this project will be used in future work to create a complete and high-performance virtual Portals4 network device for Palacios, as part of a funded DOE/ASCR X-stack project. References [1] W. Huang, J. Liu, B. Abali, and D. Panda, A case for high performance computing with virtual machines, in Proceedings of the 20th International Conference on Supercomputing (ICS), June [2] J. Lange, K. Pedretti, P. Dinda, P. G. Bridges, C. Bae, P. Soltero, and A. Merritt, Minimal-overhead virtualization of a large scale supercomputer, in Proceedings of the 2011 ACM SIGPLAN/SIGOPS International Conference on Virtual Execution Environments (VEE 11), (Newport Beach, CA), March [3] R. Russell, virtio: Towards a De-Facto Standard For Virtual I/O Devices, ACM SIGOPS Operating Systems Review, vol. 42, pp , July [4] Sandia National Laboratories, The Kitten Lightweight Kernel. kitten. 5
Implementing a GDB Stub in Lightweight Kitten OS
Implementing a GDB Stub in Lightweight Kitten OS Angen Zheng, Jack Lange Department of Computer Science University of Pittsburgh {anz28, jacklange}@cs.pitt.edu ABSTRACT Because of the increasing complexity
More informationImplementation and Analysis of Large Receive Offload in a Virtualized System
Implementation and Analysis of Large Receive Offload in a Virtualized System Takayuki Hatori and Hitoshi Oi The University of Aizu, Aizu Wakamatsu, JAPAN {s1110173,hitoshi}@u-aizu.ac.jp Abstract System
More informationInitial Performance Evaluation of the Cray SeaStar Interconnect
Initial Performance Evaluation of the Cray SeaStar Interconnect Ron Brightwell Kevin Pedretti Keith Underwood Sandia National Laboratories Scalable Computing Systems Department 13 th IEEE Symposium on
More informationTopic: A Deep Dive into Memory Access. Company: Intel Title: Software Engineer Name: Wang, Zhihong
Topic: A Deep Dive into Memory Access Company: Intel Title: Software Engineer Name: Wang, Zhihong A Typical NFV Scenario: PVP Guest Forwarding Engine virtio vhost Forwarding Engine NIC Ring ops What s
More informationA Case for High Performance Computing with Virtual Machines
A Case for High Performance Computing with Virtual Machines Wei Huang*, Jiuxing Liu +, Bulent Abali +, and Dhabaleswar K. Panda* *The Ohio State University +IBM T. J. Waston Research Center Presentation
More informationLiMIC: Support for High-Performance MPI Intra-Node Communication on Linux Cluster
LiMIC: Support for High-Performance MPI Intra-Node Communication on Linux Cluster H. W. Jin, S. Sur, L. Chai, and D. K. Panda Network-Based Computing Laboratory Department of Computer Science and Engineering
More informationFast packet processing in the cloud. Dániel Géhberger Ericsson Research
Fast packet processing in the cloud Dániel Géhberger Ericsson Research Outline Motivation Service chains Hardware related topics, acceleration Virtualization basics Software performance and acceleration
More informationMinimal-overhead Virtualization of a Large Scale Supercomputer
Minimal-overhead Virtualization of a Large Scale Supercomputer John R. Lange Department of Computer Science University of Pittsburgh Pittsburgh, PA 1526 jacklange@cs.pitt.edu Kevin Pedretti Scalable System
More informationNetchannel 2: Optimizing Network Performance
Netchannel 2: Optimizing Network Performance J. Renato Santos +, G. (John) Janakiraman + Yoshio Turner +, Ian Pratt * + HP Labs - * XenSource/Citrix Xen Summit Nov 14-16, 2007 2003 Hewlett-Packard Development
More information6.9. Communicating to the Outside World: Cluster Networking
6.9 Communicating to the Outside World: Cluster Networking This online section describes the networking hardware and software used to connect the nodes of cluster together. As there are whole books and
More informationVALE: a switched ethernet for virtual machines
L < > T H local VALE VALE -- Page 1/23 VALE: a switched ethernet for virtual machines Luigi Rizzo, Giuseppe Lettieri Università di Pisa http://info.iet.unipi.it/~luigi/vale/ Motivation Make sw packet processing
More informationMemory Management Strategies for Data Serving with RDMA
Memory Management Strategies for Data Serving with RDMA Dennis Dalessandro and Pete Wyckoff (presenting) Ohio Supercomputer Center {dennis,pw}@osc.edu HotI'07 23 August 2007 Motivation Increasing demands
More informationPerformance Evaluation of InfiniBand with PCI Express
Performance Evaluation of InfiniBand with PCI Express Jiuxing Liu Amith Mamidala Abhinav Vishnu Dhabaleswar K Panda Department of Computer and Science and Engineering The Ohio State University Columbus,
More informationSR-IOV Support for Virtualization on InfiniBand Clusters: Early Experience
SR-IOV Support for Virtualization on InfiniBand Clusters: Early Experience Jithin Jose, Mingzhe Li, Xiaoyi Lu, Krishna Kandalla, Mark Arnold and Dhabaleswar K. (DK) Panda Network-Based Computing Laboratory
More informationEfficient Shared Memory Message Passing for Inter-VM Communications
Efficient Shared Memory Message Passing for Inter-VM Communications François Diakhaté 1, Marc Perache 1,RaymondNamyst 2, and Herve Jourdren 1 1 CEA DAM Ile de France 2 University of Bordeaux Abstract.
More informationHPMMAP: Lightweight Memory Management for Commodity Operating Systems. University of Pittsburgh
HPMMAP: Lightweight Memory Management for Commodity Operating Systems Brian Kocoloski Jack Lange University of Pittsburgh Lightweight Experience in a Consolidated Environment HPC applications need lightweight
More informationElectrical Engineering and Computer Science Department
Electrical Engineering and Computer Science Department Technical Report NWU-EECS-11-7 July 14, 211 VNET/P: Bridging the Cloud and High Performance Computing Through Fast Overlay Networking Lei Xia Zheng
More informationReducing CPU and network overhead for small I/O requests in network storage protocols over raw Ethernet
Reducing CPU and network overhead for small I/O requests in network storage protocols over raw Ethernet Pilar González-Férez and Angelos Bilas 31 th International Conference on Massive Storage Systems
More informationEVALUATING INFINIBAND PERFORMANCE WITH PCI EXPRESS
EVALUATING INFINIBAND PERFORMANCE WITH PCI EXPRESS INFINIBAND HOST CHANNEL ADAPTERS (HCAS) WITH PCI EXPRESS ACHIEVE 2 TO 3 PERCENT LOWER LATENCY FOR SMALL MESSAGES COMPARED WITH HCAS USING 64-BIT, 133-MHZ
More informationDistributing Application and OS Functionality to Improve Application Performance
Distributing Application and OS Functionality to Improve Application Performance Arthur B. Maccabe, William Lawry, Christopher Wilson, Rolf Riesen April 2002 Abstract In this paper we demonstrate that
More informationCan Memory-Less Network Adapters Benefit Next-Generation InfiniBand Systems?
Can Memory-Less Network Adapters Benefit Next-Generation InfiniBand Systems? Sayantan Sur, Abhinav Vishnu, Hyun-Wook Jin, Wei Huang and D. K. Panda {surs, vishnu, jinhy, huanwei, panda}@cse.ohio-state.edu
More informationMVAPICH-Aptus: Scalable High-Performance Multi-Transport MPI over InfiniBand
MVAPICH-Aptus: Scalable High-Performance Multi-Transport MPI over InfiniBand Matthew Koop 1,2 Terry Jones 2 D. K. Panda 1 {koop, panda}@cse.ohio-state.edu trj@llnl.gov 1 Network-Based Computing Lab, The
More informationNon-blocking Java Communications Support on Clusters
Non-blocking Java Communications Support on Clusters Guillermo L. Taboada*, Juan Touriño, Ramón Doallo UNIVERSIDADE DA CORUÑA SPAIN {taboada,juan,doallo}@udc.es 13th European PVM/MPI Users s Meeting (EuroPVM/MPI
More informationMessage Passing Architecture in Intra-Cluster Communication
CS213 Message Passing Architecture in Intra-Cluster Communication Xiao Zhang Lamxi Bhuyan @cs.ucr.edu February 8, 2004 UC Riverside Slide 1 CS213 Outline 1 Kernel-based Message Passing
More informationIsoStack Highly Efficient Network Processing on Dedicated Cores
IsoStack Highly Efficient Network Processing on Dedicated Cores Leah Shalev Eran Borovik, Julian Satran, Muli Ben-Yehuda Outline Motivation IsoStack architecture Prototype TCP/IP over 10GE on a single
More informationVIDAS: OBJECT-BASED VIRTUALIZED D A T A S H A R I N G F O R H I G H PERFORMANCE STORAGE I/O
Grupo de Arquitectura de Computadores, Comunicaciones y Sistemas VIDAS: OBJECT-BASED VIRTUALIZED D A T A S H A R I N G F O R H I G H PERFORMANCE STORAGE I/O Pablo Llopis, Javier Garcia Blas, Florin Isaila,
More informationCOMPUTER ARCHITECTURE. Virtualization and Memory Hierarchy
COMPUTER ARCHITECTURE Virtualization and Memory Hierarchy 2 Contents Virtual memory. Policies and strategies. Page tables. Virtual machines. Requirements of virtual machines and ISA support. Virtual machines:
More informationAdvanced Computer Networks. End Host Optimization
Oriana Riva, Department of Computer Science ETH Zürich 263 3501 00 End Host Optimization Patrick Stuedi Spring Semester 2017 1 Today End-host optimizations: NUMA-aware networking Kernel-bypass Remote Direct
More informationHigh Performance VMM-Bypass I/O in Virtual Machines
High Performance VMM-Bypass I/O in Virtual Machines Jiuxing Liu Wei Huang Bulent Abali Dhabaleswar K. Panda IBM T. J. Watson Research Center 9 Skyline Drive Hawthorne, NY 532 {jl, abali}@us.ibm.com Computer
More informationHigh Performance Distributed Lock Management Services using Network-based Remote Atomic Operations
High Performance Distributed Lock Management Services using Network-based Remote Atomic Operations S. Narravula, A. Mamidala, A. Vishnu, K. Vaidyanathan, and D. K. Panda Presented by Lei Chai Network Based
More informationEfficient shared memory message passing for inter-vm communications
Efficient shared memory message passing for inter-vm communications François Diakhaté, Marc Pérache, Raymond Namyst, Hervé Jourdren To cite this version: François Diakhaté, Marc Pérache, Raymond Namyst,
More information10-Gigabit iwarp Ethernet: Comparative Performance Analysis with InfiniBand and Myrinet-10G
10-Gigabit iwarp Ethernet: Comparative Performance Analysis with InfiniBand and Myrinet-10G Mohammad J. Rashti and Ahmad Afsahi Queen s University Kingston, ON, Canada 2007 Workshop on Communication Architectures
More informationA Userspace Packet Switch for Virtual Machines
SHRINKING THE HYPERVISOR ONE SUBSYSTEM AT A TIME A Userspace Packet Switch for Virtual Machines Julian Stecklina OS Group, TU Dresden jsteckli@os.inf.tu-dresden.de VEE 2014, Salt Lake City 1 Motivation
More informationThe latency of user-to-user, kernel-to-kernel and interrupt-to-interrupt level communication
The latency of user-to-user, kernel-to-kernel and interrupt-to-interrupt level communication John Markus Bjørndalen, Otto J. Anshus, Brian Vinter, Tore Larsen Department of Computer Science University
More informationVirtualization and memory hierarchy
Virtualization and memory hierarchy Computer Architecture J. Daniel García Sánchez (coordinator) David Expósito Singh Francisco Javier García Blas ARCOS Group Computer Science and Engineering Department
More information2017 Storage Developer Conference. Mellanox Technologies. All Rights Reserved.
Ethernet Storage Fabrics Using RDMA with Fast NVMe-oF Storage to Reduce Latency and Improve Efficiency Kevin Deierling & Idan Burstein Mellanox Technologies 1 Storage Media Technology Storage Media Access
More informationHigh Performance MPI on IBM 12x InfiniBand Architecture
High Performance MPI on IBM 12x InfiniBand Architecture Abhinav Vishnu, Brad Benton 1 and Dhabaleswar K. Panda {vishnu, panda} @ cse.ohio-state.edu {brad.benton}@us.ibm.com 1 1 Presentation Road-Map Introduction
More informationGot Loss? Get zovn! Daniel Crisan, Robert Birke, Gilles Cressier, Cyriel Minkenberg, and Mitch Gusat. ACM SIGCOMM 2013, August, Hong Kong, China
Got Loss? Get zovn! Daniel Crisan, Robert Birke, Gilles Cressier, Cyriel Minkenberg, and Mitch Gusat ACM SIGCOMM 2013, 12-16 August, Hong Kong, China Virtualized Server 1 Application Performance in Virtualized
More informationThe influence of system calls and interrupts on the performances of a PC cluster using a Remote DMA communication primitive
The influence of system calls and interrupts on the performances of a PC cluster using a Remote DMA communication primitive Olivier Glück Jean-Luc Lamotte Alain Greiner Univ. Paris 6, France http://mpc.lip6.fr
More informationAccelerating NVMe I/Os in Virtual Machine via SPDK vhost* Solution Ziye Yang, Changpeng Liu Senior software Engineer Intel
Accelerating NVMe I/Os in Virtual Machine via SPDK vhost* Solution Ziye Yang, Changpeng Liu Senior software Engineer Intel @optimistyzy Notices & Disclaimers Intel technologies features and benefits depend
More informationXen Network I/O Performance Analysis and Opportunities for Improvement
Xen Network I/O Performance Analysis and Opportunities for Improvement J. Renato Santos G. (John) Janakiraman Yoshio Turner HP Labs Xen Summit April 17-18, 27 23 Hewlett-Packard Development Company, L.P.
More informationLive Virtual Machine Migration with Efficient Working Set Prediction
2011 International Conference on Network and Electronics Engineering IPCSIT vol.11 (2011) (2011) IACSIT Press, Singapore Live Virtual Machine Migration with Efficient Working Set Prediction Ei Phyu Zaw
More informationHIGH-PERFORMANCE NETWORKING :: USER-LEVEL NETWORKING :: REMOTE DIRECT MEMORY ACCESS
HIGH-PERFORMANCE NETWORKING :: USER-LEVEL NETWORKING :: REMOTE DIRECT MEMORY ACCESS CS6410 Moontae Lee (Nov 20, 2014) Part 1 Overview 00 Background User-level Networking (U-Net) Remote Direct Memory Access
More informationEvolution of the netmap architecture
L < > T H local Evolution of the netmap architecture Evolution of the netmap architecture -- Page 1/21 Evolution of the netmap architecture Luigi Rizzo, Università di Pisa http://info.iet.unipi.it/~luigi/vale/
More informationPerformance Analysis and Evaluation of Mellanox ConnectX InfiniBand Architecture with Multi-Core Platforms
Performance Analysis and Evaluation of Mellanox ConnectX InfiniBand Architecture with Multi-Core Platforms Sayantan Sur, Matt Koop, Lei Chai Dhabaleswar K. Panda Network Based Computing Lab, The Ohio State
More informationBetter than Native: Using virtualization to improve compute node performance
Better than Native: Using virtualization to improve compute node performance Brian Kocoloski John Lange {briankoco,jacklange}@cs.pitt.edu Department of Computer Science University of Pittsburgh Pittsburgh,
More informationEvaluating the Impact of RDMA on Storage I/O over InfiniBand
Evaluating the Impact of RDMA on Storage I/O over InfiniBand J Liu, DK Panda and M Banikazemi Computer and Information Science IBM T J Watson Research Center The Ohio State University Presentation Outline
More informationRDMA programming concepts
RDMA programming concepts Robert D. Russell InterOperability Laboratory & Computer Science Department University of New Hampshire Durham, New Hampshire 03824, USA 2013 Open Fabrics Alliance,
More informationSpeeding up Linux TCP/IP with a Fast Packet I/O Framework
Speeding up Linux TCP/IP with a Fast Packet I/O Framework Michio Honda Advanced Technology Group, NetApp michio@netapp.com With acknowledge to Kenichi Yasukata, Douglas Santry and Lars Eggert 1 Motivation
More informationThe LiMIC Strikes Back. Hyun-Wook Jin System Software Laboratory Dept. of Computer Science and Engineering Konkuk University
The LiMIC Strikes Back Hyun-Wook Jin System Software Laboratory Dept. of Computer Science and Engineering Konkuk University jinh@konkuk.ac.kr Contents MPI Intra-node Communication Introduction of LiMIC
More informationPerformance Evaluation of InfiniBand with PCI Express
Performance Evaluation of InfiniBand with PCI Express Jiuxing Liu Server Technology Group IBM T. J. Watson Research Center Yorktown Heights, NY 1598 jl@us.ibm.com Amith Mamidala, Abhinav Vishnu, and Dhabaleswar
More informationVARIABILITY IN OPERATING SYSTEMS
VARIABILITY IN OPERATING SYSTEMS Brian Kocoloski Assistant Professor in CSE Dept. October 8, 2018 1 CLOUD COMPUTING Current estimate is that 94% of all computation will be performed in the cloud by 2021
More informationExperience in Offloading Protocol Processing to a Programmable NIC
Experience in Offloading Protocol Processing to a Programmable NIC Arthur B. Maccabe, Wenbin Zhu Computer Science Department The University of New Mexico Albuquerque, NM 87131 Jim Otto, Rolf Riesen Scalable
More informationIdentifying the Sources of Latency in a Splintered Protocol
Identifying the Sources of Latency in a Splintered Protocol Wenbin Zhu, Arthur B. Maccabe Computer Science Department The University of New Mexico Albuquerque, NM 87131 Rolf Riesen Scalable Computing Systems
More informationOngoing work on NSF OCI at UNH InterOperability Laboratory. UNH IOL Participants
Ongoing work on NSF OCI-1127228 at UNH InterOperability Laboratory Robert D. Russell InterOperability Laboratory & Computer Science Department University of New Hampshire Durham, New Hampshire
More informationOne-Sided Append: A New Communication Paradigm For PGAS Models
One-Sided Append: A New Communication Paradigm For PGAS Models James Dinan and Mario Flajslik Intel Corporation {james.dinan, mario.flajslik}@intel.com ABSTRACT One-sided append represents a new class
More informationVirtualization, Xen and Denali
Virtualization, Xen and Denali Susmit Shannigrahi November 9, 2011 Susmit Shannigrahi () Virtualization, Xen and Denali November 9, 2011 1 / 70 Introduction Virtualization is the technology to allow two
More informationSharing High-Performance Devices Across Multiple Virtual Machines
Sharing High-Performance Devices Across Multiple Virtual Machines Preamble What does sharing devices across multiple virtual machines in our title mean? How is it different from virtual networking / NSX,
More informationResearch on the Implementation of MPI on Multicore Architectures
Research on the Implementation of MPI on Multicore Architectures Pengqi Cheng Department of Computer Science & Technology, Tshinghua University, Beijing, China chengpq@gmail.com Yan Gu Department of Computer
More informationEXTENDING AN ASYNCHRONOUS MESSAGING LIBRARY USING AN RDMA-ENABLED INTERCONNECT. Konstantinos Alexopoulos ECE NTUA CSLab
EXTENDING AN ASYNCHRONOUS MESSAGING LIBRARY USING AN RDMA-ENABLED INTERCONNECT Konstantinos Alexopoulos ECE NTUA CSLab MOTIVATION HPC, Multi-node & Heterogeneous Systems Communication with low latency
More informationHigh bandwidth, Long distance. Where is my throughput? Robin Tasker CCLRC, Daresbury Laboratory, UK
High bandwidth, Long distance. Where is my throughput? Robin Tasker CCLRC, Daresbury Laboratory, UK [r.tasker@dl.ac.uk] DataTAG is a project sponsored by the European Commission - EU Grant IST-2001-32459
More informationAgilio CX 2x40GbE with OVS-TC
PERFORMANCE REPORT Agilio CX 2x4GbE with OVS-TC OVS-TC WITH AN AGILIO CX SMARTNIC CAN IMPROVE A SIMPLE L2 FORWARDING USE CASE AT LEAST 2X. WHEN SCALED TO REAL LIFE USE CASES WITH COMPLEX RULES TUNNELING
More informationPARAVIRTUAL RDMA DEVICE
12th ANNUAL WORKSHOP 2016 PARAVIRTUAL RDMA DEVICE Aditya Sarwade, Adit Ranadive, Jorgen Hansen, Bhavesh Davda, George Zhang, Shelley Gong VMware, Inc. [ April 5th, 2016 ] MOTIVATION User Kernel Socket
More informationKVM PV DEVICES.
K DEVICES dor.laor@qumranet.com Agenda Kernel Virtual Machine overview Paravirtualized s intro & brief history VirtIO Enhanced VirtIO with K support 2 Kernel Virtual Machine overview is a regular Linux
More informationEnabling Fast, Dynamic Network Processing with ClickOS
Enabling Fast, Dynamic Network Processing with ClickOS Joao Martins*, Mohamed Ahmed*, Costin Raiciu, Roberto Bifulco*, Vladimir Olteanu, Michio Honda*, Felipe Huici* * NEC Labs Europe, Heidelberg, Germany
More informationIO virtualization. Michael Kagan Mellanox Technologies
IO virtualization Michael Kagan Mellanox Technologies IO Virtualization Mission non-stop s to consumers Flexibility assign IO resources to consumer as needed Agility assignment of IO resources to consumer
More informationIX: A Protected Dataplane Operating System for High Throughput and Low Latency
IX: A Protected Dataplane Operating System for High Throughput and Low Latency Adam Belay et al. Proc. of the 11th USENIX Symp. on OSDI, pp. 49-65, 2014. Presented by Han Zhang & Zaina Hamid Challenges
More informationBackground. IBM sold expensive mainframes to large organizations. Monitor sits between one or more OSes and HW
Virtual Machines Background IBM sold expensive mainframes to large organizations Some wanted to run different OSes at the same time (because applications were developed on old OSes) Solution: IBM developed
More informationMPI Programming Techniques
MPI Programming Techniques Copyright (c) 2012 Young W. Lim. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any
More informationKVM PV DEVICES.
K DEVICES dor.laor@qumranet.com 1 Agenda Introduction & brief history VirtIO Enhanced VirtIO with K support Further implementation 2 General & history Fully virtualized devices performs bad 55 Mbps for
More informationImproving performance of Virtual Machines by Virtio bridge Bypass for PCI devices
www.ijecs.in International Journal Of Engineering And Computer Science ISSN:2319-7242 Volume 6 Issue 4 April 2017, Page No. 20931-20937 Index Copernicus value (2015): 58.10 DOI: 10.18535/ijecs/v6i4.24
More informationMulti-Threaded UPC Runtime for GPU to GPU communication over InfiniBand
Multi-Threaded UPC Runtime for GPU to GPU communication over InfiniBand Miao Luo, Hao Wang, & D. K. Panda Network- Based Compu2ng Laboratory Department of Computer Science and Engineering The Ohio State
More informationImproving Application Performance and Predictability using Multiple Virtual Lanes in Modern Multi-Core InfiniBand Clusters
Improving Application Performance and Predictability using Multiple Virtual Lanes in Modern Multi-Core InfiniBand Clusters Hari Subramoni, Ping Lai, Sayantan Sur and Dhabhaleswar. K. Panda Department of
More informationAC: COMPOSABLE ASYNCHRONOUS IO FOR NATIVE LANGUAGES. Tim Harris, Martín Abadi, Rebecca Isaacs & Ross McIlroy
AC: COMPOSABLE ASYNCHRONOUS IO FOR NATIVE LANGUAGES Tim Harris, Martín Abadi, Rebecca Isaacs & Ross McIlroy Synchronous IO in the Windows API Read the contents of h, and compute a result BOOL ProcessFile(HANDLE
More informationChangpeng Liu. Cloud Storage Software Engineer. Intel Data Center Group
Changpeng Liu Cloud Storage Software Engineer Intel Data Center Group Notices & Disclaimers Intel technologies features and benefits depend on system configuration and may require enabled hardware, software
More informationreferences Virtualization services Topics Virtualization
references Virtualization services Virtual machines Intel Virtualization technology IEEE xplorer, May 2005 Comparison of software and hardware techniques for x86 virtualization ASPLOS 2006 Memory resource
More informationCommunication has significant impact on application performance. Interconnection networks therefore have a vital role in cluster systems.
Cluster Networks Introduction Communication has significant impact on application performance. Interconnection networks therefore have a vital role in cluster systems. As usual, the driver is performance
More informationBest 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 informationIBM POWER8 100 GigE Adapter Best Practices
Introduction IBM POWER8 100 GigE Adapter Best Practices With higher network speeds in new network adapters, achieving peak performance requires careful tuning of the adapters and workloads using them.
More informationPCI Express x8 Single Port SFP+ 10 Gigabit Server Adapter (Intel 82599ES Based) Single-Port 10 Gigabit SFP+ Ethernet Server Adapters Provide Ultimate
NIC-PCIE-1SFP+-PLU PCI Express x8 Single Port SFP+ 10 Gigabit Server Adapter (Intel 82599ES Based) Single-Port 10 Gigabit SFP+ Ethernet Server Adapters Provide Ultimate Flexibility and Scalability in Virtual
More informationMay 1, Foundation for Research and Technology - Hellas (FORTH) Institute of Computer Science (ICS) A Sleep-based Communication Mechanism to
A Sleep-based Our Akram Foundation for Research and Technology - Hellas (FORTH) Institute of Computer Science (ICS) May 1, 2011 Our 1 2 Our 3 4 5 6 Our Efficiency in Back-end Processing Efficiency in back-end
More informationMeasurement-based Analysis of TCP/IP Processing Requirements
Measurement-based Analysis of TCP/IP Processing Requirements Srihari Makineni Ravi Iyer Communications Technology Lab Intel Corporation {srihari.makineni, ravishankar.iyer}@intel.com Abstract With the
More informationChangpeng Liu. Senior Storage Software Engineer. Intel Data Center Group
Changpeng Liu Senior Storage Software Engineer Intel Data Center Group Legal Notices and Disclaimers Intel technologies features and benefits depend on system configuration and may require enabled hardware,
More informationAbstract. Testing Parameters. Introduction. Hardware Platform. Native System
Abstract In this paper, we address the latency issue in RT- XEN virtual machines that are available in Xen 4.5. Despite the advantages of applying virtualization to systems, the default credit scheduler
More informationLinux 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 informationLighting the Blue Touchpaper for UK e-science - Closing Conference of ESLEA Project The George Hotel, Edinburgh, UK March, 2007
Working with 1 Gigabit Ethernet 1, The School of Physics and Astronomy, The University of Manchester, Manchester, M13 9PL UK E-mail: R.Hughes-Jones@manchester.ac.uk Stephen Kershaw The School of Physics
More informationOptimizing TCP Receive Performance
Optimizing TCP Receive Performance Aravind Menon and Willy Zwaenepoel School of Computer and Communication Sciences EPFL Abstract The performance of receive side TCP processing has traditionally been dominated
More informationRack Disaggregation Using PCIe Networking
Ethernet-based Software Defined Network (SDN) Rack Disaggregation Using PCIe Networking Cloud Computing Research Center for Mobile Applications (CCMA) Industrial Technology Research Institute 雲端運算行動應用研究中心
More informationPM2: High Performance Communication Middleware for Heterogeneous Network Environments
PM2: High Performance Communication Middleware for Heterogeneous Network Environments Toshiyuki Takahashi, Shinji Sumimoto, Atsushi Hori, Hiroshi Harada, and Yutaka Ishikawa Real World Computing Partnership,
More informationAn FPGA-Based Optical IOH Architecture for Embedded System
An FPGA-Based Optical IOH Architecture for Embedded System Saravana.S Assistant Professor, Bharath University, Chennai 600073, India Abstract Data traffic has tremendously increased and is still increasing
More informationPerformance Considerations of Network Functions Virtualization using Containers
Performance Considerations of Network Functions Virtualization using Containers Jason Anderson, et al. (Clemson University) 2016 International Conference on Computing, Networking and Communications, Internet
More informationDPDK Integration within F5 BIG-IP BRENT BLOOD, SR MANAGER SOFTWARE ENGINEERING VIJAY MANICKAM, SR SOFTWARE ENGINEER
x DPDK Integration within F5 BIG-IP BRENT BLOOD, SR MANAGER SOFTWARE ENGINEERING VIJAY MANICKAM, SR SOFTWARE ENGINEER F5 Company Snapshot Founded: 1996 IPO: June 1999 Employees: 4,395 Headquarters: Seattle,
More informationFUJITSU PHI Turnkey Solution
FUJITSU PHI Turnkey Solution Integrated ready to use XEON-PHI based platform Dr. Pierre Lagier ISC2014 - Leipzig PHI Turnkey Solution challenges System performance challenges Parallel IO best architecture
More informationAchieving Performance Isolation with Lightweight Co-Kernels
Achieving Performance Isolation with Lightweight Co-Kernels ABSTRACT Jiannan Ouyang, Brian Kocoloski, John Lange Department of Computer Science University of Pittsburgh {ouyang, briankoco, jacklange} @cs.pitt.edu
More informationDesign challenges of Highperformance. MPI over InfiniBand. Presented by Karthik
Design challenges of Highperformance and Scalable MPI over InfiniBand Presented by Karthik Presentation Overview In depth analysis of High-Performance and scalable MPI with Reduced Memory Usage Zero Copy
More informationLinux Operating System
Linux Operating System Dept. of Computer Science & Engineering 1 History Linux is a modern, free operating system based on UNIX standards. First developed as a small but self-contained kernel in 1991 by
More informationDirected Point: An Efficient Communication Subsystem for Cluster Computing. Abstract
Directed Point: An Efficient Communication Subsystem for Cluster Computing Chun-Ming Lee, Anthony Tam, Cho-Li Wang The University of Hong Kong {cmlee+clwang+atctam}@cs.hku.hk Abstract In this paper, we
More informationIntroduction to Ethernet Latency
Introduction to Ethernet Latency An Explanation of Latency and Latency Measurement The primary difference in the various methods of latency measurement is the point in the software stack at which the latency
More informationThe NE010 iwarp Adapter
The NE010 iwarp Adapter Gary Montry Senior Scientist +1-512-493-3241 GMontry@NetEffect.com Today s Data Center Users Applications networking adapter LAN Ethernet NAS block storage clustering adapter adapter
More informationSupporting Soft Real-Time Tasks in the Xen Hypervisor
Supporting Soft Real-Time Tasks in the Xen Hypervisor Min Lee 1, A. S. Krishnakumar 2, P. Krishnan 2, Navjot Singh 2, Shalini Yajnik 2 1 Georgia Institute of Technology Atlanta, GA, USA minlee@cc.gatech.edu
More information