LEVERAGING EMC FAST CACHE WITH SYBASE OLTP APPLICATIONS

Transcription

1 White Paper LEVERAGING EMC FAST CACHE WITH SYBASE OLTP APPLICATIONS Abstract This white paper introduces EMC s latest innovative technology, FAST Cache, and emphasizes how users can leverage it with Sybase ASE 15.5 database applications. This white paper covers several use cases that were tested in EMC labs, along with their test results. This paper also discusses best practices for implementing FAST Cache with Sybase OLTP database applications. February 2011

2 Copyright 2011 EMC Corporation. All Rights Reserved. EMC believes the information in this publication is accurate of its publication date. The information is subject to change without notice. The information in this publication is provided as is. EMC Corporation makes no representations or warranties of any kind with respect to the information in this publication, and specifically disclaims implied warranties of merchantability or fitness for a particular purpose. Use, copying, and distribution of any EMC software described in this publication requires an applicable software license. For the most up-to-date listing of EMC product names, see EMC Corporation Trademarks on EMC.com. All other trademarks used herein are the property of their respective owners. Part Number h

3 Table of Contents Executive summary... 4 The need for FAST Cache... 5 Introduction... 6 Technology overview... 6 EMC FAST Cache technology... 6 Sybase Adaptive Server Enterprise... 7 Applying FAST Cache technology with a Sybase ASE 15.5 database... 8 Application I/O characteristics... 8 Understanding the database working set... 8 Benchmark setup... 8 Workload description... 9 Use case descriptions Baseline Use Case Use Case Results summary Conclusion References

4 Executive summary EMC unified storage platform arrays continue to provide increased storage utilization and optimization, enhanced capabilities, and greater interoperability and security, as well as multiple ease-of-use improvements. In 2008, EMC introduced Flash drive technology to its midrange storage arrays. EMC FAST Cache technology introduced at FLARE 30 leverages the use of Flash drives to add an extra layer of cache between DRAM cache and rotating spindles, which improves I/O service responsiveness. By effectively extending the caching of a database application s hot data on Flash drives, FAST Cache reduces data access latencies to the most frequently accessed business-critical data, thus improving overall application response times. As the focus of business-critical changes, the data that becomes the most important, and most frequently used, will be automatically cached in these Flash drives. Hence, the business applications will be taking full advantage of the investments made in these Flash drives, delivering optimal application service time improvement and improving the TCO of the solution. EMC FAST Cache technology is immediately available on the EMC unified storage platform. This white paper addresses the consideration for deploying FAST Cache with Sybase ASE OLTP applications. An understanding of the principles discussed in this paper will allow users to efficiently implement FAST Cache with Sybase ASE databases. Figure 1. FAST Cache 4

5 The need for FAST Cache Sybase ASE Online Transaction Processing (OLTP) database applications tend to be mission-critical and have stringent I/O latency requirements. Sybase ASE server performance depends heavily on the I/O subsystem. The ultimate limit of throughput and IOPS for ASE server data read and write is typically limited by disk I/O. Physical I/O performance can make or break a mission-critical Sybase ASE database application due to its huge impact on the overall user experience. Traditionally, these OLTP databases are deployed on a huge number of rotating Fibre Channel (FC) spindles that may be short stroked to meet the low I/O latency requirement. Therefore, when OLTP databases are deployed on a large number of rotating spindles, the effective capacity utilization of these spindles is very low, thus increasing TCO. Over the past few decades, the disk drive technology has been hardpressed to keep up with the rapid speed increase in CPU technology. In real-world applications, even the fastest rotating spindle with a 15k rpm speed performs only around 200 random 8K IOPS in most application workloads. Flash drives are a disruptive technology that significantly improves IOPS but comes at a relative higher cost. In an environment where storage resources are shared among various applications, addressing unexpected workload spikes becomes critical. FAST Cache technology creates a faster medium for storing frequently accessed data on Flash drives. This significantly reduces the need to buy more Flash drives and provides the Flash latencies to the most frequently accessed data for the business at any particular moment. The hot data of the database application is constantly cached in FAST Cache automatically and transparently as the business data need changes, thus eliminating the need for users to manually classify the hot data. To make space for more hot data chunks, the LRU mechanism is used to flush cold data to the backend spindles. Once the hot data is cached by FAST Cache, the FAST Cache handles the majority of I/O and the underlying rotating spindles receive very few IOPS. This effect enables these possibilities: You can deploy the database on fewer rotating spindles. You can deploy the database on slower spindles like 10k rpm drives or SATA drives. You can utilize more of the rotating spindles available capacity without significantly impacting overall application latency. FAST Cache can be easily expanded as operational needs evolve. This incremental approach allows customers to start with only a small initial investment. FAST Cache automatically absorbs unpredicted spikes in application workloads. Therefore, FAST Cache provides an optimal I/O solution to meet Sybase ASE performance needs for all read and write workloads. 5

6 Introduction The main goal of this paper is to introduce EMC FAST Cache technology and explain how it can leverage Sybase ASE OLTP applications. This white paper details the positive effects of deploying EMC FAST Cache. Some of these benefits include improved TPMs, generally accompanied by corresponding average transaction service latency reduction. The paper covers the test configurations and performance results of various tests conducted at EMC engineering labs. The testing procedure involved establishing the baseline by running the database benchmark on pure rotating Fibre Channel drives, comparing these baseline numbers with application metrics after adding FAST Cache to the storage subsystem, and enabling it on the pool with all database LUNs where Sybase devices reside. Technology overview EMC FAST Cache technology In traditional storage arrays, the DRAM caches are relatively too small to be able to maintain the hot data for long periods of time. Very few storage arrays enable users to nondisruptively expand DRAM cache, even if they support DRAM cache expansion. FAST Cache technology builds a faster medium (sort of cache) for data on Flash drives, thereby allowing the hotter data to stay longer on a faster medium. FAST Cache tracks the number of hits at a 64 KB chunk size and promotes the chunks to the Flash tier once the chunks get three or more hits. After a data chunk gets promoted to FAST Cache, the subsequent accesses to that chunk of data will be served at Flash latencies. Eventually when the hotness of the data chunk decreases, the data chunk is evicted from FAST Cache and replaced by new, hot data. FAST Cache uses a simple least re-used (LRU) mechanism to evict the data chunks; only when it makes room for additional hot data has it met the access threshold. A complete introduction to FAST Cache technology is beyond the scope of this document. This paragraph will provide a brief overview of FAST Cache. FAST Cache is built on the premise that overall application latencies can improve when most frequently accessed data is maintained on a relatively smaller sized, but faster, storage medium, such as Flash drives. FAST Cache technology identifies the frequently accessed data used by currently running business applications and moves it to Flash drives automatically and nondisruptively. The data movement is completely transparent to applications, thereby making this technology application-agnostic and management-free. It is as simple as turning a switch on or off. For example, you can enable or disable the FAST Cache on any pool simply by selecting or clearing the FAST Cache checkbox settings. This is shown in Figure 2. 6

7 Figure 2. How to enable FAST Cache The LUN/pool-level granularity enables FAST Cache to be selectively enabled on a few LUNs or pools within a storage array depending on the application performance requirements and SLAs. Sybase Adaptive Server Enterprise Sybase Adaptive Server Enterprise (ASE) is a database management system (DBMS) designed to support transaction-intensive, mission-critical OLTP, decision support, and mixed-load applications. Adaptive Server also manages multiple databases and users, keeps track of the data s location on disks, maintains the mapping of the logical data description to the physical data storage, and maintains data and procedure caches in the memory. ASE 15 meets the demands of large databases and high transaction volumes while providing a DBMS. The most recent release, ASE 15.5, adds several features that provide immense value to enterprise-class customers. Its key features include on-disk encryption, smart partitions, In-Memory databases and Incremental data transfer utility, multiple Tempdb groups, efficient backup compression, deferred name resolution for user-defined stored procedures, new query processing technology that demonstrates a significant increase in performance, and enhanced support for unstructured data management. 7

8 Applying FAST Cache technology with a Sybase ASE 15.5 database Users need to thoroughly understand application workload characteristics before implementing any storage technology, including FAST Cache. While comprehending basic information like workload I/O characteristics is important, it is vital that users understand the active working set of the workload for right-sizing FAST Cache. The following sections explain in detail how to determine application I/O characteristics and the application working set. Application I/O characteristics The relative improvement from any faster technology is dependent on whether the application is bottlenecked by the storage subsystem. For Sybase DBAs, it is frequently easier to work with storage administrators and turn to the I/O subsystem to avoid I/O bottlenecks than to redesign the application to minimize the amount of inefficient or extra data access by the database system. Nevertheless, some amount of analysis at the database level should be done to identify I/O bottlenecks, and to ascertain that the proper application of FAST Cache leveraging Flash drives will in fact be worthwhile. Understanding the database working set This step is critical to maximize the benefits from FAST Cache. Generally, FAST Cache is the most suitable technology for workloads with small random I/O and relatively small working sets that fit in the FAST Cache nicely. Most Sybase OLTP databases tend to be highly random in nature with small working sets as compared to the total database size. Normally, databases have datasets with varied I/O patterns because of the following reasons: OLTP databases tend to be temporal in nature, as the most recent data is more important than older data. The subset of data that is most often needed by the business applications currently running is generally referred to as a working set of a database. The relative importance of data changes from object to object. Some tables may be accessed more often than some other tables. The number of IOPS per gigabyte size of an object, also known as object intensity, changes quite significantly. The best example is comparing the index to the table in a database. The relative IOPS received by a database block occupied by an index object are very high compared to the IOPS received by a database block consumed by a table object. Benchmark setup To characterize FAST Cache with Sybase ASE OLTP workloads, the following benchmark setup is used: The benchmark testing is done on a single Sybase ASE 15.5 database deployed on a CLARiiON CX4-960 storage array and Dell R601 server. The hardware configuration is shown in the figures below. 8

9 Back-end Connectivity Front-end Connectivity SP A SP B Dell R610 Brocade 5100B Bus 3 Bus 2 Bus 0 Bus 1 CLARiiON CX4-960 CLARiiON CX GB Solid State Drives 600 GB Fibre Channel Drives Figure 3. Hardware setup In the figure, all the available four back-end buses on the CX4-960 are used just to ensure that the benchmark is not bus-bound. There is no real requirement to utilize all back-end buses to realize FAST Cache benefits. Configure your back-end buses depending on I/O requirements and following EMC best practices for bus and disk layout. The Flash drives were placed randomly but evenly on four disk enclosures to take full bandwidth advantage of all four buses. Workload description The database workload simulates an OLTP application (Order Entry) environment fashioned after the business transaction profile described in the popular industry TPC-C benchmark. The test workload is TPC-C-like (but not strictly TPC-C specification conforming), implemented to include a mixture of read-only and update-intensive transactions. The Sybase ASE database devices were distributed against LUNs in such a way that both storage processors handle an equal number of I/Os. To ensure that FAST Cache can handle workloads involving a high write mix, EMC created a suitable test scenario. Generally, most of the OLTP databases tend to have a read/write mix of 70/30. The Sybase OLTP database benchmark used in the test has the characteristics shown in the following table. 9

10 Table 1. Workload characteristics Property Value / Description Database size 750 GB Warehouses 10,000 Database version Sybase ASE 15.5 Read / write ratio 70 / 30 Database metric TPM Number of 200 concurrent users Storage pools SybaseASE, SybaseLog Total LUNs 22 LUNs 5 x 100 GB LUNs for database devices and segments; 1 x 300 GB LUN for a Log device; 8 x 50 GB LUNs for a Stock table; 4 x 100 GB LUNs for a Customer table; and 4 x 100 GB LUNs for an OrderLine table EMC kept consistent workload characteristics for all the use cases described next so users can analyze the application metrics. Use case descriptions EMC conducted the following tests to understand the impact of FAST Cache on Sybase ASE OLTP databases and to compare the impact of FAST Cache on the workload using the application-level and database-level metrics. The goal of these use cases is to show that FAST Cache can significantly reduce the number of IOPS received by rotating spindles once the hot data is cached by FAST Cache. The reduction in IOPS of the rotating spindle count opens up several possibilities either to create a configuration with a fewer number of rotating Fibre Channel drives or to completely replace the rotating Fibre Channel drives with SATA drives. The following use cases are crafted to demonstrate this very basic point. Baseline: The baseline metric is established on a storage pool of 45 x 600 GB Fibre Channel drives. Use Case 1: Enable FAST Cache with 8 x 73 GB Flash drives to the baseline environment. Use Case 2: Repeat Use Case 1 where 45 x 600 GB Fibre Channel drives are replaced by 25 x 600 GB Fibre Channel drives and front-ended by FAST Cache created on 8 x 73 GB Flash drives. Baseline The baseline was established on all Fibre Channel rotating drives in this configuration. There was a total of 45 x 600 GB FC disks used to create two storage pools (SybaseASE, SybaseLog) in this setup. Then, 22 LUNs (see Table 1) were carved out to store a Sybase ASE database master device, log device, and segments. Since the Stock table, Customer table, and OrderLine table are bigger tables and will be frequently accessed during TPCC-C testing, the Stock table was stored on a 10

11 stripped logical volume created on 8 x 50 GB LUNs and similarly the Customer and OrderLine tables were stored on a stripped logical volume created on 4 x 100 GB LUNs and on 4 x 100 GB LUNs, respectively. This configuration helps in distributing I/O load equally among storage processors. The next figure shows the exact disk and database layout for this use case. Physical Layer Storage Pool 1- SybaseASE {40 * 600 GB FC disks} Logical Layer 100 GB LUN for Sybase ASE Master device 100 GB LUN for TPCC Orders table Segment 100 GB LUN for TPCC History table Segment 100 GB LUN for TPCC Cache Segment 100 GB LUN for TPCC Cust_Index Segment 4 * 100 GB LUNs Stripped Logical Volume for OrderLine table 4 * 100 GB LUNs Stripped Logical Volume for Customer table 8 * 50 GB LUNs Stripped Logical Volume for Stock table Storage Pool 2- Sybase Log { 5 FC drives} 1 * 300 GB LUN Sybase ASE Log Disk Figure 4. Baseline disk and database layout 11

12 Use Case 1 The database layout for in this use case is identical to the baseline scenario except that FAST Cache (read/write) was created on 8 x 73 GB Flash drives and was enabled on the storage pool containing Sybase devices and segments. FAST Cache was not enabled on the storage pool having a LUN for ASE s log device because you will yield better results by enabling FAST Cache on a storage pool with LUNs for database devices and segments. After FAST Cache was enabled for Sybase ASE, the frequently accessed data by the TPC-C test was cached by FAST Cache. This warming up of FAST Cache resulted in improved latencies to the most accessed data, thereby enhancing overall application performance. This configuration obtained ~2.5 times the number of transactions as compared to the baseline number of transactions after four 2-hour runs of the TPC-C test. This significant improvement in transactions was without performing any kind of performance tuning on the Sybase ASE 15.5 database. This is a significant return on a small investment made into FAST Cache. The next figures show the impact of FAST Cache warm-up on the TPM rate after four 2-hour runs of the TPC-C test Transactions per Minute (FC 45) Use case 1 (FC 45) 1st TPC-C Run (FAST Cache) 2nd TPC-C Run (FAST Cache) 3rd TPC-C Run (FAST Cache) 4th TPC-C Run (FAST Cache) * Test duration= 2hr/Run Figure 5. FAST Cache warm-up 12

13 Relative Transactions per Minute (FC45) Relative Transactions per Minute FC Without FAST Cache With FAST Cache (Read- Write) Figure 6. Impact on FAST Cache on the TPM rate Before enabling FAST Cache, the server running the Sybase OLTP application was I/Obound with a disk wait of 25 percent. Once FAST Cache was enabled for the storage pool with database segments, the disk wait decreased to 7 percent, which eventually improved server CPU utilization. These results revealed that Sybase OLTP application performance can be improved significantly just by enabling FAST Cache and without performing any fine-tuning at the OLTP application level. Use Case 2 Based on the performance data from Use Case 1, the 45 underlying rotating FC drives were replaced with 25 FC drives of the same size and speed. FAST Cache (read/write) was also created on 8 x 73 GB Flash drives and was enabled on the storage pool with database segments and not on the storage pool with a LUN for the ASE log disk. This change reduces the disk costs by almost 50 percent. You can lower costs even further when reductions in the data center footprint, power, and cooling are taken into consideration. Reducing the number of rotating Fibre Channel drives certainly decreases the capacity you can use, with OLTP databases; however, the entire capacity of drives was never used in real-world deployments because short-stroking was typically done to guarantee very low latencies. FAST Cache technology enables applications to tap into the available capacity of these drives for the first time. By implementing FAST Cache, the rotating medium receives only a small fraction of overall I/Os and the application will not be majorly impacted by those few extra disk seeks. This allows more workloads to be consolidated to the same set of available 13

14 spindles. A careful analysis needs to be done to determine the right number of rotating spindles before attempting consolidation. Figure 7 shows the exact disk and database layout for this use case. Physical Layer Storage Pool 1- SybaseASE {20 * 600 GB FC disks} Logical Layer 100 GB LUN for Sybase ASE Master device 100 GB LUN for TPCC Orders table Segment 100 GB LUN for TPCC History table Segment 100 GB LUN for TPCC Cache Segment 100 GB LUN for TPCC Cust_Index Segment 4 * 100 GB LUNs Stripped Logical Volume for OrderLine table 4 * 100 GB LUNs Stripped Logical Volume for Customer table 8 * 50 GB LUNs Stripped Logical Volume for Stock table Storage Pool 2- Sybase Log { 5 FC drives} 1 * 300 GB LUN Sybase ASE Log Disk Figure 7. Use Case 2 disk and database layout The Use Case 2 (25 x 600 GB rotating Fibre Channel) configuration without FAST Cache resulted in a 39 percent drop of TPMs as compared to the baseline configuration. The next figure revealed the impact of decreasing the number of spindles on the TPM rate. 14

15 Without FAST Cache With FAST Cache (Read-Write) FC 45 FC 25 Figure 8. Impact of the number of spindles on the TPM rate The 25 x 600 GB rotating Fibre Channel configuration with FAST Cache yields the same TPMs as that of the 45 x 600 GB Fibre Channel drive configuration with FAST Cache. This leads to a reduction in TCO by 25 percent, which justifies the added cost of FAST Cache. Finally, this solution achieves approximately 2.5 times the number of transactions compared to the baseline configuration after the warm-up of FAST Cache. The next figure shows the impact of FAST Cache on the ASE OLTP application s TPM rate while FAST Cache was warming up Transactions per Minute (FC 25) Use case 2 (FC 25) 1st TPC-C Run (FAST Cache) 2nd TPC-C Run (FAST Cache) 3rd TPC-C Run (FAST Cache) 4th TPC-C Run (FAST Cache) * Test duration= 2hr/Run Figure 9. TPM vs. FAST Cache warm-up 15

16 Results summary Figure 10 represents a comparison of TCO and TPMs for Use Case 1 and Use Case 2. TCO advantage of FAST Cache for Sybase 120% 100% 100% 100% 100% 80% 75% 60% FC 45 (Use Case 1) FC 25 (Use Case 2) 40% 20% 0% Price (TCO) TPMs (FAST Cache enabled) Figure 10. The TCO advantage of FAST Cache The figure indicates that irrespective of the underlying number of spindles, once the hot data is cached by FAST Cache, the Use Case 1 and Use Case 2 configurations support the same amount of TPMs and Use Case 2 improves TCO. It is important to note that the application is able to perform almost 2.5 times the number of transactions with enabled FAST Cache. The increased transactions and reduced latency allow the application to scale much better and improve the end-user experience significantly. Conclusion In an environment where storage resources are shared among various applications, enabling FAST Cache on a particular OLTP application will result in improved I/O performance for that period of time interval. FAST Cache technology significantly reduces the TCO of a database solution by reducing the investment for the Flash tier and improving the capacity utilization of the rotating drives. The application-agnostic nature of FAST Cache technology makes it applicable to any application with a low I/O latency requirement. EMC unified storage arrays support various sizes of FAST Cache; therefore, the right-sizing of FAST Cache is very important to maximize the benefit. 16

17 Even though FAST Cache technology can be used with any application with any type of I/O pattern, it is especially suited for applications with small random I/Os and relatively small working sets compared to the total database size. The following best practices should be followed to realize the full advantage of FAST Cache: Disable FAST Cache on the storage pool that does not require it. Disable FAST Cache on the storage pool where ASE s transaction logs reside. Enabling FAST Cache on the transaction logs may or may not help, depending on workload characteristics. References The following white papers and documents are available on EMC.com and Powerlink : EMC CLARiiON and Celerra Unified FAST Cache - A Detailed Review Leveraging EMC CLARiiON CX4 with Enterprise Flash Drives for SAP Deployments Maximize the Performance Benefit from Enterprise Flash Drives by Sharing through Virtual Provisioning in EMC CLARiiON CX4 EMC CLARiiON Virtual LUN Technology - A Detailed Review EMC CLARiiON Virtual Provisioning Applied Technology EMC CLARiiON CX4 Model 960 Networked Storage System Specification Sheet Introduction to the EMC CLARiiON CX4 Series Featuring UltraFlex Technology EMC CLARiiON Best Practices for Performance and Availability: Release 30.0 Firmware Update The following can be found on the Sybase website: Monitoring performance with sp_sysmon /html/monitoring/x99294.htm Solid State Disks: Why are you still using metal platters??? 17