Partner Solution Case Study
/ / New Database Benchmark Hits Enterprise “Sweet Spot”
Overview
Country or Region:United States
Industry:Computer Software
Partner Profile
Network Appliance, Inc. (NetApp) is a world leader in unified storage solutions for today's data-intensive enterprise.
Business Situation
NetApp, IBM, and Microsoft realized that together they had the capability to establish enterprise-level database performance with a much less expensive system than had been required in the past.
Solution
Together, the three companies constructed a reference system and conducted a TPC-C transaction processing benchmark.
Benefits
Enterprise-level performance
Relatively low cost
Rapid database restore
Multiple snapshot states
Transparent disk recovery / “By using NetApp FAS3000 series storage systems, IBM xSeries servers, and Microsoft SQL Server 2005, customers can enjoy enterprise-class performance at relatively low cost.”
Patrick Rogers, V.P. Partners and Products, Network Appliance
Network Appliance, IBM, and Microsoft have collaborated on a TPC-C transaction processing benchmark for a 16-way, 64-bit Intel Xeon MP-based computer using Microsoft® SQL Server™ 2005 Enterprise Edition (64-bit). The system, priced at a total of U.S.$3.14 million, achieved 492,307 tpmC, with a cost per tpmC of $6.37. This benchmark demonstrates the performance of SQL Server 2005 on NetApp RAID-DP storage systems and IBM’s new eServer xSeries 460. In addition, it shows how SQL Server 2005 is able to exploit technology advances from key vendors like Intel, IBM, and Network Appliance to deliver advances in performance and cost efficiency.
Situation
In the highly competitive world of database transaction processing, the din of marketing claims and counter-claims for “the fastest transaction-processing system” was quieted in the late 1980s by the formation of the Transaction Processing Performance Council (TPC). Using a series of benchmarks and guidelines for its conduct, reporting, and auditing, the TPC ( eventually succeeded in enforcing fair benchmarking in the database transaction processing space.
The premier benchmark for transaction processing systems is now the TPC-C, originally approved in 1992. The first TPC-C result, published in September 1992 was 54 tpmC result with a cost per tpmC of U.S.$188,562, which means that that the total cost of the hardware and software required was a little over $10 million.
The current high TPC-C result for Microsoft® SQL Server™ 2005 is 1.08 million tpmC, with a cost per tpmC of $5.38. This result comes from a HP Integrity Superdome system, with a total system cost of $5.8 million. The Superdome has 64 Intel Itanium 2 processors, and runs Microsoft SQL Server 2005 Enterprise Edition (64-bit) on Microsoft Windows Server™ 2003 Datacenter x64 Edition. SQL Server 2005 and Windows Server 2003 are part of Microsoft Windows Server System™ integrated server software.
Using the new IBM eServer xSeries 460 with 16 dual-core, 64-bit Intel Xeon MP processors, and Network Appliance’s FAS3050 Enterprise Storage System, IBM, Network Appliance (NetApp), and Microsoft realized that it might be possible to achieve almost half the tpmC throughput of the Superdome—492 thousand tpmC—on a much less expensive system that is compatible with more standard Microsoft Windows® applications.
NetApp is a world leader in unified storage solutions for today's data-intensive enterprise. Since its inception in 1992, Network Appliance has delivered technology, product, and partner firsts that simplify data management. NetApp storage solutions include specialized hardware, software, and services, providing seamless storage management for open network environments.
IBM strives to lead in the invention, development and manufacture of the industry's most advanced information technologies, including computer systems, software, storage systems and microelectronics.
SQL Server 2005, the next-generation Microsoft data management and analysis software, delivers increased scalability, availability, and security to enterprise data and analytical applications while making them easier to create, deploy, and manage.
Each of the three critical components of this combination contributes to the cost reduction: the 16-processor eServer xSeries 460 is significantly less costly than the 64-processor Superdome; the SQL Server database is licensed per-processor on dual-core processors such as those in the IBM eServer xSeries 460, which effectively doubles the software price/performance; and, finally, the NetApp FAS3050 storage system can deliver high-performance enterprise-class storage at a reasonable price.
The use of dual-core Intel Xeon processors means that 16 processors represent 32 processor cores. Each HT-enabled core supports two threads, making 64 threads available to the database server. SQL Server 2005 has been tuned to take full advantage of large numbers of threads.
Seeing the potential synergy, the three companies agreed to collaborate on a TPC-C benchmark for such a system.
Solution
“This change improves the SQL Server 2005 product, since dual-core processors and hyperthreading will be dominant technologies in the future.”Susan Goodwin, Senior Performance Engineer (IBM xSeries), IBM
The IBM eServer xSeries 460, running Windows Server 2003 Datacenter x64 Edition, is the main database server. As configured for this benchmark, it has 17 Fibre Channel cards, which are directly connected to 17 NetApp FAS3050 storage systems. Sixteen of the FAS3050 storage systems hold 84 72GB 15K RPM disks each. These 16 systems hold the actual database files.
The seventeenth storage system holds the database logs. This storage system has 56 144GB 15K RPM disks.
The SQL Server database is partitioned into two file groups. Each file group is divided into 16 database files, so that the entire file group is spread evenly across the 16 storage systems used to hold the database files. Therefore, this yields a total of 32 Fibre Channel logical units that store all of the data needed for the TPCC database. An additional logical unit stores all of the log information for the SQL database on the seventeenth storage system.
“To get to this configuration, NetApp started with its tpmC performance goals, in combination with latency requirements, and figured out what that meant in terms of I/O operations per second; that number, in turn, determined the number of disks and how they could be allocated among the storage systems,” explained Sanjay Gulabani, Senior Performance Engineer, Network Appliance.
There are 16 IBM eServer x336 computers that act as client systems to drive the load to the xSeries 460. The client computers are connected to a 48-port Cisco 4948 switch using one copper-based Gigabit Ethernet connection. The xSeries 460 also has four copper-based Gigabit Ethernet connections going into that same switch. This is how the generated load for the transactions makes its way to the database server.
The IBM xSeries 460 runs the Microsoft Windows Server 2003 Datacenter x64 Edition operating system, and the RTE and client computers run the Windows Server 2003 Standard Edition operating system.
The systems were assembled at a NetApp laboratory in Research Triangle Park, North Carolina. The benchmarks were run from an IBM facility in Kirkland, Washington, as well as the NetApp facility in Research Triangle Park, North Carolina. To accommodate the remote operation, the systems were built into special racks with remotely operated power switches.
Once the system was assembled, the team built a “virgin” database for the TPC-C test, in a matter of some 52 hours, and took a snapshot of the database. The ability to quickly restore the initial database from the snapshot between test runs facilitated the system tuning required to reach the performance goals. Loading databases from snapshots is a new feature of SQL Server 2005; this was boosted by the NetApp Snapshot feature of the storage system.
Tuning proceeded in two stages. In the first stage, tests were run in a two-tier client-server configuration, and the hardware was adjusted between tests. In the second stage, tests were run in the official three-tier configuration required by the TPC-C, with RTE machines controlling clients, and the clients in turn performing transactions on the server.
One of the challenges with a benchmark this large is simply the volume of moving parts. All told there were over 1,400 disk drives employed on the database server. With this many drives, a disk failure is simply a matter of time. “All other TPC-C benchmarks use minimal RAID protection to drive down the cost, making life very difficult for the benchmark team,” said Susan Goodwin, Senior Performance Engineer (IBM xSeries),IBMCenter for Microsoft Technologies. “Loss of a drive typically means hours of lost productivity while a database restore is run. The TPC-C benchmark rules require that data is recoverable, but they don’t specify how quickly.”
The NetApp technology provided the benchmark system with inexpensive, redundant, high-performance storage and insulated the benchmark team from the inevitable failures of individual disks. Goodwin said there were no detectable problems with the NetApp storage system. “We never would have known that there had been any disk failures unless NetApp had told us.”
NetApp, however, noticed a completely different problem that was unrelated to the disk failures: periodic spikes in latency. NetApp proactively found an anomaly in the version of its Data ONTAP software used for testing, verified a fix, and implemented it. “That had minimal impact to the database server, didn’t corrupt the database, and only caused a few minutes of delay to reboot the system,” said Goodwin.
“Using NetApp storage technology made it possible to work within a very compressed schedule,” said Goodwin.
The IBM team made changes to some disk driver parameters, and implemented some SQL Server configuration parameters for client connection affinity, which is a standard feature of SQL Server 2005. In addition, the team adjusted the SQL Servermultitasking algorithm. “We had to adjust some latch parameters in SQL Server,” said Goodwin, “because the SQL Server algorithm was designed and tested for 64 physical single-core CPUs, while we have only 16 dual-core processors with hyperthreading enabled, giving 64 logical processors. This change improves the SQL Server 2005 product, since dual-core processors and hyperthreading will be dominant technologies in the future.”
Every SQL Server 2005 customer using large numbers of dual-core hyper-threaded processors will benefit from this optimization, which was made during the course of this benchmark test.
Once the performance goals were reached, the audited TPC-C tests were performed, yielding the result of 492,307 tpmC, with a cost per tpmC of $6.37. The total cost of the test system, as computed according to the TPC’s accounting standards, was $3.14 million.
Benefits
The strong results demonstrated a new price point for enterprise-class transaction-processing databases. “By using NetApp FAS3000 series storage systems, IBM xSeries servers, and Microsoft SQL Server 2005, customers can enjoy enterprise-class performance at relatively low cost,” said Patrick Rogers, V.P. Partners and Products, Network Appliance. “The powerful combination of NetApp, IBM, and Microsoft means that customers can run mission-critical applications on SQL Server 2005 and enjoy high-performance storage, computing and database functionality without major capital outlays.”
In this benchmark, SQL Server 2005 demonstrated its ability to exploit leading-edge server hardware, such as dual core processors and hyperthreading. The price for a SQL Server 2005 configuration across 64 “virtual” processors is the same as the price for SQL Server 2005 configuration on 16 single-core, single-threaded processors. As far as the software cost goes, the benchmark showed four times the performance for almost no additional cost.
In addition, the benchmark showed how SQL Server 2005 can exploit inexpensive, high-performance, high-reliability data stores such as the NetApp FAS3000 Series storage systems. New features such as fast database Snapshot copies and restores mean that customers can guarantee quick database recovery from natural and artificial disasters.
Finally, the benchmark demonstrated great SQL Server 2005 performance on IBM’s eServer xSeries line of servers. Using standard Intel Xeon processors, which are compatible with standard x86 applications, this system hits a “sweet spot” for enterprise database servers.
------
Performance is an Internal Throughput Rate (ITR) ratio based on measurements and projections using standard benchmarks in a controlled environment. The actual throughput that any user will experience will vary depending upon considerations such as the amount of multiprogramming in the user's job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve throughput improvements equivalent to the performance ratios stated here.
Microsoft Windows Server System
Microsoft Windows Server System is a line of integrated and manageable server software designed to reduce the complexity and cost of IT. Windows Server System enables you to spend less time and budget on managing your systems so that you can focus your resources on other priorities for you and your business.
For more information about Windows Server System, go to:
Microsoft SQL Server 2005
Microsoft SQL Server 2005 is comprehensive, integrated data management and analysis software that enables organizations to reliably manage mission-critical information and confidently run today’s increasingly complex business applications. By providing high availability, security enhancements, and embedded reporting and data analysis tools, SQL Server 2005 helps companies gain greater insight from their business information and achieve faster results for a competitive advantage. And, because it’s part of Windows Server System, SQL Server 2005 is designed to integrate seamlessly with your other server infrastructure investments.
For more information about SQL Server 2005, go to: