Last week, IBM announced a world-record industry-standard performance benchmark result using Red Hat Enterprise Linux 5. This benchmark follows a long line of Red Hat Enterprise Linux world-record results published recently. Here are some highlights.
To give credit where it’s due, we should emphasize that these benchmarks are completed by Red Hat partners. This includes leading hardware OEMs such as Dell, Fujitsu, HP and IBM. Also, some benchmarks require the use of a database – recent results include both IBM DB2 and Oracle database products. Red Hat’s role is to work with these partners to assist them with tuning and configuration of the Red Hat Enterprise Linux platform.
For Red Hat, working with partners to deliver world-record benchmarks helps us create a better product, which in turn benefits our customers – so, thank you, to all our benchmark partners!
Four specific benchmark areas are worth highlighting in this blog:
- Last week’s benchmark from IBM announces the delivery of world-record TPC-C performance for a 16-thread system – an IBM x3850 M2 server configured with four quad-core x86-64 processors. The benchmark used IBM’s DB2 9.5 64-bit database and achieved 516,752 tpmC (at a cost of $2.59 USD/tpmC, with a system availability date of March 14, 2008). That more than half a million transactions/minute can be delivered by a system with such a small physical footprint is astonishing.
- Late last year, HP published the world-record TPC-H performance result for the 300GB database size, based on a BladeSystem ProLiant BL480c configured as a cluster with 16 dual core processors. The result was 40,411 Queries/hour @300GB. This result highlights the performance capabilities of modern blade cluster configurations.
In fact, Red Hat dominates the TPC-H 300GB database performance category. Six of the top seven results are held by Red Hat Enterprise Linux. Meanwhile, the smaller database category (100GB) is a focus area for Microsoft. And many operating systems, including Red Hat Enterprise Linux, publish results in the larger categories (1000GB and over).
- Believe it or not, as of today, the top 13 (yes, 13!) SPECweb2005 results are all held by Red Hat Enterprise Linux. These results have been achieved on a range of HP, Dell and Fujitsu systems. The latest result, the first ever to exceed 30,000 connections, used an HP ProLiant DL580 G5 server, with 4 quad core chips (for 16 threads) and 64GB of memory.
- One of the more interesting results published recently was posted by HP and Oracle. This was for the world-record TPC-C price/performance result. Historically, this category of results has been dominated by Microsoft SQLserver configurations due to differences in database and support pricing. Oracle’s result, published in June this year, ranked Oracle Enterprise Linux with Oracle 10G as the leading TPC-C price/performance solution. As you will know, Oracle Enterprise Linux is a clone of Red Hat Enterprise Linux and, if you delve into the benchmark documentation, you will find that Oracle actually used Red Hat Enterprise Linux 4 to achieve this result. Interestingly, Microsoft trumped the result three months later using the same hardware configuration, coming in with 1.5 percent improved performance and a 1c/transaction price advantage (with system availability targeted for the end of this year). As you can see, this category of results is highly competitive! Of course, Red Hat Enterprise Linux 5 would be used for the benchmark today, which we would expect to deliver even higher performance.
Perhaps the most important takeaway from all of this information is that these results truly demonstrate the ability of Red Hat Enterprise Linux to support large SMP systems and high I/O throughputs. Historically, Linux was considered suitable for small 2-4 way SMP configurations, but the majority of recent results have been achieved using servers with four quad-core processors (in other words, 16-way SMP). And this is with industry-standard, commodity Intel and AMD chips, not specialized architectures. Benchmarks such a TPC-C are highly I/O intensive (the IBM result mentioned above was delivering over 8,000 transactions per second) and Red Hat Enterprise Linux has proven itself quite capable of handling the demands of these very high I/O loads.