The Four Hundred

Six months ago, I told you about a 32-way Itanium-based server from NEC that showed Unix-class performance and price/performance running the forthcoming Windows Server 2003 Datacenter Edition. Now NEC is showing off the much-improved performance of the "AzuzA" NEC Express5800/1320Xc server, which has tweaked hardware and software. The 433,108 transactions per minute (TPM) throughput of the box has just about given the iSeries and Unix crowds heart attacks. Scalability and performance on big database jobs was the key differentiator between Unix and Windows servers.

I said was intentionally. NEC has just made that distinction a thing of the past, and as NEC rolls out Linux on the box--as it intends to do--the pressure will be even greater on the Unix crowd, which still accounts for about 45 percent of server sales a year. This year, Intel-based machines are expected to tie or surpass the aggregate Unix server market in terms of sales; Intel-based machines already account for upward of 90 percent of server shipments worldwide.

Way back when the first-generation "Merced" Itaniums made their debut, in 2000, a 16-way version of the AzusA machines was demonstrated with the 733 MHz and 800 MHz Merceds. These 16-way servers were sold by NEC in Japan and by Hewlett-Packard in North America and Europe, running HP's HP-UX 11i variant of the Unix operating system or the open-source Linux operating system. HP reckoned that the 16-way machine using an 800 MHz Merced could do about 140,000 TPM on the TPC-C test and about 51 gigaflops on supercomputer workloads. That puts the first generation of AzusA servers in the same power class as the most powerful Unix and proprietary servers (such as IBM's AS/400 and iSeries line and its S/390 and zSeries mainframe line) of about 18 months earlier, when Unix machines with 24 or 32 processors were able to hit around 150,000 TPM.

Back at the fall Intel Developer Forum, NEC was showing off an early version of the AzuzA machine using the "McKinley" Itanium 2 processors. This AzuzA machine was sold in Japan under the TX7 brand name (mostly to supercomputer customers) and in North America under the unwieldy Express5800/1320Xc brand name. NEC's test results on this 32-way TX7 server, which at the time only supported HP-UX and Linux with support for Windows Server 2003 promised when Microsoft got it out the door, posted an impressive 308,621 TPM on the TPC-C test at a cost of $14.96 per TPM. That NEC TX7 server was equipped with Itanium 2 chips with 3 MB L3 caches, 256 GB of main memory, and 23.6 TB of disk capacity. That TX7 server cost around $1.4 million, including the license for Windows Server 2003 Datacenter Edition, and the storage cost just under $2 million. SQL Server cost another $529,312, and client hardware and three years of support brought the total cost to $5.3 million, a figure that dropped to $4.6 million after a 13 percent discount. This was Unix-class performance at a substantial price break compared to Unix, especially given that the Unix servers pushed by IBM and Hewlett-Packard using the TPC-C tests have deep discounts ranging from 20 to 40 percent in their price/performance calculations. At list price--the price customers pay to upgrade their machines, more or less--an Itanium-based server like the earlier TX7 offers considerably better bang for the buck. Think about that for a second.

Now consider that the latest test on this same AzuzA server, using the same 1 GHz McKinley processors with 3 MB of L3 cache, posted a 40 percent gain in performance, with a 21 percent increase in system cost, due mainly to doubling main storage from 256 to 512 GB, nearly doubling disk storage to 40.9 TB (necessary because of TPC-C rules about scaling simulated users and storage), and using improved disk arrays with bigger cache memories. The latest AzuzA server in the TPC-C test had a list price of $2.17 million, with three years of maintenance costing $1.16 million. The prior TX7 configuration pegged the lost price of the 32-way machine at $1.41 million with $983,965 for three years of maintenance. How much of the price increase in the new TPC-C configuration is due to doubling memory and how much is due to a big price increase is unclear. But it looks like NEC, seeing the performance of the box, has raised its list prices on the AzuzA boxes. The reason why it probably felt justified is that the improved storage arrays and switches that are part of the TPC-C configuration are less expensive. SQL Server 2000 Enterprise Edition cost the same on this machine, at $529,312. The resulting machine, with client hardware and software thrown in, yielded a price/performance of $12.98 per TPM after discounts, down 13 percent compared with the tests NEC ran last fall using Windows and SQL Server.

Here's some iSeries perspective. By my estimates, the fastest iSeries Model 890 server, a 32-way machine using 1.3 GHz Power4 processors, can probably hit around 350,000 TPM on the TPC-C test, maybe 400,000 with some database tuning and large memory support (like 512 GB of main memory, which is not yet supported on the box). The same machine with a pSeries 690 label running AIX and Oracle has been tested and came in at just under 430,000 TPM with that half-terabyte of main memory recently.

The only Unix server that can beat these numbers out there in the market is the Fujitsu Siemens "Kaiser" PrimePower 2000 server, a 128-way behemoth that runs Sun Microsystems' Solaris Unix variant. That PrimePower machine, using 563 MHz Sparc64-GP processors, was able to handle 455,818 TPM on the TPC-C test using an obscure database called SymfoWare from Fujitsu. Fujitsu-Siemens will begin shipping 1.3 GHz Sparc64-GP processors this quarter, and it is likely that the PrimePowers will be able to hit 900,000, or maybe even 1 million, TPM. IBM is shooting for something in the range of 1 to 2 million TPM with its Power5 servers early next year, and HP is probably going to push 1 million TPM with its 64-way "Pinnacles" servers running the dual-core PA-8800 RISC processors by late summer (that's an effective 128-way processing); in early 2004, when the "Hondo" mx2 dual "Madison" Itanium 2 processor modules are ready, HP could hit this performance level with its Itanium boxes, too. However, NEC will probably only be able to hit somewhere between 600,000 and 650,000 TPM, with the 32-way AzuzA machines using the 1.5 GHz Madison processors when they come out in mid-2003. IBM and HP eventually will be able to field bigger iron. But NEC has what is effectively the best Itanium server in the market right now, and one that, for processor-based database and middleware pricing schemes, makes it a convincing alternative to RISC-Unix servers.

This is exactly the moment that Microsoft and Intel have been waiting for since the 1993 launch of Windows New Technology. The fifth time, thanks in large measure to 64-bit processing on database transactions, has been the charm.

The AzuzA servers come in three flavors. They all support hardware partitioning at the cell-board level, with a four-way cell board being the basic component of the server, and they can run either Linux or Windows Server 2003. The Express5800/1080Rc is an eight-way Itanium 2 server with 16 GB of main memory and 26 PCI I/O slots; it fits into an 8U rack-mounted chassis. The Express5800/1160Xc is a 16-way machine that supports up to 128 GB of main memory and has 56 I/O slots. The Express5800/1320Xc is a 32-way server that has up to 512 GB of main memory and 112 I/O slots. The latter two configurations come with their own stand-alone chassis. Both the 900MHz/1.5 MB L3 cache and 1 GHz/3 MB L3 cache versions of the McKinley Itanium processors are available on these servers, and the future 1.5 GHz/6 MB L3 Madison, due in the summer, and the faster Madisons with the expected 9 MB L3 cache will also plug into these machines when they are available early next year. The HP mx2 dual Madison modules will not, however, plug into the AzuzA frames, unless NEC and HP work out a licensing deal, since this is HP technology. NEC may just come up with its own way of plugging two chips into one slot, now that HP has shown the proof of concept. We'll see.

Sponsored By

MIDRANGE PERFORMANCE GROUP

Performance Navigator ® at Boise Cascade Office Products

Doug Mewmaw, Technical Support Manager for Boise Cascade Office Products, manages an iSeries environment that includes 14 separate servers. Doug's responsibilities include performance management and reporting. For several years, management was receiving regular weekly reports on the performance of each system. This time consuming task fell to Doug. Until March of 2000, Doug used a combination of IBM's Performance Tools, and various desktop tools such as spreadsheets and graphing packages. In March of 2000, Doug attended COMMON, and registered for a vendor course offering that discussed Performance Navigator.

Doug immediately saw the huge impact this product would have in his operation. "With Performance Navigator, everything is a simple point-and-click" said Mewmaw. He continues- "What I used to have to do involved 6 or 7 steps for each graph, for each system. With PerfNav, I can set many of the reports up on a scheduler, and they're done automatically." Mewmaw estimated that he used to spend over a day each week doing routine reporting. Now that is reduced to an hour or two.

Performance Navigator also provides Boise Cascade with the ability to constantly tune their configurations for optimum return on investment. Consolidations of multiple workloads onto a single server can be modeled in minutes, with complete confidence in the results. Increasing the workload of a specific job can be modeled to provide clear impact on future growth. As Doug Mewmaw said, "If Performance Navigator says we have a bottleneck, we almost certainly do. Better yet, PerfNav can tell us which users or which jobs are causing a performance problem."

Another issue that Performance Navigator addressed was Mewmaw's comfort level. "Best/1® has always been my performance and capacity planning tool. To be honest, I've not only felt frustration with the 'hardware promoting' tool, but I have never felt 100 percent confident in its predictions. I hate that. After switching to Performance Navigator, I don't have that uneasy feeling anymore. I really think this product will become the new industry standard".

Mewmaw continues, "MPG's support is just terrific, however there is so much that Performance Navigator does I actually went to their headquarters for a 2 day course. Even though I had worked with the product almost daily for over a year, I learned an incredible amount in those 2 days."

Free download at
www.mpginc.com