In exploring SPEC.org's site for yesterday's blog I came across something interesting: specifically a configuration problem with a benchmark result for the dual Xeon, eight core, 2.66Ghz Dell 2950.
The composite score for this machine on SpecWeb_2005 is 16,830 - a number that puts it third from the top (behind a Fujitsu and the 20,235 achieved by an eight core, four Opteron, HP ProLiant DL585 G2).
The test date for this is given as "Nov 06", the hardware availability date is given as "Dec 06" and all of the software is listed as available. So, obviously, I wanted to know what the basic machine cost -because SPEC doesn't tell you.
According to the SPEC data report this machine had:
- Intel Xeon processor X5355
- Processor Speed (MHz) 2660
- # Processors 8 cores, 2 chips, 4 cores/chip
- Primary Cache 32KB(I) + 32KB(D) on chip, per core
- Secondary Cache 4MB on chip, per chip
- Memory 32GB (8x4GB) 667MHz ECC CL5 DDR2 FB-DIMM
- Disk Subsystem 4 x 73GB 15K RPM SAS, 55 x 36GB 15K RPM Fibre Channel
- Disk Controllers Integrated PERC5i/SAS, Emulex LPe-1150-E Fibre Channel
- Operating System Red Hat Enterprise Linux 4 AS Update 3 (default kernel 2.6.9-34.ELsmp)
- File System ext2
- Other Hardware 2 x PowerConnect 5224, PowerConnect 2624, 3 x PowerConnect 5012, Dell/EMC CX3-40 SAN Array
- Other Software Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_04-b05). Java HotSpot(TM)
- 64-Bit Server VM (build 1.5.0_04-b05, mixed mode)
What's interesting is that Dell does show a 2950 on its medium to large enterprise store site - but the unit can't be configured this way on-line. First, the fastest four core processor shown is only a " Quad Core Intel® Xeon® E5320, 2x4MB Cache, 1.86GHz, 1066MHz FSB" - not the Intel E5355 2.6Ghz machine shown on the SPEC report - and there is no obvious way to configure it with more than 16GB of 667Mhz memory.
More oddly, attempting to configure one with Red Hat 4/3 and 32GB of RAM produced these error messages:
The Operating System you selected supports a maximum of 16 GB Memory.
The Operating System you selected requires Broadcom Dual Port TCP/IP Offload Engine Not Enabled for Linux.
You may want to read that second one again: "The Operating System you selected [i.e. Linux] requires Broadcom Dual Port TCP/IP Offload Engine Not Enabled for Linux."
Since we're actually still in "Dec 06" what this suggests to me is that Dell's left hand is decidedly more enthusiastic about Linux in general and Red Hat in particular than its right - meaning in practice that they prefer seeing potential sales go away to correcting errors in their configurator.
- Dell's per transaction cost, based on partially configuring the available 1.86Ghz model ($36,247 with dual CPUs and 4 x 8GB RAM) has to be at least $2.15 per WebMark - and that's before disk, OS, networking, or upgrades to the configuration claimed.
- Fujitsu Siemens has similar machine, with a "Jan 07" availability date, and a better SpecWeb_2005 result: 18,160, but has a long standing policy of not making it easy to find out what anything costs.
- The comparable UltraSPARC T1 result, from a a 32GB/1.2Ghz machine is 14,001. That machine is openly available at a Sun store list price of $25,995 ($1.85 per WebMark) for the basic single processor, 32GB unit - and the software, or course, is free unless you want to buy support.
That result, incidently, was obtained before ZFS became available in the commercial product release -i.e. with traditional RAID hardware. Do it today, with ZFS on a pure JBOD, and you'd see better performance at a lower total cost.
The configuration issues aren't the only things that raise questions here. For example, the Sun machine recorded no validation errors versus 346 for the Dell (and 513 for the Fujitsu). More interestingly, the Sun machine's results are consistent across the three iterations allowed for each of the three benchmark components. On the banking test, for example, the Sun machine produced composite scores of 32157.5, 33290.2, and 33203.8 for a high - low difference of only 133 - 0.03%
In contrast both Xeon results, that for the Dell 2950 and that for the comparable Fujitsu, show enormous variation. Thus Dell's three banking runs produced scores of 40,333.6, 23,989.6, and 23,710.9 for a variation of 16,623 - 70% of the last two; Similarly, the eight core Fujitsu records scores of 68,659.4, 23,375.7, and 22,908.3 for a variation of 45,701 - very nearly twice the average (23,495) of all four low scores!
Since this isn't how real systems behave in production, the temptation is to recalculate the composite scores omitting consideration of the outlier results, but doing that assumes we know what's going on here: and I certainly don't - not with respect to these results, not with respect to improbable configurations, and not with respect to absurd error messages from Dell's configurator.