Slashdot Mirror
IBM Launches p690
edyavno writes: "IBM just announced the launch of their new high-end Unix server p690. It's based on its new Power 4 chip, and is in the same category as just announced Sun's SunFire 15K. It also includes some mainframe level features and can be used either as a single large server or divided into up to 16 "virtual" servers, running any combination of AIX 5L and Linux. Here's yahoo article, and here it is from IBM itself."
I dont agree with your facts and also I *hate* sun.
"First of all, let me mention that the RS/6k S80 (two releases ago - prior to the p680) outdid the 10k, at reduced cost, with the previous generation of procs."
This is bullcrap. IBM was able to get a high TPC-C score. TPC-C is a really lame, 6-year database benchmark designed when the typical RISC CPU ran at 80 MHz, had 256KB cache, 4GB RAM max, less than a 1 GB/sec backplane, and storage was done on 5,400 RPM SCSI-2 drives.
The S80 got a high score because it's CPU had a very short pipeline (5-stage) which a database benchmark with a lot of locality benefits from tremedously. Another benefit was the S80's CPUs had twice the cache as the rest of the industry. Throw in some really good database benchmark engineers and you can really tune for TPC-C's locality. Also, the S80 could hold as memory as the E10K, so a big Oracle SGA means fewer disk accesses.
Then with the p680, IBM boosted performance by turning on an experimental CPU feature called Hardware Multi Threading. Real cool, but no datacenter manager in his right mind would run HMT on a production machine.
What all this proved was nothing. When you looked at real-world bechmarks such as Oracle Applications, PeopleSoft, SAP, etc., Sun servers with 400 MHz UltraSPARC II CPUs performed as well as, and sometimes better than IBM boxes with 450 MHz CPUs.
The whole pitch IBM was making that the S80's 450 MHz CPUs were equal to three 400 MHz E10K CPUs were absolute fraudulent. Fortunately, in my job, I got the chance to blow IBM's story out of the water. There was no more fun to see an IBM sales rep at a total loss for words in front of our CIO when I called him on the 3 IBM CPUs = 1 Sun CPU. Of course, when the CIO said: "You know, I really don't mind marketing, but I can't stand lying." The IBM guy started sweating.
I guess the reality was we didn't really like IBM anyway, were mostly a Sun and HP shop. But the way IBM muscled their way into the opportunity really pissed us all off, but we decided to give them a shot. But they made EMC sales reps look like kindergarten teachers.
"16 processors outdid 64 in many, many tests (including ones with real-world data movement)."
Total BS. IBM did no S80/p680 benchmarks with "real-world data movement" because the S80 has a big bottleneck called the I/O hub, which is by design limited to 1 GB/sec input max. So IBM did no TPC-H or similar data warehousing benchmark. In fact, the last one IBM did was an old SP2 result.
"As for partitioning... hmmm... let's think. IBM has been doing logical partitioning in AS/400 for a while, and on the S/390 (now the z-series) for quite some time... a few decades now. A lot of that experience went into this."
Decades, huh? The S/390 got LPARS in 1987. That is 14 years ago. That is 1.4 decades, not a "few decades". Maybe it is the IBM math that is the problem. Sort of like how a 32-CPU p690 becomes a 16 CPU p690 when running HPC code.
Most of the customers at this level are more interested in hypereliability than hyperperformance. And most are not going to trust version 1.0 of AIX LPARs (actually AIX VM Guest OS would be a more accurate mainframe feature comparison).
That said I don't doubt you got a really powerful CPU is a tight litte package, but don't overhype it.
You know, I really don't mind marketing, but I can't stand lying.