Dual CPU Database Server Comparison
by Johan De Gelas on December 2, 2004 12:11 AM EST- Posted in
- IT Computing
Analyses: IBM DB2 8.1.3 32 bit and 64 bit
Let us summarize what we have learned so far: L3-caches and Hyperthreading offer very limited performance increases. Hyperthreading is essentially free of charge, so we are happy with whatever little bonus we get. However, one can seriously question the wisdom of paying $800 more for a dual Xeon 3.2 GHz with 2 MB L3 instead of one with 1 MB L3-cache. Of course, it is a small amount when compared to license and labour costs, but still...However, we must not be blind to the possible limitations of our benchmarks. The limited effect of the large caches made us think. Did we randomize a little too well? Often, some rows will be requested far more often than others. For example, if you have a forum, the most recent messages/threads will be accessed much more often than the oldest. If you have an e-commerce site, the items appearing on the front page or on major pages will be loaded much more often. While our real world queries do not access all rows equally, more research will be needed to see closer how our randomize function could mimick real world behavior.
Still, we don't expect spectacular improvements, as the benchmark scales well with CPU clock speed.
64 bit offers a very decent improvement (12% - 16%), although it might be less than what some reports speculated.
We certainly didn't expect Nocona to do so well. SQL databases access data in the memory sometimes randomly, courtesy of, for example, linked lists, and processing a single SQL database query requires parsing and checking the query, optimising it, and ordering results for output, which is very branch and integer intensive, similar to interpreting programming languages. The Nocona Xeon with its extremely long pipeline (and hence, high branch misprediction penalties), slightly higher latency caches (compared to Opteron and previous Xeons) and no L3-cache seemed like the worse architecture that one could think of for database serving.
The new Nocona Xeon surprised us in a positive way, and it outperformed the previous Xeons, even on a clock per clock basis. More in-depth research with CPU counters must give us the exact reasons why, but right now, we can only conclude that the faster memory bus (800 MHz versus 533 MHz) and the twice as large L1- and L2-caches outweigh the fact that it has no L3-cache and a longer pipeline than the previous Xeons. Maybe the improved branch predictor performs miracles, but we are not sure right now.
Facebook
Twitter
RSS
46 Comments
View All Comments
JohanAnandtech - Thursday, December 2, 2004 - link
About SLES9 and NUMA: NUMA is also supported by Linux kernel 2.4.21 and it boosts performance only a tiny bit. The reason are the very speedy HT links which keep latency at a minimum.It is still possible that kernel 2.6 NUMA support is far better of course, but I doubt it makes a difference for quad or dual systems as there is only hop in quad systems. With two hops (8 CPUs) from CPU 1 to 8 for example, this will become important.
AtaStrumf - Thursday, December 2, 2004 - link
A TYPO:So for now, the Opteron has an advantage still, but it ***can*** /can't/ knock out the Xeon, as it could have a few months ago, before the Xeon Nocona arrived.
HardwareD00d - Thursday, December 2, 2004 - link
There have been enough benchmarks on the web for a long time which show that Opteron generally wipes Xeon's a$$ hands down, and scales far better in multi processor configurations. The latest Xeon is nothing special compared to prior versions and will no doubt preform better mostly due to its increased clock speed. Xeon will never be better than Opteron no matter how much cache and tweaks Intel adds.Maybe Intel's next server architecture will be something to woo, but that's a ways off.
jshaped - Thursday, December 2, 2004 - link
as a long-time reader of aceshardware, i'll be the first to welcome Johan here, great first article. keep them coming!!!!HardwareD00d - Thursday, December 2, 2004 - link
I don't think there are enough variations of the way requests are handled to make a realistic conclusion for either chip. I'm sure you could create a situation where Intel bests AMD in My Sql, and vice versa. This article really needs more benchmarks and more in-depth analysis. Still, it provides enough information to conclude that both Xeon and Opteron have their strengths and weaknesses.mczak - Thursday, December 2, 2004 - link
Nice read. I really think though you should have used SLES 9. Not only does it use kernel 2.6, but it's also NUMA-aware (and DB2 should specifically support it, though it might not have been released yet). SLES 9 also ought to be faster especially on x86_64 due to newer compiler (not that it would matter much with precompiled databases, but every bit counts...). Though for 2-cpu boxes, NUMA might not be that important - but it's safe to predict a landslide victory for a 4-cpu opteron with NUMA support vs. a 4-cpu xeon box. Xeons simply don't scale to 4 cpus, intel might sell them but they are useless (especially since the Xeon MPs are still limited to 400 (or was that 533?) Mhz FSB.A pity though the quad opterons don't support ddr-400. I guess manufacturers decided it's more important to have a boatload of ram slots than fewer slots (with shorter traces) with higher speeds...
And btw, where are the 90nm Opterons? AMD's latest roadmap shows them as available in 2004, which doesn't leave too much time...
bthomas - Thursday, December 2, 2004 - link
Bogus conclusions about IBM tests IMO. From the
article:
> If we had published a similar report back in
> August, the Opteron would enjoyed a landslide
> victory. Luckily for Intel, Nocona is very
> competitive and is about 5% faster than the Opteron 250.
and later in the "conclusion":
> Nevertheless, AMD cannot sit on its laurels.
> Intel made a very good comeback with Nocona, as > this 3.6 GHz CPU is just a tiny bit faster in >
DB2.
It has not.
You fail to specify that this is comparing the _32 bit_ mode for the Opteron. IF you compare the Nocoma performance to the Opteron 64 bit capability...it sweeps the the Nocona in all tests.
The true conclusion is that based on the results in the article, for neither of the databases tested do *any* of the Intel processors compete with the Opteron.
fitten - Thursday, December 2, 2004 - link
Randomized benchmarks are hard to verify as well. You could get a "good" distribution that really takes advantage of cache locality while another randomization may be very cache unfriendly. I agree with #5 to a degree. A database that fits entirely inside of RAM isn't very interesting, ultimately.Still, I am happy that AnandTech is going down these paths of benchmarking instead of just being about Doom3, HL2, and FarCry like most other sites. I eagerly await further database benchmark articles.
PrinceXizor - Thursday, December 2, 2004 - link
#5 - Since when do top tier e-commerce cites compare to mid-level company database users as the beginning of the article mentioned?My company is an engineering firm that does custom electronics. Our database server handles all the transactions for our Inventory/MRP system which is mostly reads. These benchmarks are very appropriate. I wish I could have convinced my boss to go Opteron. Its funny, they had Athlon MP's before and then switched to Xeons when Opteron was out. Go figure.
Anyway, great article. I'm not IT guy by any stretch, but I enjoyed the article.
P-X
Jason Clark - Thursday, December 2, 2004 - link
#6, done ages ago..http://www.anandtech.com/IT/showdoc.aspx?i=2205
http://www.anandtech.com/IT/showdoc.aspx?i=1982