August 29, 2014

Intel Woodcrest vs AMD Opteron for MySQL

AnandTech published Intel Woodcrest preview benchmarks which have some numbers for MySQL as well.

From these numbers performance looks great and it looks like finally Intel has something to respond to AMD Opterons on Server market. Now competition heats up and we’ll see what AMD will have to respond. As Opteron did not have serious updates for a while I guess there is something on a way.

Same site publishes Core vs K8 architecture comparison which is very interesting reading.

This article also looks at performance of Irwindale based Xeon and Sun T2000. Results are close to the ones I had in MySQL Performance Landscape presentation – neither of them was match to Opteron performance in MySQL workloads. As benchmark used in this article is very different from what I used I guess it is rather general situation.

Few words about P4 architecture. It is very interesting to see in most benchmarks Woodcrest at 3 Ghz beats Irwindale at 3.6 Ghz up to 3 times , which means performance per clock cycle is back where it should be. With P4 architecture it looks like increasing frequency was the main goal. Possibly marketing thought in peoples mind frequency is what defines performance. No one measures their CPU with SpecInt results or something like that. This strategy however could not work well with cores and hypperthreading as you need people to understand you want to pay more for multiple cores and threads, not only for Ghz. Also AMD with their using “Comparison Rank” rather than true frequency tricked people. Also with Server market this did not work as well as these are bought by more savvy people who had always better understanding of Frequency != Performance.

My expectation for performance per clock cycle was – it should be increasing with each new CPU family, This was true for 286,386,486,Pentium, Pentinum Pro/Pentium II but it failed for Pentium 4. Performance per clock cycle could be 1.5-1.7 times worse than PIII. Ie I remember being very surprised to find P4-2.0 Ghz runs just 30% faster than my PIII-1.0Ghz system. From preview Woodcrest seems to match my expectation for new processor family.

This article also has very interesting observation on MySQL Multiple CPU scalability . As you can see MySQL could be slower on Linux with many connections with increasing number of CPUs. This is known bug which being worked on. However it looks like it is not the case with Solaris. The scalability is not perfect but at least there is gain not loss from increasing number of CPUs. This makes me think this is not only MySQL but also Linux what has the problem. I had similar results in my tests and now happy to see these can be confirmed.

A bit of disclosure. As you could read in the article I was working with article authors advising them how to tune MySQL and what is the reason for negative scalability etc.

About Peter Zaitsev

Peter managed the High Performance Group within MySQL until 2006, when he founded Percona. Peter has a Master's Degree in Computer Science and is an expert in database kernels, computer hardware, and application scaling.

Comments

  1. guest says:

    Interesting to see that competition has returned to the server-cpu-world.
    Intel back on track and the ball is in AMD’s court now…

  2. sharikou says:

    AnandTech’s tests have no credibility.

    See the following analysis

    http://sharikou.blogspot.com/2006/06/busting-anandtechs-woodcrest-vs.html

  3. Ajay says:

    Correct me if I am wrong. Follow the link below to Anandtech’s own earlier benchmarks. Goto the last table on the page and check results of “Dual Dual Core 875″ and “Dual Opteron 248″ from 5 Concurrencies onwards. The increase is slight, but there is definetly no performance degradation.

    http://www.anandtech.com/IT/showdoc.aspx?i=2447&p=5

    The earlier review too uses Opterons+Linux+MySQL+InnoDB, the same as this setup used, though versions may differ. Why do you get totally different results sets this time around when moving to dual dual core opteron setup?

  4. peter says:

    Hi,

    Yes the link for DB2 and MySQL benchmark I guess is using different benchmark kit. That one was flawed in respect of MySQL results and I contacted author to explain what was not done well.

    The general big problem I have with anadtech results is using publicly unavailable benchmarks for MySQL. So unfortunately noone can validate these numbers.

    This particular benchmark is hopefully better. As I mentioned in the article on other data points (old Xeons and Opterons) it corelates very well with our internal results.

    Speaking about Innodb degradation – this link just tests it on 1 and 2 CPU this is not when problems typically happen. For most workloads you need more than 4CPUs to see degradation.

    Thank you for insightfool comments.

  5. NooB says:

    sharikou is telling us we should be running bzip compression program instead of apache, iis, sql, 3d rendering apps, etc…

    LOL!!

  6. sharikou says:

    Weoodcrest was soundly beaten in Apache benchmark, see

    http://www.gamepc.com/labs/view_content.asp?id=woodcrest&page=9

    A 2.33GHZ Woodcrest 5140 is slower than a 2GHZ low end Opteron 270.

    Unlike those review sites which live on ad money, GamePC is a system vendor that sells servers, workstations, desktops and notebooks, both AMD and Intel. In fact. you can pre-order Woodcrest CPUs from GamePC.com. GamePC’s customers include big names such as Dolby, Hitachi, Lockheed Martin, FAA.

  7. peter says:

    Sharikou,

    Unfortunately these results do not tell much – no one really uses Apache in production on Windows, plus they did not really specify what did they test – was it static html page ? What is really interesting for modern application is dynamic pages.

    Not to mention the benchmark description itself looks strange 50.000 users with concurrency of 10. What does this suppose to mean ? The concurrency is number of requests submitted at the same time by apache benchmark, so what is number of users when ?
    I suspect it is really number of requests not number of users.

    Anyway I will refrain from speculations at this point. Hopefully I will have a chance to benchmark Woodchrest vs Opteron myself soon.

  8. devon says:

    Amandtech doesn’t do a fair comparison, they compare a Woodcrest 3ghz (5160) to a AMD 275…if anything they should compare a single Woodcrest 3ghz to a AMD 285…or pick a lower end Xeon to compare with a 275…

    of course the 1st in line Xeon is going to beat the 3rd in line for AMD.

  9. peter says:

    Devon,

    To be honest I normally do not care about performance of first in line CPU – it costs much more while offering only marginal improvement compared to second and 3rd model.

    Speaking about results… well they say what they compare and they do not claim 3ghz Woodcrest will be faster than Opteron 285.

    Keep in mind results were published a while ago and it took a while to prepare so they simply might not have 285 available.

    Also what is the most important is to see the giant different from old Xeons. I personally do not care if Opteron or Xeon is faster 5-10 percent. This is something which can be offset by having better deal from vendors. While 40%+ performance difference might be showstopper, not to mention giant power consumption.

Speak Your Mind

*