This discussion is archived
8 Replies Latest reply: Dec 11, 2012 5:47 AM by rainiermorilla RSS

How to evaluate the impact of removing CPU?

rainiermorilla Newbie
Currently Being Moderated
One of my clients need to remove three(of four) CPU to comply the licensing agreement with Oracle.

To avoid problems and also to list the possible problems that removing the CPU can bring, I wish to make a survey of the possible impacts, especially in performance, that removal can cause.

How can I get this information?
  • 1. Re: How to evaluate the impact of removing CPU?
    sb92075 Guru
    Currently Being Moderated
    rainiermorilla wrote:
    One of my clients need to remove three(of four) CPU to comply the licensing agreement with Oracle.

    To avoid problems and also to list the possible problems that removing the CPU can bring, I wish to make a survey of the possible impacts, especially in performance, that removal can cause.

    How can I get this information?
    application will continue to run; but might be slower
  • 2. Re: How to evaluate the impact of removing CPU?
    Justin Cave Oracle ACE
    Currently Being Moderated
    It's hard enough to make predictions when you're contemplating a marginal change in the resources available to the server. Removing 75% of the server's CPU capacity is going to make predictions very difficult.

    I would probably start by looking at an AWR report (if you're licensed to use the AWR) or a Statspack report (if you've installed Statspack) to see how much CPU you use during peak processing. Obviously, if you're using more than 25% of the available CPU during peak load, cutting back to 1 CPU is going to significantly throttle performance. Even if you're not using that much CPU over the snapshot window, though, you may be using more than 25% of the available CPU for shorter periods during that window-- massively restricting CPU will obviously throttle that performance. Plus, if you're going from using, say, 20% of the available CPU on average to 80% of the available CPU on average by removing 3 processors, you're likely to start encounter some significant queueing effects long before you hit 100% utilization.

    Justin
  • 3. Re: How to evaluate the impact of removing CPU?
    rp0428 Guru
    Currently Being Moderated
    >
    One of my clients need to remove three(of four) CPU to comply the licensing agreement with Oracle.

    To avoid problems and also to list the possible problems that removing the CPU can bring, I wish to make a survey of the possible impacts, especially in performance, that removal can cause.

    How can I get this information?
    >
    Make a survey? Why don't you just make that change on your client's test system and see what the results are? That's part of what testing is all about: to identify potential problem areas BEFORE you roll changes to production.

    And if your client doesn't have a test environment then your very first, and immediate, recommendation should be to invest in one. You won't even need to conduct a survey to provide evidence that a test system is 'best practices'.

    Nothing that any 4-to-1 cpu survey would show, if there even were such a thing, will necessarily apply to your client's particular usage patterns.

    Your client doesn't need a survey; they need proper testing.
  • 4. Re: How to evaluate the impact of removing CPU?
    jgarry Guru
    Currently Being Moderated
    Could you be a little more clear about the exact requirements? I seem to recall some language about the hardware having the capability to plug in more cpu being the driving factor, so just pulling them out won't help the licensing side.

    Never take any licensing advice from the Internet, by the way.
  • 5. Re: How to evaluate the impact of removing CPU?
    rainiermorilla Newbie
    Currently Being Moderated
    Excuse me, I am Brazilian and I am still learning English. I must have made ​​a mistake in translation. It is not a survey, but rather: "identify potential problem areas BEFORE you roll changes to production." - Is precisely that!

    I really want to do tests and get hard data on what is the real impact. The customer uses about 20% of its processing (cpu + cpu waits) but I know that does not mean he can just cut their processing to 25% safely.

    Yes, my client has a testing environment, but does not have a replay database, then it is not possible to perform tests on the workload precisely.

    How can I test without this tool, or at least collect data that will give me more precise information than AWR? It would be appropriate to open an SR on MOS?
  • 6. Re: How to evaluate the impact of removing CPU?
    Justin Cave Oracle ACE
    Currently Being Moderated
    Do you have any load testing tool that would allow you to simulate roughly the production load in the test environment?

    If you are saying that they're actually using 20% of the available CPU on average, I would expect that removing 3 of the 4 CPUs would have very substantial negative performance implications. That's going to take average CPU utilization to ~80% of 1 CPU which is likely to be beyond the knee in the performance curve.

    You could potentially work through Craig Shallahamer's Forecasting Oracle Performance but the margin of error on a change this big is going to be very substantial.

    Justin
  • 7. Re: How to evaluate the impact of removing CPU?
    rp0428 Guru
    Currently Being Moderated
    >
    I really want to do tests and get hard data on what is the real impact. The customer uses about 20% of its processing (cpu + cpu waits) but I know that does not mean he can just cut their processing to 25% safely.

    Yes, my client has a testing environment, but does not have a replay database, then it is not possible to perform tests on the workload precisely.

    How can I test without this tool, or at least collect data that will give me more precise information than AWR? It would be appropriate to open an SR on MOS?
    >
    There is nothing to open an SR for.

    The type of system change you are talking about can only be tested on a system that closely matches the production system. There are too many potential interactions between CPU, memory, cache and external (disk) storage systems to be able to use a non-matching test system and extrapolate the results.

    You will have to do equivalent performance testing to get meaningful results but those tests need to be done on equivalent systems.

    If I were you I would start by contacting your Oracle support person and identifying the actual costs with continuing to use your 4-cpu system and the costs associated with perhaps a 2-cpu system.

    Then I would estimate the costs of the work involved in any modifications to your test system to make it as near-equivalent to the production system as possible and add in the costs of actually performing the load testing and add a fudge factor to account for possible downtime or financial loss due to degradation of the system if your estimates are wrong.

    You may just find that it is less costly, and less risky, to just pony up for at least a 2-cpu license. That would at least buy your client enough comfort and time to design a more reliable long-term solution.
  • 8. Re: How to evaluate the impact of removing CPU?
    rainiermorilla Newbie
    Currently Being Moderated
    Through AWR and environment monitoring, I gave to my custumer an estimate, not very accurate, but realistically, that the environment would not support this abrupt removal of CPU.

    I requested him to get in touch with Oracle vendor to budget the CPU licenses because the cost of stopping everything can be much worse.

    Thank you Justin Cave, for the tips. I have always worked with the Database Replay, but in this case the customer had no way to license software. I will try to implement an opensource solution.

    Thanks to all

Legend

  • Correct Answers - 10 points
  • Helpful Answers - 5 points