0

Looking for a way to profile hardware utilization. While it'd be nice to correlate the utilization to the cause of the hardware being use, at this point I'm more interested in as detailed as possible logging of utilization to the hardware.

Must work on Linux, would be nice if it worked on Windows 7; tie breaker if it works on OSX too.

I'm not stating what needs to be logged, because I'm not 100% sure what can be logged.

Goal is to profile hardware utilization to optimize the build of the hardware. For example, SSD do better (based on my understanding) when they're used in a high-read; among other reasons.

blunders
  • 813
  • 7
  • 14
  • 30
  • 3
    I don't know if you can get perfect numbers on this unless it's from the manufacturer since drivers play a role in performance, which are OS dependent. The best numbers you can get is to configure it for real-world use and measure it in real-world performance for your use cases. – Bart Silverstrim May 25 '11 at 13:05
  • @Bart Silverstrim: Hmm, right -- that's what I want to do, not trying to guessing how hardware will perform, want logs of real world data. Make sense, or no? – blunders May 25 '11 at 13:50
  • @Bart Silverstrim: Thought about your comment, are you saying that it's impossible to do. In thinking about it, makes sense, but strange that hardware utilization profiling could not be done directly. – blunders May 25 '11 at 14:23
  • You can get numbers on performance straight from hardware, but it won't necessarily work in real-world performance unless you are measuring the whole environment. For example, you can have a database that works really really well as long as you're accessing a small amount of data repeatedly, but once you're accessing clusters of data all over perhaps that configuration has trouble pulling things that aren't always in cache. – Bart Silverstrim May 25 '11 at 14:31
  • 1
    Likewise things like driver versions and driver quality can affect performance (you can get a video card with high pure hardware tested numbers, but the Windows driver gives one benchmark while a Linux driver gives another and the binary from the manufacturer gives another versus an open source version of the driver yet another) – Bart Silverstrim May 25 '11 at 14:32
  • So the only way to test is to actually build the server or system and put it in real-world use and benchmark it from there. – Bart Silverstrim May 25 '11 at 14:32
  • @Bart Silverstrim: Thanks... :-) ...part of the problem is I'm not 100% sure what I'm asking for. I know things like OS, Kernel, drivers, climate, etc. all play a role in performance, but I want RAW data from the hardware. How can I get get numbers on performance straight from hardware? – blunders May 25 '11 at 15:41
  • Only the manufacturer could give you that because you're asking for numbers that could only come from a lab, without OS drivers or components getting in the way. – Bart Silverstrim May 25 '11 at 17:24
  • Otherwise your numbers can be skewed just by changing a motherboard out or doing a firmware update. – Bart Silverstrim May 25 '11 at 17:25
  • What do you need the numbers for? Perhaps you're looking for the wrong question to answer. – Bart Silverstrim May 25 '11 at 17:25
  • @Bart Silverstrim: Okay, yeah, didn't think it was possible, but reading your comment thought it might be; who knows, there might be a standard "secret" slot on a motherboard that I've never heard of that would feedback to a USB slot that could be read by an application... :-) ...so, seem like there really is not an answer to my question. One thing that was wrong about my original question was that I included the OS; which would not be the same as raw access to what the hardware is doing. – blunders May 25 '11 at 18:33

1 Answers1

1

You can't benchmark in the way you are looking to do because it implies there is no difference in usage based on OS. The drivers and the way the OS uses a divers will vary greatly depending on OS. Linux drivers in and of themselves generally get less attention from manufacturers than windows drivers and many times this leads to less optimal situations. As an example windows uses a hybrid drive differently than a normal drive and different still than a solid state drive. Does Linux? If it does did the manufacturer present it properly? Lets presume it treats all drives the same (for the sake of arguement as I have no idea). in that case your disk benchmarks will show a vast difference between windows and linux, for no reason other than OS settings and drivers. Linux drivers in and of themnselves generally get less attention from manufacturers than windows drivers and many times this leads to less optimal situations

Jim B
  • 24,081
  • 4
  • 36
  • 60
  • +1 @Jim B: Okay; who knows, there might be a standard "secret" slot on a motherboard that I've never heard of that would feedback to a USB slot that could be read by an application... :-) ...so, seem like there really is not an answer to my question. One thing that was wrong about my original question was that I included the OS; which would not be the same as raw access to what the hardware is doing. – blunders May 25 '11 at 18:33
  • 1
    When you say "raw access" what does that mean? in the case of a USB port for instance are you looking for transmission performance specs? – Jim B May 26 '11 at 21:30
  • +1 @Jim B: My reference to the USB port "magical" in the sense it makes no sense. Basically what I was trying to say, though likely still not explaining it clearly is that I'd like to "see" what the hardware is doing with as few proxies in the middle as possible. The reference to a magical slot on the motherboard was just that, a slot that streams data on what the hardware is doing independently. Make sense, or is there something flawed conceptual with that request. – blunders May 27 '11 at 12:08