Perfmon and IOPS - Storage

This is a discussion on Perfmon and IOPS - Storage ; Hi there - we are having some performance issues with a particular application (works with very large image files) and I've been trying to get some benchmarks on it while running in different configurations. As a baseline, I'm trying to ...

+ Reply to Thread
Results 1 to 9 of 9

Thread: Perfmon and IOPS

  1. Perfmon and IOPS

    Hi there - we are having some performance issues with a particular
    application (works with very large image files) and I've been trying
    to get some benchmarks on it while running in different
    configurations. As a baseline, I'm trying to run everything off a
    single local SATA drive (typical 7200RPM), while running perfmon to
    capture various statistics such as CPU, Physical Disk Reads/Sec,
    Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    Sec.

    Now usually, reads/sec hover between 70-100 range, which makes sense
    to me if these are representative of single-disk IOPS. However, when
    I first start off the app, reads/sec shoots up to well over 1000 and
    stays there for several minutes. Isn't this impossible? Or does
    reads/sec on perfmon not match up well to IOPS?

    I always assumed IOPS would equal reads/sec + writes/sec.

  2. Re: Perfmon and IOPS

    You have to take into account cache and queuing.



    "maethlin" wrote in message
    news:dc0bdd03-5359-434c-a73b-941fbd8f500b@l28g2000prd.googlegroups.com...
    > Hi there - we are having some performance issues with a particular
    > application (works with very large image files) and I've been trying
    > to get some benchmarks on it while running in different
    > configurations. As a baseline, I'm trying to run everything off a
    > single local SATA drive (typical 7200RPM), while running perfmon to
    > capture various statistics such as CPU, Physical Disk Reads/Sec,
    > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    > Sec.
    >
    > Now usually, reads/sec hover between 70-100 range, which makes sense
    > to me if these are representative of single-disk IOPS. However, when
    > I first start off the app, reads/sec shoots up to well over 1000 and
    > stays there for several minutes. Isn't this impossible? Or does
    > reads/sec on perfmon not match up well to IOPS?
    >
    > I always assumed IOPS would equal reads/sec + writes/sec.




  3. Re: Perfmon and IOPS


    "maethlin" wrote in message
    news:dc0bdd03-5359-434c-a73b-941fbd8f500b@l28g2000prd.googlegroups.com...
    > Hi there - we are having some performance issues with a particular
    > application (works with very large image files) and I've been trying
    > to get some benchmarks on it while running in different
    > configurations. As a baseline, I'm trying to run everything off a
    > single local SATA drive (typical 7200RPM), while running perfmon to
    > capture various statistics such as CPU, Physical Disk Reads/Sec,
    > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    > Sec.
    >
    > Now usually, reads/sec hover between 70-100 range, which makes sense
    > to me if these are representative of single-disk IOPS. However, when
    > I first start off the app, reads/sec shoots up to well over 1000 and
    > stays there for several minutes. Isn't this impossible? Or does
    > reads/sec on perfmon not match up well to IOPS?
    >
    > I always assumed IOPS would equal reads/sec + writes/sec.


    Before focusing on the application, I would recommend creating a performance
    baseline of the system using an I/O generating tool.

    Moojit



  4. Re: Perfmon and IOPS

    Could read-cache and queuing really account for a 10-fold increase in
    IOPS on a single-disk system? I haven't done a lot of extensive
    performance testing on disk I/O in the past so this is somewhat new to
    me.


    On Jun 16, 1:03 pm, "John Fullbright"
    wrote:
    > You have to take into account cache and queuing.
    >
    > "maethlin" wrote in message
    >
    > news:dc0bdd03-5359-434c-a73b-941fbd8f500b@l28g2000prd.googlegroups.com...
    >
    > > Hi there - we are having some performance issues with a particular
    > > application (works with very large image files) and I've been trying
    > > to get some benchmarks on it while running in different
    > > configurations. As a baseline, I'm trying to run everything off a
    > > single local SATA drive (typical 7200RPM), while running perfmon to
    > > capture various statistics such as CPU, Physical Disk Reads/Sec,
    > > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    > > Sec.

    >
    > > Now usually, reads/sec hover between 70-100 range, which makes sense
    > > to me if these are representative of single-disk IOPS. However, when
    > > I first start off the app, reads/sec shoots up to well over 1000 and
    > > stays there for several minutes. Isn't this impossible? Or does
    > > reads/sec on perfmon not match up well to IOPS?

    >
    > > I always assumed IOPS would equal reads/sec + writes/sec.



  5. Re: Perfmon and IOPS

    Moojit - I just ran IOMeter on this machine, using 32k transfer size,
    100% read, 100% sequential, I get a reading of over 2k IOPS - does
    that sound correct? If I switch it to 100% random the IOPS reading on
    IOMeter drops down to 160ish.


    On Jun 16, 2:48 pm, "moojit" wrote:
    > "maethlin" wrote in message
    >
    > news:dc0bdd03-5359-434c-a73b-941fbd8f500b@l28g2000prd.googlegroups.com...
    >
    >
    >
    > > Hi there - we are having some performance issues with a particular
    > > application (works with very large image files) and I've been trying
    > > to get some benchmarks on it while running in different
    > > configurations. As a baseline, I'm trying to run everything off a
    > > single local SATA drive (typical 7200RPM), while running perfmon to
    > > capture various statistics such as CPU, Physical Disk Reads/Sec,
    > > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    > > Sec.

    >
    > > Now usually, reads/sec hover between 70-100 range, which makes sense
    > > to me if these are representative of single-disk IOPS. However, when
    > > I first start off the app, reads/sec shoots up to well over 1000 and
    > > stays there for several minutes. Isn't this impossible? Or does
    > > reads/sec on perfmon not match up well to IOPS?

    >
    > > I always assumed IOPS would equal reads/sec + writes/sec.

    >
    > Before focusing on the application, I would recommend creating a performance
    > baseline of the system using an I/O generating tool.
    >
    > Moojit



  6. Re: Perfmon and IOPS


    "maethlin" wrote in message
    news:1b1a137f-13f5-42c3-99fc-8aeebf86df62@w34g2000prm.googlegroups.com...
    > Moojit - I just ran IOMeter on this machine, using 32k transfer size,
    > 100% read, 100% sequential, I get a reading of over 2k IOPS - does
    > that sound correct? If I switch it to 100% random the IOPS reading on
    > IOMeter drops down to 160ish.


    IOPS calculations are typically performed using 512B transfers, not 32K
    transfers. If you're system under test is connected to the internet, I can
    provide you with a demo license for datamover. It will perform an automated
    performance test across many different transfer sizes and provide a report
    that can be saved to a CSV file for later import into excel. Shameless plug
    for my product, but setting up IoMeter can be difficult. Sequential I/O is
    a must, RANDOM will always slow you down.

    www.moojit.com

    Da Moojit
    >
    >
    > On Jun 16, 2:48 pm, "moojit" wrote:
    >> "maethlin" wrote in message
    >>
    >> news:dc0bdd03-5359-434c-a73b-941fbd8f500b@l28g2000prd.googlegroups.com...
    >>
    >>
    >>
    >> > Hi there - we are having some performance issues with a particular
    >> > application (works with very large image files) and I've been trying
    >> > to get some benchmarks on it while running in different
    >> > configurations. As a baseline, I'm trying to run everything off a
    >> > single local SATA drive (typical 7200RPM), while running perfmon to
    >> > capture various statistics such as CPU, Physical Disk Reads/Sec,
    >> > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    >> > Sec.

    >>
    >> > Now usually, reads/sec hover between 70-100 range, which makes sense
    >> > to me if these are representative of single-disk IOPS. However, when
    >> > I first start off the app, reads/sec shoots up to well over 1000 and
    >> > stays there for several minutes. Isn't this impossible? Or does
    >> > reads/sec on perfmon not match up well to IOPS?

    >>
    >> > I always assumed IOPS would equal reads/sec + writes/sec.

    >>
    >> Before focusing on the application, I would recommend creating a
    >> performance
    >> baseline of the system using an I/O generating tool.
    >>
    >> Moojit

    >




  7. Re: Perfmon and IOPS

    If you are not using an application which has the ability to perform
    DIRECT_IO (also known as DIO), then you must ensure that your filesize
    you are reading/writing is at least 2x the size of your system memory.
    (2x is not really needed, but that's the simplest method).

    Then you run a long term test. While running perfmon, simply remove the
    beginning and end of the test and now you have a fairly accurate
    representation of your throughput.

    As someone stated before, for IOPS, use 512B (2k is also common) and for
    throughput, you usually want to use 32k or 64k depending on the cache
    alignment and flushing algorithm used by your storage provider.

    The best thing ... experiment experiment experiment. You will become a
    performance guru by the time you are done and figure out all the nuances
    of performance testing.

    ~kenny

    maethlin wrote:
    > Hi there - we are having some performance issues with a particular
    > application (works with very large image files) and I've been trying
    > to get some benchmarks on it while running in different
    > configurations. As a baseline, I'm trying to run everything off a
    > single local SATA drive (typical 7200RPM), while running perfmon to
    > capture various statistics such as CPU, Physical Disk Reads/Sec,
    > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    > Sec.
    >
    > Now usually, reads/sec hover between 70-100 range, which makes sense
    > to me if these are representative of single-disk IOPS. However, when
    > I first start off the app, reads/sec shoots up to well over 1000 and
    > stays there for several minutes. Isn't this impossible? Or does
    > reads/sec on perfmon not match up well to IOPS?
    >
    > I always assumed IOPS would equal reads/sec + writes/sec.


  8. Re: Perfmon and IOPS


    >
    > As someone stated before, for IOPS, use 512B (2k is also common) and for
    > throughput, you usually want to use 32k or 64k depending on the cache
    > alignment and flushing algorithm used by your storage provider.
    >


    I fully agree with Kenny here

    just to add: three is a difference between IOPS and throughput, and the way
    you take measurements are different, as Kenny has indicated above

    Rgds,
    edwin.



  9. Re: Perfmon and IOPS

    Yes. Also, if you are running this from a remote client you have client
    side redirector caching as well as the server side caching. It is not
    unusual to see a Windows Client (Xp/Vista) and Server effective throughput
    exceed the wireline speed.


    "maethlin" wrote in message
    news:cefd170b-00bf-4523-aef9-f8eb42425f0e@z24g2000prf.googlegroups.com...
    > Could read-cache and queuing really account for a 10-fold increase in
    > IOPS on a single-disk system? I haven't done a lot of extensive
    > performance testing on disk I/O in the past so this is somewhat new to
    > me.
    >
    >
    > On Jun 16, 1:03 pm, "John Fullbright"
    > wrote:
    >> You have to take into account cache and queuing.
    >>
    >> "maethlin" wrote in message
    >>
    >> news:dc0bdd03-5359-434c-a73b-941fbd8f500b@l28g2000prd.googlegroups.com...
    >>
    >> > Hi there - we are having some performance issues with a particular
    >> > application (works with very large image files) and I've been trying
    >> > to get some benchmarks on it while running in different
    >> > configurations. As a baseline, I'm trying to run everything off a
    >> > single local SATA drive (typical 7200RPM), while running perfmon to
    >> > capture various statistics such as CPU, Physical Disk Reads/Sec,
    >> > Physical Disk Writes/Sec, and Disk Read Bytes/Sec and Disk Write Bytes/
    >> > Sec.

    >>
    >> > Now usually, reads/sec hover between 70-100 range, which makes sense
    >> > to me if these are representative of single-disk IOPS. However, when
    >> > I first start off the app, reads/sec shoots up to well over 1000 and
    >> > stays there for several minutes. Isn't this impossible? Or does
    >> > reads/sec on perfmon not match up well to IOPS?

    >>
    >> > I always assumed IOPS would equal reads/sec + writes/sec.

    >



+ Reply to Thread