Topic
  • 5 replies
  • Latest Post - ‏2013-06-10T09:35:36Z by CG_Karthik
CG_Karthik
CG_Karthik
3 Posts

Pinned topic NMON & IOPS

‏2013-05-17T08:43:30Z |

Hi Guys,

All our AIX servers have nmon running on them daily and are captured on to a central reposiotry, We also use NMON2RRD to host it in web.

Now I have a question on how to determine the total IOPS of a server using the nmon files. My requirement is to identify the total IOPS of the system for a month and also to identify the IOPS for each VG.

Appreciate your help this.

Thanks

Karthik

  • nagger
    nagger
    1640 Posts
    ACCEPTED ANSWER

    Re: NMON & IOPS

    ‏2013-05-20T09:00:58Z  

    You question makes no sense you want: total I/O operations per second per month.

    So what is it per second or per month?

    if you sort the nmon files you can find the section that looks like this

    DISKXFER,Disk transfers per second gold6,hdisk1,hdisk2,hdisk0,hdisk3,cd0,cd1
    DISKXFER,T0001,0.0,0.0,0.0,0.0,0.0,0.0
    DISKXFER,T0043,98.9,23.1,56.8,0.0,0.0,0.0
    

    As you can see these are transfers per second and transfers = I/O ops.

    So you need to add up all the stats to get the total IOPS for that snapshot.

    To get the total operations you need to multiply by the elapsed seconds.

    For VG you need to include -V in the nmon command but I don't think (from memory) nmon2rrd hands the VG stats but you could add them yourself - assuming you are a C programmer. If not read a book :-)

     

    cheers, Nigel Griffiths

  • nagger
    nagger
    1640 Posts

    Re: NMON & IOPS

    ‏2013-05-20T09:00:58Z  

    You question makes no sense you want: total I/O operations per second per month.

    So what is it per second or per month?

    if you sort the nmon files you can find the section that looks like this

    DISKXFER,Disk transfers per second gold6,hdisk1,hdisk2,hdisk0,hdisk3,cd0,cd1
    DISKXFER,T0001,0.0,0.0,0.0,0.0,0.0,0.0
    DISKXFER,T0043,98.9,23.1,56.8,0.0,0.0,0.0
    

    As you can see these are transfers per second and transfers = I/O ops.

    So you need to add up all the stats to get the total IOPS for that snapshot.

    To get the total operations you need to multiply by the elapsed seconds.

    For VG you need to include -V in the nmon command but I don't think (from memory) nmon2rrd hands the VG stats but you could add them yourself - assuming you are a C programmer. If not read a book :-)

     

    cheers, Nigel Griffiths

  • CG_Karthik
    CG_Karthik
    3 Posts

    Re: NMON & IOPS

    ‏2013-05-20T12:45:03Z  
    • nagger
    • ‏2013-05-20T09:00:58Z

    You question makes no sense you want: total I/O operations per second per month.

    So what is it per second or per month?

    if you sort the nmon files you can find the section that looks like this

    <pre dir="ltr">DISKXFER,Disk transfers per second gold6,hdisk1,hdisk2,hdisk0,hdisk3,cd0,cd1 DISKXFER,T0001,0.0,0.0,0.0,0.0,0.0,0.0 DISKXFER,T0043,98.9,23.1,56.8,0.0,0.0,0.0 </pre>

    As you can see these are transfers per second and transfers = I/O ops.

    So you need to add up all the stats to get the total IOPS for that snapshot.

    To get the total operations you need to multiply by the elapsed seconds.

    For VG you need to include -V in the nmon command but I don't think (from memory) nmon2rrd hands the VG stats but you could add them yourself - assuming you are a C programmer. If not read a book :-)

     

    cheers, Nigel Griffiths

    Thanks Nigel.

    The value that I'm trying to fetch is the average IOPS for a month & Peak IOPS for a month. Sorry I didn't made my query clear earlier. 

    I already have scripts to sort and fetch the data from NMON file. Now I can addup this one to it to identify the IOPS.

    Since you stated "So you need to add up all the stats to get the total IOPS for that snapshot."

     I have the system with 800 LUNs. Now if I need to find the Average IOPS say for a day, Will the following procedure work ? 

    1. Identify the IOPS for each of the disks (from .nmon files)
    2. Sum them up to arrive at the final IOPS for that period in time( Say T0001,T0002)
    3. Identify the Average & Peak values among the Total IOPS calculated in step b

     Yep, Nigel I'm a C programmer ;) But all that was in my college. Now-a-days all I write is only on Bash :)

    Thanks

    Karthik

     

    Updated on 2013-05-20T14:08:11Z at 2013-05-20T14:08:11Z by CG_Karthik
  • nagger
    nagger
    1640 Posts

    Re: NMON & IOPS

    ‏2013-05-20T12:57:59Z  

    Thanks Nigel.

    The value that I'm trying to fetch is the average IOPS for a month & Peak IOPS for a month. Sorry I didn't made my query clear earlier. 

    I already have scripts to sort and fetch the data from NMON file. Now I can addup this one to it to identify the IOPS.

    Since you stated "So you need to add up all the stats to get the total IOPS for that snapshot."

     I have the system with 800 LUNs. Now if I need to find the Average IOPS say for a day, Will the following procedure work ? 

    1. Identify the IOPS for each of the disks (from .nmon files)
    2. Sum them up to arrive at the final IOPS for that period in time( Say T0001,T0002)
    3. Identify the Average & Peak values among the Total IOPS calculated in step b

     Yep, Nigel I'm a C programmer ;) But all that was in my college. Now-a-days all I write is only on Bash :)

    Thanks

    Karthik

     

    Yep - that is exactly the way to do it.

    There is an alternative of extracting the numbers from the rrd files using rrdtool - that has the benefits that is can do lots of the maths for you for any period you like and you not dealing with 31 nmon files for the month but you need to sort out the rrdtool command options to get that (from memory (slightly suspect) it can give you the average/peak of a saved stat = like one for each disk). And of course it could graph all the disks in a stacked graph for the period and you extract an average and peak by eye!

  • BruceSpencer
    BruceSpencer
    297 Posts

    Re: NMON & IOPS

    ‏2013-05-29T01:31:06Z  

    800 LUNs? Using IOADAPT would reduce the complexity, assuming you're not worried about the individual disks.

    IOADAPT,Disk Adapter aixtest,vscsi1_read-KB/s,vscsi1_write-KB/s,vscsi1_xfer-tps,vscsi0_read-KB/s,vscsi0_write-KB/s,vscsi0_xfer-tps
    IOADAPT,T0001,0.0,0.0,0.0,24.6,8.0,2.0
    IOADAPT,T0002,0.0,0.0,0.0,0.0,3.7,0.8
     

  • CG_Karthik
    CG_Karthik
    3 Posts

    Re: NMON & IOPS

    ‏2013-06-10T09:35:36Z  
    • nagger
    • ‏2013-05-20T12:57:59Z

    Yep - that is exactly the way to do it.

    There is an alternative of extracting the numbers from the rrd files using rrdtool - that has the benefits that is can do lots of the maths for you for any period you like and you not dealing with 31 nmon files for the month but you need to sort out the rrdtool command options to get that (from memory (slightly suspect) it can give you the average/peak of a saved stat = like one for each disk). And of course it could graph all the disks in a stacked graph for the period and you extract an average and peak by eye!

    Thanks Nigel,

    I may need to spend much time with the rrdtool for it. I will give a try.. Thanks :)

    Updated on 2013-06-10T10:26:56Z at 2013-06-10T10:26:56Z by CG_Karthik