AIO VP's -how many?

We are using AIX 4.3.3, and IDS 7.31UC3, with 2 processors but MULTIPROCESSOR=0, so we are using one CPU VP. I have 2 AIO VP's configured. We are not using kaio as it caused Informix crashes when we implemented a web server to access the Informix database. I have heard differing ideas about how to configure AIO VP's. I have looked at "onstat -g ioq" which shows me quite consistently an aio len of 0 and maxlen of 1. One book I have read, "Informix Performance Tuning" by Elizabeth Suto says to use this command to see if there are many requests in the aio queue. So my results seem to suggest that it probably isn't necessary to add more aio vp's. However, I have also heard that I should look at "onstat -g iov" and look at io/s and ours shows around 30 - this would suggest that we should have more than 2 AIO VP's. AND I have also read in other books that the number of AIO VP's should be equal to the same number of CPU VP's (one) AND I have also read in more books that AIO VP's should be equal to the number of disks that I/O is being performed on. That seems to suggest we should be using 6 of them. SO.....the truth is out there......Do you know how I find it or what it is? Cathy Geier Unix/Web/DB Admin Northwestern College St. Paul, MN 651-631-5284

These are all good rules of thumb some of which (like ==#CPUS and ==#disks) will give you starting values while others (like -g ioq and -g iov) give you monitoring techniques to tune the number of AIO VPs. The rules of thumb that I use are thus: Initial # AIO VPs: IDS Version # VIO VPS ------------ ---------- 7.1x 2.5 * # chunks (or use disks if discrete disks chunks if 7.20-7.23 2.0 * # chunks (chunks are part of disk arrays 7.24, 9.1x 1.5 * # chunks 7.3x, 9.2x 1.0 * # chunks + 2-4 for overhead Monitoring: onstat -g iov and look at io/wkup. If ALL AIO VPS have io/wkup >= 1.0then you probably can benefit from more AIO VPS. There should always be at least one AIO VP with io/wkup <1.0 indicating that often it was awakened and another AIO VP had already accepted the job so there was nothing to do. In addition monitoring the queue lengths of the chunks (-g ioq) and the VPS themselves (-g iov) will also clue you in to a need for more VPS. Art S. Kagel CKGeier@nwc.edu wrote: > > We are using AIX 4.3.3, and IDS 7.31UC3, with 2 processors but > MULTIPROCESSOR=0, so we are using one CPU VP. I have 2 AIO VP's > configured. We are not using kaio as it caused Informix crashes when we > implemented a web server to access the Informix database. > > I have heard differing ideas about how to configure AIO VP's. I have > looked at "onstat -g ioq" which shows me quite consistently an aio len of > 0 and maxlen of 1. One book I have read, "Informix Performance Tuning" by > Elizabeth Suto says to use this command to see if there are many requests > in the aio queue. So my results seem to suggest that it probably isn't > necessary to add more aio vp's. However, I have also heard that I should > look at "onstat -g iov" and look at io/s and ours shows around 30 - this > would suggest that we should have more than 2 AIO VP's. > > AND I have also read in other books that the number of AIO VP's should be > equal to the same number of CPU VP's (one) > > AND I have also read in more books that AIO VP's should be equal to the > number of disks that I/O is being performed on. That seems to suggest we > should be using 6 of them. > > SO.....the truth is out there......Do you know how I find it or what it > is? > > Cathy Geier > Unix/Web/DB Admin > Northwestern College > St. Paul, MN > 651-631-5284

FYI - We had a similar problem with your same configuration. Our kaio problems were solved with IDS 7.31UC5 CKGeier@nwc.edu wrote: > We are using AIX 4.3.3, and IDS 7.31UC3, with 2 processors but > MULTIPROCESSOR=0, so we are using one CPU VP. I have 2 AIO VP's > configured. We are not using kaio as it caused Informix crashes when we > implemented a web server to access the Informix database. > > I have heard differing ideas about how to configure AIO VP's. I have > looked at "onstat -g ioq" which shows me quite consistently an aio len of > 0 and maxlen of 1. One book I have read, "Informix Performance Tuning" by > Elizabeth Suto says to use this command to see if there are many requests > in the aio queue. So my results seem to suggest that it probably isn't > necessary to add more aio vp's. However, I have also heard that I should > look at "onstat -g iov" and look at io/s and ours shows around 30 - this > would suggest that we should have more than 2 AIO VP's. > > AND I have also read in other books that the number of AIO VP's should be > equal to the same number of CPU VP's (one) > > AND I have also read in more books that AIO VP's should be equal to the > number of disks that I/O is being performed on. That seems to suggest we > should be using 6 of them. > > SO.....the truth is out there......Do you know how I find it or what it > is? > > Cathy Geier > Unix/Web/DB Admin > Northwestern College > St. Paul, MN > 651-631-5284