On May 10, 2010, at 12:16 AM, Roy Sigurd Karlsbakk wrote: > ----- "Michael Schuster" <michael.schus...@oracle.com> skrev: > >> On 10.05.10 08:57, Roy Sigurd Karlsbakk wrote: >>> Hi all >>> >>> It seems that if using zfs, the usual tools like vmstat, sar, top >> etc are quite worthless, since zfs i/o load is not reported as iowait >> etc. Are there any plans to rewrite the old performance monitoring >> tools or the zfs parts to allow for standard monitoring tools? If not, >> what other tools exist that can do the same? >> >> "zpool iostat" for one.
The traditional tools are quite useful. But you have to know how to use them properly. The tools I use most often are: iostat, fsstat, nfsstat, iosnoop, and nicstat. > > I know that, and iostat, etc, but wouldn't it be rather consistent to > integrate with the tools that have been used the latest two or three decades? > wio shouldn't be reported as 0% when the disks are the bottleneck... Absolutely not. Wait for I/O is a processor state and has no direct relation to I/O bottlenecks. As a result, it caused confusion for the better part of the past 30 years. In Solaris 10, wio is always zero. Alan talks about this and refers to an Infodoc describing how wio is useless. http://blogs.sun.com/tpenta/entry/how_solaris_calculates_user_system However, in the brave new world, I can't find a reference to the infodoc. Perhaps someone with a SunSolve account can find it? Suffice to say, this still trips people up and you'll find many references to posts where people try to clarify this if you google a bit. -- richard -- ZFS storage and performance consulting at http://www.RichardElling.com _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
