Managers, Larry Scott has asked me to point out
the following:
>Just a quick comment regarding the summary. It's important to note that
>the smoothsync capability is important for performance. Trading off the
>clicking for smoothsync might be okay on a lightly-loaded system, but I
>would not recommend this for any system with a serious load.
>
>Smoothsync can decrease a workload's net I/O cost by allowing each page to
>age for a specified time period. Allowing each page to age can correspond
>to an increased buffer cache hit rate, hence, fewer disk requests. The
>smsync2 mount option further extends this by not flushing pages until they
>are both dirty and idle.
>
>Smoothsync can minimize stalls resulting from a heavy system I/O load.
>I/O throttling, which requires smoothsync, enforces a limit on the number
>of delayed I/O requests allowed to be on a device queue at any point in
>time. This allows the system to be more responsive to any synchronous
>requests added to the device queue, such as a read or the loading of a new
>program into memory. This can also decrease the duration of process
>stalls for specific dirty pages, as pages remain available until placed on
>the device queue.
>
>I felt it important that system managers had this information before
>taking action to address the disk clicks.
Hopefully the situation where click-madness is the risk
(i.e. personal workstations) wont in general experience
heavy system I/O, but for big disk servers, managers should
be aware of the above.
Cheers,
Terry.
Received on Wed Apr 04 2001 - 15:32:58 NZST