Back in 2005 there was a short discussion about "Wait I/O Calculations".
Since the first contact with Solaris 10 I've been missing useful %wio values 
and things
did not become better over the last years. Hope, you don't mind raising this 
issue again.

The basic statement was "%wio is a meaningless statistic ... we've removed the 
statistic in Solaris 10. %wio will now always return zero."

IMHO the creators of this "meaningless value" did have something in mind (all 
SVR4 systems and most other UNIX-Systems _do_ maintain this "meaningless 
value"). 
I can agree with most arguments from the discussion in 2005. Above all a 
certain amount of %wio doesn't say anything about IO problems or its nature. If 
there are IO problems a process-/threadwise analysis (dtrace, prex ...) is more 
useful.

The %wio value is a percentage of time where "CPU is idle waiting for IO". In 
other words: It gives us an idea of how much more CPU time could be spent for 
calculations if the CPU(s) would not have to wait for IO with the programs they 
are currently executing (load profile). Nothing more but nothing less!!! As 
such this value gives us an excellent overall indicator of the CPU/IO-subsystem 
balance related to the current load profile.

To take one example from the discussion:
> A user would see 100% I/O wait in vmstat report when the CPU is idle 
> and there's only one thread waiting for I/O,
This means faster program execution is blocked by waiting on IO-requests. If 
this can be seen for a long time (and the program isn't changed) the most 
promising way to make things faster is to care for faster IO. Detailed analysis 
could be done with dtrace.
> he would also see 0% wait when there're 100 threads waiting for I/O 
> but just one thread keeps the CPU busy at the same time,
This means:
a) The user is running Solaris 10 and has no useful wio value
or:
b) the IO system is fast enough to serve the CPU. If it was faster we would
see little effect only (compared to the situation above) since the program 
would have to wait for CPU in case of IO returning faster.

I see all the advantages of detailed statistics e. g. with dtrace but I cannot 
see any advantage of removing the _useful_ %wio-value from kstat (sar -d, 
mpstat ...). So I'd like to ask for re-insertion of this value.
 
 
This message posted from opensolaris.org
_______________________________________________
perf-discuss mailing list
[email protected]

Reply via email to