Re: Interpreting disk-zone-profile

On Fri, Jul 20 2007, Erwan Velu wrote:
> Sorry to send so many mails on your list, hope you don't find me too
> verbose.

No it's great, you are finding a lot of corner case bugs! So just keep
going :-)

> I've been running the disk-zone-profile on my disk.
>
> After 5mn, it reaches 100% done and eta = 0. I'm now in this situation for
> 10mn, I can see the average rate falling down slowly.
>
> [root_at_max1 ~]# fio disk-zone-profile
> /dev/sdb: (g=0): rw=read, bs=64K-64K/64K-64K, ioengine=libaio, iodepth=2
> Starting 1 process
> Jobs: 1 (f=1): [R] [100.0% done] [ 70802/ 0 kb/s] [eta 00m:00s]
>
> I thought it could be a cache flushing effect but we are O_DIRECT so
> couldn't be an explanation....
> I see see are libaio so the 100% done could may significate that all ios
> have been queued but in that case eta should be set to 0 because as all IOs
> haven't been ended we should know how many are remaning and the average
> rate of reading.
>
> This jobs reads 256M every 2G, so on a 150G disk and reading at 70MB/sec,
> this should be very quick isn't it ? Around 5mn maybe...
>
> Do you have any idea how I must interpret this situation ?

Please attach the jobfile you use (that btw goes for any bug report).

-- 
Jens Axboe