problem while executing fio on multiple disks

problem while executing fio on multiple disks

[Shrirang Phansalkar]

Hi,

I was trying to execute fio with the following job file, but I have some issues with it.

[globals]
rw=randrw
size=10M
ioengine=sync
blocksize=4k
numjobs=16
direct=1
iodepth=32
thread

[thread1]
filename=/dev/sdb
stonewall

[thread2]
filename=/dev/sdc
stonewall

[thread3]
filename=/dev/sdd
stonewall


After its completion in Disk stats I got util as :

sdb: util=99.64%
sdc: util=51.72%
sdd: util=34.76%

where as expected util for all 3 disks should be somewhat similar ( all 3 disks are of same size)

But if I run fio using separate jobfile for every disk, then I am getting almost similar disk util for each disk (around util=98% for each disk).

Pl let me know that if there is some problem with the fio or there is something wrong with my jobfile.

Thanks and regards,
Shrirang.

Re: problem while executing fio on multiple disks

[Jens Axboe]

On 2011-09-08 11:55, Shrirang Phansalkar wrote:
> Hi,
> 
> I was trying to execute fio with the following job file, but I have some issues with it.
> 
> [globals]
> rw=randrw
> size=10M
> ioengine=sync
> blocksize=4k
> numjobs=16
> direct=1
> iodepth=32
> thread
> 
> [thread1]
> filename=/dev/sdb
> stonewall
> 
> [thread2]
> filename=/dev/sdc
> stonewall
> 
> [thread3]
> filename=/dev/sdd
> stonewall
> 
> 
> After its completion in Disk stats I got util as :
> 
> sdb: util=99.64%
> sdc: util=51.72%
> sdd: util=34.76%
> 
> where as expected util for all 3 disks should be somewhat similar ( all 3 disks are of same size)
> 
> But if I run fio using separate jobfile for every disk, then I am getting almost similar disk util for each disk (around util=98% for each disk).
> 
> Pl let me know that if there is some problem with the fio or there is something wrong with my jobfile.

Hmm, I'm assuming the performance is about the same (whether run with
this file, or separately)?

Looks like the utilization is done prior to the job being started, hence
you get ~100%, ~50%, and ~33%. So looks like just an accounting bug,
I'll take a look at it.

-- 
Jens Axboe

Re: problem while executing fio on multiple disks

[Jens Axboe]

On 2011-09-08 13:32, Jens Axboe wrote:
> Hmm, I'm assuming the performance is about the same (whether run with
> this file, or separately)?
> 
> Looks like the utilization is done prior to the job being started, hence
> you get ~100%, ~50%, and ~33%. So looks like just an accounting bug,
> I'll take a look at it.

Does this work? Completely untested...

diff --git a/fio.c b/fio.c
index 4514840..b492889 100644
--- a/fio.c
+++ b/fio.c
@@ -1502,10 +1502,8 @@ static void run_threads(void)
 	for_each_td(td, i) {
 		print_status_init(td->thread_number - 1);
 
-		if (!td->o.create_serialize) {
-			init_disk_util(td);
+		if (!td->o.create_serialize)
 			continue;
-		}
 
 		/*
 		 * do file setup here so it happens sequentially,
@@ -1533,8 +1531,6 @@ static void run_threads(void)
 					td_io_close_file(td, f);
 			}
 		}
-
-		init_disk_util(td);
 	}
 
 	set_genesis_time();
@@ -1573,6 +1569,8 @@ static void run_threads(void)
 				break;
 			}
 
+			init_disk_util(td);
+
 			/*
 			 * Set state to created. Thread will transition
 			 * to TD_INITIALIZED when it's done setting up.

-- 
Jens Axboe