G'Day Victor,
On Thu, Jun 26, 2008 at 11:56:26AM -0700, victor wrote:
> Thank you for the suggestion here are the results. I hate to scale it down to
> a 34 meg file, 128 megs took too long to write out. Compression is off.
>
> [EMAIL PROTECTED]:~# ptime dd if=/dev/urandom of=foo bs=128k count=256
> 256+0 records in
> 256+0 records out
> 33554432 bytes (34 MB) copied, 29.7683 s, 1.1 MB/s
>
> real 29.770
> user 0.000
> sys 1.335
Ok, that is indeed slow!
> [EMAIL PROTECTED]:~# iostat -xne 5
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 4.8 4.0 250.5 102.7 0.3 0.1 39.6 8.5 4 3 0 0 0 0
> c5t0d0
> 0.6 6.8 37.3 280.7 8.9 2.4 1217.4 327.3 51 24 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 0.8 14.6 0.3 1020.9 3.1 0.2 204.1 11.8 18 18 0 0 0 0
> c5t0d0
> 0.8 12.4 3.4 1002.3 8.8 0.4 666.2 31.7 80 42 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 0.0 20.8 0.0 1139.2 2.1 0.1 99.2 6.4 13 13 0 0 0 0
> c5t0d0
> 0.8 18.8 3.4 664.0 4.8 0.2 242.8 12.3 53 24 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 1.6 11.2 0.6 50.0 0.3 0.1 20.2 3.9 5 5 0 0 0 0
> c5t0d0
> 1.4 15.8 5.1 543.8 5.4 0.3 313.5 15.8 57 27 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 0.0 21.0 0.0 1104.7 5.3 0.3 253.0 13.6 31 28 0 0 0 0
> c5t0d0
> 0.4 21.4 3.3 1104.7 7.6 0.4 350.2 16.2 87 35 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 1.6 26.0 0.6 1102.3 0.0 3.1 0.0 111.7 0 12 0 0 0 0
> c5t0d0
> 1.6 25.6 6.8 1102.3 0.0 4.3 0.0 157.3 0 19 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
> extended device statistics ---- errors ---
> r/s w/s kr/s kw/s wait actv wsvc_t asvc_t %w %b s/w h/w trn tot
> device
> 1.0 22.8 25.9 1102.3 0.1 4.0 2.2 169.1 5 17 0 0 0 0
> c5t0d0
> 0.2 22.2 12.8 1102.3 0.0 7.8 0.0 346.7 0 45 0 0 0 0
> c5t1d0
> 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 10 0 10
> c4t1d0
Hmm - if the first output is the summary since boot, and that shows 6.8 w/s
for c5t1d0 at asvc_t of 327.3 ms. SATA 7200 RPM drives should be faster
than this, although it's a bit hard to know for sure with only 6.8 w/s.
The asvc_t is measuring time for the disk subsystem to return - meaning
this is looking like a disk or disk driver issue, not ZFS. To check further,
I'd test that disk with something simple:
window1# dd if=/dev/rdsk/c5t1d0 of=/dev/null bs=48k
window2# iostat -xne 5
which performs a raw read test with about the same I/O size as your writes.
Performing a raw write test would be more interesting, but you'd need to
destroy the pool first - since it would overwrite the disk.
I'd also check /var/adm/messages to see if something else odd was going on.
Brendan
--
Brendan
[CA, USA]
_______________________________________________
perf-discuss mailing list
[email protected]
