IHAC with an X4500 (no SSDs) that got burned with huge performance hit when 
applying patch 141445-09 (ZVOLs did not correctly implement the SCSI write 
cache behavior and given an unfortunate sequence of events write operations to 
the iSCSI lun could be lost.)
As a last resort, I suggested that he try snv_134a. Initial testing went very 
well, and performance was even better than when they started.  However, he now 
suspects that OpenSolaris has the same bug as Solaris (CR6770534). In his words:
+++++
Looking at the iostat writing to a comstar volume I only see the flushes. 
(missing data means there was no disk I/O for that time period)

                               r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t 
 %w  %b device
Tue May 11 10:47:56 EDT 2010
Tue May 11 10:48:01 EDT 2010
Tue May 11 10:48:06 EDT 2010
Tue May 11 10:48:11 EDT 2010
Tue May 11 10:48:16 EDT 2010
Tue May 11 10:48:21 EDT 2010    0.0  127.6    0.0  609.5  0.0  0.2    0.0    
1.3   0   3 c9t0d0 - Flush
Tue May 11 10:48:26 EDT 2010
Tue May 11 10:48:31 EDT 2010
Tue May 11 10:48:36 EDT 2010
Tue May 11 10:48:41 EDT 2010
Tue May 11 10:48:46 EDT 2010
Tue May 11 10:48:51 EDT 2010    0.0  167.8    0.0  785.0  0.0  0.2    0.0    
1.3   0   4 c9t0d0 - Flush
Tue May 11 10:48:56 EDT 2010
Tue May 11 10:49:01 EDT 2010
Tue May 11 10:49:06 EDT 2010
Tue May 11 10:49:11 EDT 2010    2.0    0.4   32.4    0.0  0.0  0.0    0.0    
0.0   0   0 c9t0d0 - Read Activity
Tue May 11 10:49:16 EDT 2010
Tue May 11 10:49:21 EDT 2010    0.0  156.2    0.0  736.0  0.0  0.2    0.0    
1.3   0   3 c9t0d0 - Flush
Tue May 11 10:49:26 EDT 2010
Tue May 11 10:49:31 EDT 2010
Tue May 11 10:49:36 EDT 2010
Tue May 11 10:49:41 EDT 2010
Tue May 11 10:49:46 EDT 2010
Tue May 11 10:49:51 EDT 2010    0.0  166.8    0.0  813.9  0.0  0.2    0.0    
1.4   0   4 c9t0d0 - Flush
Tue May 11 10:49:56 EDT 2010
Tue May 11 10:50:01 EDT 2010
Tue May 11 10:50:06 EDT 2010
Tue May 11 10:50:11 EDT 2010
Tue May 11 10:50:16 EDT 2010
Tue May 11 10:50:21 EDT 2010    0.0  174.4    0.0  802.2  0.0  0.2    0.0    
1.3   0   4 c9t0d0 - Flush
Tue May 11 10:50:26 EDT 2010
Tue May 11 10:50:31 EDT 2010    2.0    0.4   32.4    0.0  0.0  0.0    0.0    
0.0   0   0 c9t0d0 - Read activity
Tue May 11 10:50:36 EDT 2010
Tue May 11 10:50:41 EDT 2010
Tue May 11 10:50:46 EDT 2010
Tue May 11 10:50:51 EDT 2010    0.0  148.6    0.0  719.2  0.0  0.2    0.0    
1.2   0   3 c9t0d0 - Flush

So, what am I missing? The ZIL is enabled - I checked:

# mdb -kw
> zil_disable/X
zil_disable:
zil_disable:    0
I disabled it and didn't get any better performance and see the same io pattern.

>From what I can see the COMSTAR iSCSI target suffers from the same possible 
>loss of data bug the pre-patched iscsi target daemon suffers from.

Is there any way you can verify I have everything right. Do I have to configure 
COMSTAR to use the ZIL?
-- 
This message posted from opensolaris.org
_______________________________________________
storage-discuss mailing list
storage-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to