[Lustre-discuss] Low performance of 64bit OSS

[Lu Wang]

 This is the network traffic graph of one OSS( the other one is similar, since files are stripped). There are some peaks, 
and I am sure user jobs which generated the peaks had not finished when the performance dropped down. 

------------------				 
Lu Wang
2008-12-05

-------------------------------------------------------------
发件人：Lu Wang
发送日期：2008-12-05 10:11:58
收件人：lustre-discuss
抄送：
主题：Re: [Lustre-discuss] Low performance of 64bit OSS

 1) upgrade from 1.6.5.1(32bit)
 2) yes,disk speed of OSSs

 by the way, our MDS are still on 1.6.5.1(32bit). 
 Is this a problem? 
"Note - Lustre clients running on architectures with different endianness are supported. One limitation is that the PAGE_SIZE kernel macro on the client must be as large as the PAGE_SIZE of the server. In particular, ia64 clients with large pages (up to 64kB pages) can run with i386 servers (4kB pages). If you are running i386 clients with ia64 servers, you must compile the ia64 kernel with a 4kB PAGE_SIZE (so the server page size is not larger than the client page size). " 

when I strace a "cp" from Lustre to local disk, it R/W 4k every time. 
.....
read(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 4096) = 4096
write(4, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 4096) = 4096
.....

when I strace a "cp" from Lustre to Lustre, it R/W 22M every time.
read(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 2097152) = 2097152
_llseek(4, 2097152, [2097152], SEEK_CUR) = 0
read(3, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 2097152) = 2097152
_llseek(4, 2097152, [4194304], SEEK_CUR) = 0





On Thu, 2008-12-04 at 18:36 +0800, Lu Wang wrote:
> Dear list, 
> 
>     After upgrade our OSSs to  lustre-1.6.6-2.6.9_67.0.22.EL_lustre.1.6.6smp.x86_64, the phenomenon of frequent crash disappears.

Good.  What version did you upgrade from?

> However, the two OSS provide low performance: 100MB/s read, and 200MB/s write.

What was your read/write performance before the upgrade?  Is this 100
and 200 MBs/ measurements measured at the clients or is it a measurement
of the disk speed at the OSS?

> Is it a problem caused by 32bit client and 64bit OSS server?

No.  Mixed 32/64 bit clusters should not cause a slow-down.

b.


_______________________________________________
Lustre-discuss mailing list
Lustre-discuss at lists.lustre.org
http://lists.lustre.org/mailman/listinfo/lustre-discuss



_______________________________________________
Lustre-discuss mailing list
Lustre-discuss at lists.lustre.org
http://lists.lustre.org/mailman/listinfo/lustre-discuss

-------------- next part --------------
A non-text attachment was scrubbed...
Name: network traffic.JPG
Type: image/jpeg
Size: 101123 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20081205/04d06b3a/attachment.jpeg>