[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Xen disk write slowness in kernel 3.8.x

Do you know the size of your requests?
You used "iostat -m". Perhaps "iostat -xm" could be more meaningful as it will 
tell you the average request size in sectors.

I still didn't understand your environment. I think this is the first time you 
mention NFS/SMB.

Weren't you testing copies from /dev/zero to your device?

I'd also recommend you run "oprofile" to see if there are any time sinks that 
we are overlooking. You can find out how to use it here:


-----Original Message-----
From: Steven Haigh [mailto:netwiz@xxxxxxxxx] 
Sent: 03 April 2013 12:42
To: Felipe Franciosi
Cc: Xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-devel] Xen disk write slowness in kernel 3.8.x

On 03/04/13 22:36, Felipe Franciosi wrote:
> Based on what you said later, your raid stripe size is 128KB. When you plug 
> /dev/md3 to your guest, you will be going through blkback/blkfront which only 
> support 44KB per request.

This is interesting. The default size for a RAID5/6 via mdadm these days is 

> So can I ask what is the "task that is going on"?
> I mean, what is your workload and how does it look like (in terms of 
> reads/writes, (a)sync and size)?

The load is mostly sequential writes. Samba + NFS shares mainly. Files are 
usually 300Mb -> 10Gb. Its mostly writes, with occasional reads. 
Interestingly, I get 220Mb/sec+ reads - but only ~50Mb/sec writes via the DomU.

Steven Haigh

Email: netwiz@xxxxxxxxx
Web: https://www.crc.id.au
Phone: (03) 9001 6090 - 0412 935 897
Fax: (03) 8338 0299

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.