[OpenAFS] Re: vos move speed rates

Eric Chris Garrison ecgarris@iu.edu
Wed, 13 Apr 2011 13:54:12 -0400


This is a multi-part message in MIME format.
--------------000700080805000204040203
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit

On 4/13/11 12:01 PM, openafs-info-request@openafs.org wrote:
> From: Andrew Deason <adeason@sinenomine.net>
> Date: Tue, 12 Apr 2011 15:59:22 -0500
> Organization: Sine Nomine Associates
> Subject: [OpenAFS] Re: vos move speed rates
> 
> On Tue, 12 Apr 2011 16:20:25 -0400
> Eric Chris Garrison <ecgarris@iupui.edu> wrote:
> 
>> > I've asked something similar before, but I could use some insight on
>> > how vos moves actually work.  I have a volume being vos move'd from
>> > one server to another right now, poking along in the range of about
>> > 15-20 Mbit/s (on a Gigabit ethernet connection, on the same switch as
>> > the other server).  Other times, I've seen performance more like I
>> > expect, something around 300 Mbit/s.
> How are you measuring these speeds? Are you looking at the actual link
> utilization, or just looking at how long it takes a volume to transfer,
> and how large the volume is?

I've measured two ways, using "time vos move" and dividing the size by
the time, and confirmed the rough result by watching the transfer rate
on "iftop".

I'm considering doing tests with other underlying filesystems than my
usual ext3.  There have been threads about this before, and I gather
ReiserFS would be great at handling small file IO, but it's out of
development.

Anyone have experience using XFS or other filesystems instead of ext*?

In other news, I was going to try to attempt the wrapper script on
volserver and restart with "bos restart server volserver -localauth" but
bizarrely, I got this in reply:

bos: failed to restart instance volserver (no such entity)

It's most definitely in my BosConfig, and the current volserver in the
process list has the currently running bosserver as its parent.  The
script is executable, and I checked my paths and it really should work.
 I'd try the other way, by manually killing the other volserver and
running the wrapped one by hand, but I'd be afraid it wouldn't respawn
when it crashes later.

I suppose I could try crashing the volserver with a file transfer and
see whether bosserver respawns on its own...

Chris
-- 
Eric Chris Garrison | Principal Mass Storage Specialist
ecgarris@iu.edu     | Indiana University - Research Storage
317-278-1207        | Jabber IM: ecgarris@iupui.edu

--------------000700080805000204040203
Content-Type: text/x-vcard; charset=utf-8;
 name="ecgarris.vcf"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
 filename="ecgarris.vcf"

begin:vcard
org:Indiana University;Research Storage
adr:;;535 W. Michigan st;Indianapolis;IN;46202;USA
title:Principal Mass Storage Specialist
version:2.1
end:vcard


--------------000700080805000204040203--