Networker

Re: [Networker] Time differences in writing 1Gb

2004-07-21 05:12:51
Subject: Re: [Networker] Time differences in writing 1Gb
From: "Maarten Boot (CWEU-USERS/CWNL)" <Maarten.Boot AT NL.COMPUWARE DOT COM>
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Date: Wed, 21 Jul 2004 11:12:37 +0200
Thanks Mike ,

That is very helpfull as is the most accurate description of the phenomenon
that we see here. I did find hints to different drivers for HP on the web but
no luck so far for Solaris.

Your description also suggests to be of a a physical nature rather than a
software issue. It may also hint at the reason why different tape vendors
have different results. The conductivity for static electricity could be
different.

Anyway thanks for your input,

Maarten



On Tuesday 20 July 2004 18:16, Mike Schwarz wrote:
> We saw the same thing earlier this year with a new STK tape library with
> IBM FC ULTRIUM2 drives with the same firmware (38D0).  This library is
> connected to both an AIX p630 and HP L2000 servers.  I wrote a perl program
> to simulate the networker saves (2Gb files of 64Kb blocks) and report times
> and rates.  The normal file write was around 63 seconds with occasional
> files taking up to 600 seconds.  We got rid of the sympton by switching to
> the IBM Atape.driver which doesn't have a write timeout that the default
> tape driver has.  The HP server didn't timeout and so didn't show the
> problem.
>
> All this was documented and submitted to STK who then submitted it to IBM.
> I havn't got back any resolution from STK or IBM other than they are
> looking into it.
>
> The STK tech indicated they had seen problems traced to static in the tapes
> and that after the tapes are used a few times the problems went away.  I
> also noticed the problems are not consistent.  Different spots on the tape
> with different passes.  Even had days with no failures (long write times)
> and then lots of failures the next day on the same tape and tape drive.
>
> What we did was to use drivers with no (or very long) write timeouts.
> Other than the backup will ocassionally stall for several minutes,
> everything is running and the tapes are filling up (300-500Gb per tape).
>
>
> On Mon, 19 Jul 2004 12:09:30 +0200, Maarten Boot (CWEU-USERS/CWNL)
>
> <Maarten.Boot AT NL.COMPUWARE DOT COM> wrote:
> >Networkers,
> >
> >Last week I wrote a small perl program to verify my LTO-2 tapes.
> >I needed it after too many LTO tapes were marked full between 20 and 50
> > GB. On some tapes (not all) I see some interesting time differences on an
> > otherwise idle machine
> >
> >Hardware used:
> >
> >Sun E450 with 2 400 Mhz CPU, 1 GB mem
> >scsi card:  LSILogic LSI22320-R
> >Exabyte 110L with IBM ULTRIUM2 drives (firmware: 38D0)
> >
> >data written is a 128k block of gzipped random data from /dev/random

--
Maarten Boot,
Compuware Europe B.V.
Hoogoorddreef 5
1101 BA Amsterdam

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

<Prev in Thread] Current Thread [Next in Thread>