Amanda-Users

Re: tapetape (not in faq-o-matic)

2006-05-23 16:45:47
Subject: Re: tapetape (not in faq-o-matic)
From: Joshua Baker-LePain <jlb17 AT duke DOT edu>
To: Brian Cuttler <brian AT wadsworth DOT org>
Date: Tue, 23 May 2006 16:39:08 -0400 (EDT)
On Tue, 23 May 2006 at 3:55pm, Brian Cuttler wrote

Does anyone have the tape type for the LTO3 (Quantum) ?

This is what I use:

define tapetype LTO3comp {
        # All values guesswork :)  jlb, 8/31/05
        # except blocksize ;) jlb, 9/15/05
        length 420000 mbytes
        blocksize 2048
        filemark 5577 kbytes
        speed 60000 kps
        lbl-templ "3hole.ps"
}

I leave hardware compression on (why not, with LTO), but most of my data isn't all that compressible. I get much better speed with the blocksize above (2MiB) rather than amanda's default of 64K. I determined that by testing raw write speed to the tape with tar and various blocksizes.

Are there any other parameters I should tweak to get better
performance/utilization ?

This is still a reasonable default ?

tapebufs 20

On Linux at least, with the blocksize above I had to dial back tapebufs to 15 or I got this warning in my nightly emails:

  taper: attach_buffers: (20 tapebufs: 41947616 bytes) Invalid argument
  taper: attach_buffers: (19 tapebufs: 39850440 bytes) Invalid argument
  taper: attach_buffers: (18 tapebufs: 37753264 bytes) Invalid argument
  taper: attach_buffers: (17 tapebufs: 35656088 bytes) Invalid argument
  taper: attach_buffers: (16 tapebufs: 33558912 bytes) Invalid argument

I am running the StorEdge C2 jukebox with lto3 drive on a SunFire 280R
under Solaris 9 with 4 gig of memory.

My usual warning with LTO3 is to make sure that your disks can keep up with your tape. Yes, you read that right. Especially with amanda dumping to holding disk while trying to write to tape, it's tough to feed LTO3 as fast as it wants to be fed (80MB/s native write speed). LTO3 can throttle back to half that without shoeshining the drive, but you don't want to see your write speeds below that.

--
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University

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