|
RE: Slow recover speeds was amrecover failures
2006-07-11 07:58:37
> > Contention on the disk, especially if it is also
used for other
> > tasks, and compounded if inparallel is set very high. RAID5
is often
> > not high-performance. I've seen some that do well on a
single read
> > or write, but seriously degrade with a few simultaneous operations.
> > Either way, if the dump can't feed the tape fast
enough, the tape
> > keeps stopping and repositioning itself, vastly slowing your
throughput
> > (and wearing out your drive and tape). You might consider
increasing
> > tapebufs in your config to buffer more data in memory.
>
> Ok, I have increased tapebufs to 80, I'm not really sure how high
to
> increase it though. I do have plenty of free ram.
I've increased it to 1024 on my PowerEdge 2850 with
8GB RAM.
All I have is this:
NOTES:
taper: attach_buffers: (1024 tapebufs: 33570816 bytes) Invalid argument
Which, I believe, does not harm (From what I have
read till now, I'm guessing that the above message means that amanda uses
1023 tapebufs).
Note however that in my case, it didn't help (but
I'm not using amanda exactly as I should).
>
> I will see tonight how it performs. I don't think it's an amanda issue,
I
> personally believe it's a tape drive/scsi issue. The drive is
an LTO2 drive
> u320. Compression is off. Can a faulty terminator cause this speed
> degradation?
>
> Oldy write speed is much quicker than read speed! I just double checked,
> it's actually raid 1 on the /usr (& amanda dump) partition
>
> SRV# dd if=/dev/zero of=/usr/recover/test ^C5675166+0 records in
> 5675165+0 records out
> 2905684480 bytes transferred in 60.056881 secs (48382208 bytes/sec)
>
> SRV# dd if=/usr/recover/test of=/dev/null
> 5675166+0 records in
> 5675166+0 records out
> 2905684992 bytes transferred in 147.353463 secs (19719150 bytes/sec)
>
> SRV# dd if=/usr/recover/test of=/dev/null
> 5675166+0 records in
> 5675166+0 records out
> 2905684992 bytes transferred in 147.716610 secs (19670672 bytes/sec)
>
> SRV# ammt -f /dev/nsa0 rewind
> SRV# dd if=/dev/zero of=/dev/nsa0 bs=32k count=10000
> 10000+0 records in
> 10000+0 records out
> 327680000 bytes transferred in 122.037586 secs (2685074 bytes/sec)
Wow, that's bad for an LTO2!
Just a thought: Were you sitting next to the tape
drive while running the above test? Did you see it stopping and re-starting
over and over...?
>
>
> Cheers,
>
> Stavros Patiniotis
> EscapeNet ~ 08 8292 5200
>
|
|
|
