Veritas-bu

[Veritas-bu] Re: Implications of turning off s/w compresion?

2002-04-02 11:24:11
Subject: [Veritas-bu] Re: Implications of turning off s/w compresion?
From: richard.hall AT ingenta DOT com (Richard.Hall)
Date: Tue, 2 Apr 2002 17:24:11 +0100 (BST)
My thanks to Jeff Kennedy, Dave High and Larry Kingery for being very
quick off the blocks. They've convinced me to go for it. To summarise
/ interpolate their answers ...

On Tue, 2 Apr 2002, Richard.Hall wrote:

> Recent discussion here has convinced me that we should never have set up
> NBU s/w compression in the first place. ("Told you so"!!)
>
> Before we turn it off ...
>
> 1) will this affect recovery from existing images?

[JK] No, at least it's not supposed to.
[DH] No

> 2) or mess anything else up?

[JK] Nope
[DH] No

> 3) (For academic interest only, and to make sure I understand what is
>    going on)
>    Since our DLT7000s are currently managing ~5Mb/sec, I assume that the
>    data is currently compressed on the client side, not decompressed again
>    on the server side, and the drives are (roughly) achieving no further
>    compression. And that, other things being equal, I can now expect to
>    see them managing 10 Mb/sec, but with no reduction in backup time.
>    Except perhaps in the tail of the backups, where the number of jobs and
>    the total data rate is falling.

[JK]

> Maybe not double, but definitely more.  I don't run compression and my
> DLT7k's generally run at 8-10MB/sec, but with hw compression that put's
> me at approx. 4-5MB/sec compressed, which is what they should be running
> at.  Occasionally I see my filers pushing them to 14MB/sec but those are
> burst speeds.  And in fact, backup jobs may run noticeably faster if the
> clients were cpu constrained to begin with.  Not having to do
> compression will lighten cpu load tremendously.

[DH]

> What you should see is a bit more compression due to hardware compression
> being more efficient than software compression and you *should* see the
> backups speed up a bit all things being equal. Software compression
> compresses on the Client (like you said) and then sends the compressed
> packets to the server. Sending a stream is more efficient. Speed could be
> 20% or so better but that depends on your LAN.

[LK]

> I wouldn't expect 10MB/s.  You might get it, but there's no reason to
> expect it.  2:1 has nothing to do with the real world.  You'll also
> need more than 100Mb/s.

OK, my message was over-simplistic. I understand why 2:1 is only
'illustrative'.

And my media server has Gigabit Ethernet (though the clients don't). That
plus a sufficient level of multiplexing ought to give me the bandwidth I
need, I reckon.

Thanks again.

Richard Hall