Re: [Networker] 2 Part Question

2007-05-15 14:11:50
Subject: Re: [Networker] 2 Part Question
From: Darren Dunham <ddunham AT TAOS DOT COM>
Date: Tue, 15 May 2007 11:07:59 -0700
> One other comment on the discussion about faster restoring from a
> demultiplexed clone.  While it would be faster I wonder if it's really
> that much faster.

It's impossible to answer that question in isolation.  It depends on
several factors.  It may not be faster at all in some cases.

If your parallelism and backup device speed does not constrain the
client, then the doing a single client restore should go no faster with
a non-multiplexed saveset.

As an example:  
client - 100Mb/s network
drive - 60MB/s throughput

If I demultiplex my saveset, I'm still limited to 100Mb/s for data
transfer.  If I can sustain that from my original tape (presumably with
parallelism of less than 5 or so), then there's no difference.  As the
clients get faster, the parallelism has to go down for this to remain

> I guess the more streams you have in the multiplex the better
> your time saving would be from demultiplexing (more tape between each
> piece of the particular stream you are restoring increasing the seek
> time to find the next piece) so there would be a point where it becomes
> worthwhile but doubt our parallelism of 12 reaches that point.

Pretty much.  It's not an absolute number.  It's when contention for the
device slows the datastream below the speed of the client, and it's
probably best approximated as a ratio between the speed of the backup
device and the speed of the clients.  If we backed up two very fast
clients to a slow device, then they slow on the backup and slow on the
restore because of the contention.  Somewhere between the two extremes,
you're fully utilizing the device, but not contending for resources.
With fast clients, a parallelism of 12 is probably contending if you
have 12 full streams coming in.  You'd have to do a test to see how much
it matters for your restoration.

Absolute worst case is if you have a "pokey" client.  If it just pushes
data out in dribs and drabs while other clients are streaming full speed
ahead, then instead of getting 1/4th or 1/12 of a datastream, it might
be getting much less, like 1/100th.  Demultiplexing or keeping those
clients segregated can be a win when you're dealing with tape.

You also have a bit of a buffer in the fact that most clients restore
slower than they backup due to file creation overhead on the filesystem.

Darren Dunham                                           ddunham AT taos DOT com
Senior Technical Consultant         TAOS  
Got some Dr Pepper?                           San Francisco, CA bay area
         < This line left intentionally blank to confuse you. >

To sign off this list, send email to listserv AT listserv.temple DOT edu and 
type "signoff networker" in the body of the email. Please write to 
networker-request AT listserv.temple DOT edu if you have any problems with this 
list. You can access the archives at or
via RSS at