>
>
>Message: 28
>Date: Mon, 03 Oct 2005 11:22:40 -0500
>From: Les Mikesell <lesmikesell@xxxxxxxxx>
>Subject: Re: Symantec ghost and CentOS 3.4/4.0
>To: CentOS mailing list <centos@xxxxxxxxxx>
>Message-ID: <1128356560.29433.17.camel@xxxxxxxxxxxxxxxxxxxxxx>
>Content-Type: text/plain
>
>On Mon, 2005-10-03 at 11:06, Ted Kaczmarek wrote:
>
>>><SNIP>
>>>
>Or, if the machines are close to identical including the drives, just
>use dd to copy the whole disk. It's slow in real-time but takes very
>little human time.
><snip>
>
>
With proper use of the bs= parameter to dd, it can be very fast. Just
need to watch out that the jobs the node is doing at that time are not
unacceptably degraded, since we don't have a preemptive scheduler (? I
think... haven't been paying atention the last year or so.) yet.
Using a blockisize that takes advantage of cache sizes (several levels
of these) would be optimal, but the 80/20 rule seems to apply here. So I
always just use a block size that is a cylinder (not really a cyl, but
who knows what they really are these days?) based on the HD parameters.
What's really important is not the size, but the reduction in number of
system calls.
Anyway, not a major issue, but since I luv dd (and ed and other old time
real *IX stuff) I didn't want it to get a bad rap! ;)
> -- Les Mikesell lesmikesell@xxxxxxxxx
>
Bill
