At 11:37 AM 9/28/99 -0500, Dave Lorand wrote:
>Paul,
>
>Thanks for your thoughts; they'll be helpful in designing our new cell
>structure. A few questions:
>
>At 6:47 AM -0500 9/28/99, Paul Blackburn wrote:
>>Hello Dave,
>>
>>Some thoughts about cell layout. At /afs/@cell, we have:
>>
>> @sys-dirs architecture specific binaries and libraries
>> rs_aix41 rs_aix42 rs_aix43 etc
>
>Do you install each package in a separate volume per @sys, or do you have a
>common volume containing all architechtures? This seems to be the major
>issue about which folks here have waflled in the past. Some packages, like
>Matlab, seem to go better if there is a separate volume for each
>architechture because the authors wrote with single-platform installations
>in mind. Others, like Perl and Sendmail, expect to mingle architechtures
>within their trees, and they know how to separate architechture-specific
>files from architechture-independant ones. I'd appreciate anyone's
>comments on this issue.
>
>The comments about home directories were also helpful. We don't have any
>home directories in AFS yet, but we'd like to move in that direction once
>we change /bin/login.
>
>Regards,
>
>Dave
> ____________________________________________________________
>| Dave Lorand, System Administrator | [EMAIL PROTECTED] |
>| Social Science Research Computing | 773-702-3792 |
>| University of Chicago | 773-702-2101 (fax) |
>+-----------------------------------+------------------------+
> ---> finger [EMAIL PROTECTED] for my PGP key <--
>
Dave,
Several folks have already posted some good ideas and strategies
for cell organization. A few other comments..
Basically when you go to set up your cell, think big. Consider
how you would best organize your cell with the stuff that you
need to make available now. Then, try to imagine that same setup
if you were to triple (or more) the amount of stuff online.
How well does it scale?
Basically you need a good directory structure for organization.
Can I predict where in the tree I would find something?
If not, reconsider your design.
As for the per @sys volumes, I am/was in favor of them. One of the
posts mentioned that there was no need unless the volumes grew
to a large size. You do need to consider backups - volume size
vs. number of volumes (this has been discussed previously on the
list, check the archives). If you have more of one type of OS than
another, more replicants should probably be devoted to the more
prevalent OS instead of just wasting diskspace to replicate seldom
used platform binaries. As well, if you are running short on
diskspace (or your boss is cheap :-) ) it is easier to relocate a
couple of smaller volumes across a couple of servers than it is to
find space for a 1 gig volume with all OS versions of a product in it.
The bottom line is that experience will often dictate how you set
things up. If you have luck with fewer larger volumes, then
go with that. Too many volumes can also be problematic. For
instance, why create 20 - 10 Meg volumes when one 200 Meg volume
might work just as well?
Cell setup is often a gray area. There isnt so much right
and wrong to it, just make sure that whatever you decide to
do is supportable, usable, and well thought through.
-Mike-
/===================================================================\
| Mike Caudill ([EMAIL PROTECTED]) --- |
| One does not discover new continents without consenting to lose |
| sight of the shore for a very long time. - Andre Gide |
\===================================================================/
