Re: Repository services and budget

[Michael Dillon <michael@memra.com>]

On Tue, 7 Jan 1997, Dave Crocker wrote:

> >etc.), a backup system (not simply tar to DAT), but something like ADSM
> >for networkign b/u's, 256Mb ot 512MB of memory for DB systems (most
> >software runs better with more memory and the DB systems are all memory
> >hogs), RAID or SSI disks so as to improve reliablity on the DB servers.
> >Basically $10K per server is too low in my opinion.  Look at $30K.
> 
> 	Given the full redundancy of systems, I would guess that RAID is
> (parden the pun) unnecessarily redunant.  RAID is important for avoiding a
> single-point of failure problem; but that is solved differently is the
> proposed, multi-server arrangement.  (Now, RAID for performance may well be
> a different issue.  I don't have a solid enough feel for current
> performance to claim to know the definitive answer on this.)

My experience is that even small database systems with 20 users or so
can suffer a significant performance hit when somebody runs a query that
scans significant chunks of the database. In the CORE scenario this could
happen when generating DNS zone files or generating a separate whois 
data source. Now if the software is properly designed by someone who
understands database performance issues a lot of this can be sidestepped
but installing a RAID 5 system with hot-swap chassis can be considered as
a form of insurance against these problems. In addition, it is no small
task to rebuild a broken server unless you have taken the precaution to do
a few dry runs and have a clearly documented procedure in advance. 

In the RAID 5 hot-swap scenario, somebody notices that a drive has failed,
but it's almost lunch time so let's not worry about it yet. After lunch
they pull one of the spares off the shelf, yank the broken drive and
install the replacement. All the while the system runs normally because
when the drive failed, it's data was automatically rebuilt onto a hot
spare and by the time the technician returned from lunch the server was
already back to full operational speed. If the RAID array uses lots of
1 gig or 2 gig drives (10 or more) then there is not much performance hit
during the rebuild. As far as the 3 system scenario goes, I have had
some gruesome experiences while trying to get operational data onto a
backup system and it is not uncommon to see two machines out of commission
during a crisis. 

Note that RAID 5 does not necessarily imply high end server hardware. I
have set up very nice systems with RAID 5 hot swap using
IBM PC-Server 320 machines with dual Pentiums and SCO UNIX.

> 	I'd guess that 60-$100K would suffice.  What do other folks think
> about this?

It can be done for that price. A lot depends on what brands of hardware,
which UNIX, which database (licensing issues) and who is doing the work.
Some high-priced consulting firms can be absolute bumbling timewasters.
People who choose their suppliers and products carefully always pay less
in the long run.

> 	Remember this this is a back-end organization.  It's "customers"
> are the member registrars.  Initially, that is a very small number.  After
> two years, it's still only around 40-50.  My guess is that technical
> support staff are going to be more important than a purely
> adminitrative/receptionis kind of person.

If you have on-staff tech people with some sort of manager to oversee them
as opposed to outside consultants, then I would agree that there is
no need for a secretary since office visits and phone calls (other than
faxes) will be rare.


Michael Dillon                   -               Internet & ISP Consulting
Memra Software Inc.              -                  Fax: +1-604-546-3049
http://www.memra.com             -               E-mail: michael@memra.com