On Fri, 2002-01-11 at 12:42, Matthew S. Hallacy wrote:
 
> Figuring for 32K block size, each message claimed to be 7K, he could easily
> store 3,750,000 messages, with $MAIL_DAEMON set to just expire messages
> not delivered after say, 12 hours (he said after 3-4 they were useless)
> that's quite a bit of room to grow. (assuming a 120G drive dedicated to mail)
> 
> I'm not sure how important raid would be for long-term data reliability
> considering his comments about it being useless after a short amount of time.
> 

I don't think that the purpose of RAID, in this case, is for data
reliability, which doesn't sound like the key to this issue, but for
increased I/O performance.  Depending on how smart caching really is,
throwing a lot of memory at the problem might be even more rewarding,
and one way to make sure that that would scale up would be to set this
up on a Mosix cluster.  My entirely amateur understanding is that this
sort of process generates a lot of threads/processes that might well
benefit from parallelizing -- and if it does, the answer to increased
demand on the server would probably be, "well, let's just add some nodes
and add some more RAID arrays."  

But I could be wrong.

 
-- 
-------------------------------------
There's a widow in sleepy Chester
  Who weeps for her only son;
There's a grave on the Pabeng River,
  A grave that the Burmans shun,
And there's Subadar Prag Tewarri
  Who tells how the work was done.
-------------------------------------