Re: RFC mailbox interface

[James McPherson - TSG Engineer <James McPherson Sun COM>]

On 26 Nov 2001, 11:34:16 AM Carlos Morgado wrote:
> On Sun, Nov 25, 2001 at 08:30:33PM +0100, M . Thielker wrote:
> > >> Well, it's possible to find out. As a matter of fact, if I were to write 
> > >> this, I would gather perf stats while opening the mailbox and reading 
> > >> the headers, then assign "speed indices" to each...
> > > clever. except it doesn't work for nfs :) --
> > Why not? Read benchmarking should work on anything. Collect stats over 
> > time, and adjust the values at runtime, continuously.
> > 
> ohhh urgghhh

And rightly so! Melanie, have you ever done read benchmarking? How big is the 
largest network environment you have worked in on a day to day basis? How would
your dynamic runtime updates handle somebody maxing out the link through your
major mailserver in order to download a powerpoint presentation of 22 Mb? I've
known people who do that, totally killed (ie, mega-bad filesystem corruption)
the corporate mailserver, and who don't actually care about how much bandwidth
they chew up. Have you?

How would continuous runtime (realtime?) updates get generated? with a thread
in an RT scheduling class? what if the code was bad and the thread never released
the cpu? Even if you have a 100Mbit switched connection to your mailserver, your
access might be sufficient to drive it into the ground if it isn't tuned. 

It might be a nice idea in theory but with the state of the world's computing
and network hardware at the moment I don't think this baby will fly unless you're 
testing in a highly specialised lab.

Sorry to be such a wet blanket, but the idea just has so many issues associated
that I had to comment.

James

(ps, Carlos don't you sleep on the weekend?)

-- 
TSG Engineer (Kernel/Storage)           828 Pacific Highway
APAC Customer Care Centre               Gordon NSW 
Sun Microsystems Australia              2072

Failfast panic: those controlling voices in my head have 
stopped telling me what to do.....

Read about the VOS Initiative at http://www.vosinitiative.com