[rescue] SMP on intel wasteful?

[Big Endian]

>~ -----Original Message-----
>~ From: Chris Hedemark [mailto:chris at yonderway.com]
>~ Sent: Monday, June 24, 2002 1:33 PM
>~ To: rescue at sunhelp.org
>~ Subject: Re: [rescue] SMP on intel wasteful?
>~
>~
>~ Still, while that hardware is mighty studly, the dollar for dollar
>~ argument still trumps the CISC vs. RISC argument in terms of overall
>~ system performance.  The kind of people footing the bill for
>~ that exotic
>~ hardware are increasingly turning to cheapo CISC clusters and laughing
>~ all the way to the bank.
>
>
>Umm, no.
>
>Cheap hardware clusters cost same/more than RealBoxes in the end. My company
>is learning this now. Clone-Nix on Intel doesn't translate one-to-one or
>clockcycle-to-clockcylcle when compared to a box designed to be a Unix box
>running Unix. Been there, doing that now, have t-shirts. I'm in a unique
>position to know as we run identical services on equal number of solaris/sun
>boxes as linux/intel boxes. We aren't even using cheap intel boxes. Thank
>God our rewrite of the software was in something that can play on both linux
>and solaris.
>
>When you push linux/intel to the limits you push it righ off the table.

Ditto here.  I had a *PAIR* of IBM 4000R netfinities.  Dual 750 p3, 
gig ram, dual 9 gig scsi, the works.  These the best os for intel 
systems (freebsd) and postfix.  A *SINGLE* sun AXi @ 300mhz sends 
more messages with 1/100 of the system load.  Raw clock != 
performance.  It honestly depends on what you're doing.  Also 
price/performance of used Sun/SGI gear blows even brand new PC gear 
out of the water for I/O bound tasks.

daniel
-- 
-----------------------------------------------------------------
"Yay aytch lich ... trurtreen oven prerkrershetch... naw lich luleven 
... naw wush thrurteen..."
--"Thirteen fucking percosets?!?! Dude the most I ever ate was like 
six. Holy shit."
"Aywhirr schmall."
--"What, like 5's?"
"Mm mm mm... mebee..."

-posted on lowbrow.com