Fw: [Beowulf] Correct networking solution for 16-core nodes

Vincent Diepeveen diep at xs4all.nl
Fri Aug 4 12:52:06 PDT 2006


Greg,

Look who's talking.

Read the subject. you posted onto this list answerring a private email of 
mine
which i had shipped to the guy starting this thread.

No clue how that private email reached you.

Yet you posted it here.

So i felt free to do the same with an email from you.

Vincent

p.s.

Note that you as a designer of an interconnect and claim that i don't 
understand how to take
advantage of it for latency sensitive programs such as mine which sometimes 
ship a megabyte or 200
and majority of the time ship around very short messages which are latency 
sensitive,
then feel free to explain how to actually USE your hardware for my software 
when a node has many
cores and just 1 network card.

You don't design that thing to let it look good on paper.

Do you?

Don't you design that thing for CUSTOMERS?

Reality is probably even worse; that you don't design for customers, just 
for a few limited specific 'benchmarks',
just like processors get created for benchmarks too, and don't care for 
actual software that has to run on it,
let alone 'explain' how to take advantage of it for non-embarrassingly 
parallel software.

The reality is that i never see you warn for 'worst case behaviours' when 
working with highend networks, which is what helps
most software get speeded up a lot and work better with networks; OTOH is 
see you being upset when a card yours is quoted
to be 1.4 us instead of 1.395 us.

---- Original Message ----- 
From: "Greg Lindahl" <greg.lindahl at qlogic.com>
To: "Vincent Diepeveen" <diep at xs4all.nl>
Cc: <beowulf at beowulf.org>
Sent: Friday, August 04, 2006 4:43 PM
Subject: Re: Fw: [Beowulf] Correct networking solution for 16-core nodes


> p.s. Don't ever post someone's personal email to a mailing list
> without asking. It's considered unethical by many on the Internet.
> 




More information about the Beowulf mailing list