Breaking News
Design How-To

Lessons learned: Network-based processing v. host-based processing

NO RATINGS
Page 1 / 2 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
docdivakar
User Rank
Manager
re: Lessons learned: Network-based processing v. host-based processing
docdivakar   9/27/2010 11:17:45 PM
NO RATINGS
@Shainer: I would have liked to see a cost vs. performance statistic between off-loading and on-loading in your article above. But I do agree with providing meaningful benchmark comparisons to compare latencies between off-loading and on-loading approaches, a point you quite well make above. It would seem that the reduction in through put (in case of on-loading) far outweighs the gains one may make in reduced latencies of the signal path (highspeed mother board traces and on to the HCA/TCA ports). I favor with off-loading since it truly helps improving the latency of your cluster as a whole where packets can travel to and from heterogeneous interconnects. With a TOE/POE off-loaders, 10Gig Cat6A fabrics can fulfill the needs of some lesser stringent (latency) applications. Dr. MP Divakar

shainer
User Rank
Rookie
re: Lessons learned: Network-based processing v. host-based processing
shainer   9/17/2010 7:43:43 PM
NO RATINGS
Yes - look at the TOP500 development over time as an example (http://www.mellanox.com/pdf/applications/Top500_June_2010.pdf). InfiniBand started with 3 systems back in 2003, and now is connecting 208 systems, 42% of the Top500. At first InfiniBand took nearly all of the proprietary systems (in 2005 there were more than 200 Myricom connected systems, now it is around 5 if I remember correctly), and later InfiniBand start to take systems from GigE. Check the graphs on the pdf for more info.

krisi
User Rank
CEO
re: Lessons learned: Network-based processing v. host-based processing
krisi   9/16/2010 8:39:16 PM
NO RATINGS
Gilda, You write that ""By providing low-latency, high-bandwidth and extremely low CPU overhead, InfiniBand has become the most deployed high-speed interconnect, replacing proprietary or low-performance solutions"". Do you have any statistic to back up that claim? I remember that although InfiniBand was an elegant solution it was not deployed widely as other technologies as GbE was enjoining larger market share...Kris

Top Comments of the Week
August Cartoon Caption Winner!
August Cartoon Caption Winner!
"All the King's horses and all the KIng's men gave up on Humpty, so they handed the problem off to Engineering."
5 comments
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
EE Times on Twitter
EE Times Twitter Feed
Radio
LATEST ARCHIVED BROADCAST
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.
Flash Poll