very high bandwidth, low latency manner?
sp at scali.com
Thu Apr 11 20:37:17 PDT 2002
On Thu, 11 Apr 2002, Iwao Makino wrote:
> I think ... Quadrics<http://www.quadrics.com/> is another one.
Yep, sorry I forgot that one.
> Here's quick figures I have on hand....
> RH7.2, 2.4.9 kernel for i860 cluster.
> On their site, they claim;
> after protocol, of 340Mbytes/second in each direction. The
> process-to-process latency for remote write operations is2us, and 5us for
> MPI messages.
But this 340MBytes/second and 2us latency is also chipset dependent, as I
mentioned for SCI (in my examples latency was lowest on 760MPX but
bandwidth was highest on IA64 460GX...). I can't imagine that the i860 can
actually perform as well as 340MByte/sec since the Hub-Link (between
the MCH and the P64H) has a limit of 266MByte/sec (AFAIK) ....
> But pricing is MUCH higher than SCI/Myrinet.
> Best regards,
> At 4:08 +0200 5.04.2002, Steffen Persvold wrote:
> >On Thu, 4 Apr 2002, Jim Lux wrote:
> >> What's high bandwidth?
> >> What's low latency?
> > > How much money do you want to spend?
> >I don't want to start a flamewar here, but I _think_ (not knowing real
> >numbers for other high speed interconnects) that SCI has atleast the
> >lowest latency and maybe also the highest point to point bandwidth :
> >SCI application to application latency : 2.5 us
> >SCI application to application bandwidth : 325 MByte/sec
> >Note that these numbers are very chipset specific (as most high speed
> >interconnect numbers are), these numbers are from IA64. Here are numbers
> >from a popular IA32 platform, the AMD 760MPX :
> >SCI application to application latency : 1.8 us
> >SCI application to application bandwidth : 283 MByte/sec
Steffen Persvold | Scalable Linux Systems | Try out the world's best
mailto:sp at scali.com | http://www.scali.com | performing MPI implementation:
Tel: (+47) 2262 8950 | Olaf Helsets vei 6 | - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 | N0621 Oslo, NORWAY | >320MBytes/s and <4uS latency
More information about the Beowulf