[Beowulf] Correct Units for quoting Interconnect Bandwidth ?

[john.brookes at quadrics.com]

The point is that Pallas are wrong to quote Mebibytes under the heading 'MBytes/s'. Mebi is an SI standard prefix with the abbreviation 'Mi', so they should either have 'MiBytes/s' or actually quote MBytes/s. Just an oversight, I assume.

John Brookes

> -----Original Message-----
> From: Daniel Kidger 
> Sent: 28 May 2004 13:37
> To: beowulf at beowulf.org
> Subject: [Beowulf] Correct Units for quoting Interconnect Bandwidth ?
> 
> 
> 
> I have spotted an anomaly when comparing b_eff results with PMB:
>    http://www.hlrs.de/organization/par/services/models/mpi/b_eff/
>    http://www.pallas.com/e/products/pmb/
> 
> The former uses 1.e6 when displaying 'MBytes/s', where that 
> latter appears to 
> use 2^20.   (aka 'Mebibytes/s')
> viz from PMB's Output.c:
> 
> static double MEGA = 1.0/1048576.0;
> if( tmax > 0. ) throughput = (Bmark->scale_bw*SCALE*MEGA)*size/tmax;
> SCALE=1000000 (declare.h)
> size is in bytes, tmax in usec, and scale_bw=2 if say doing PingPong 
> 
> The result is that PMB figures appear *lower* than equivalent 
> b_eff figures.
> 
> 
> Do we have a concensus of what units we should be using?
> 
> So does a 1GB/s interconnect transfer 1 Gigabyte in a second 
> or 1,000,000,000 
> bytes in a second? (As interconnects move from Kilo though 
> Mega to Giga  the 
> gap is of course widening)
> 
> 
> -- 
> Yours,
> Daniel.
> 
> --------------------------------------------------------------
> Dr. Dan Kidger, Quadrics Ltd.      daniel.kidger at quadrics.com
> One Bridewell St., Bristol, BS1 2AA, UK         0117 915 5505
> ----------------------- www.quadrics.com --------------------
> 
> 
> 
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) 
> visit http://www.beowulf.org/mailman/listinfo/beowulf
>