elm@sprite.berkeley.edu (ethan miller) (06/03/91)
Judging from the information posted by Joseph Kubicky, the Touchstone
has far too little I/O bandwidth. This is already a problem with
the nCUBE, but it seems worse in the Touchstone Delta.
In article <1991Jun2.095813.17733@nntp-server.caltech.edu> jkubicky@nntp-server.caltech.edu (Joseph J. Kubicky) writes:
% - 520 numeric nodes w/i860's & 16MB each
% - 32 disk I/O nodes each w/one 1.4 GB disk (will double
% each node disk space in July '91). These nodes
% are 8MB i386's.
% - 2 tape I/O nodes
% - 2 netword nodes (Ethernet) i386, 8MB each
% - 6 service nodes - i386, 8MB
% - 2 High Performance Parallel Interfaces (HIPPI),
% each w/ i860 + 32MB mem
%Peak speed: 31.7 gigaflops (64-bit), 42.2 GFLOPS (32-bit),
% 17,000 MIPS
%Total mem: 8.4 GB
%On-line disk cap: 45 GB (increase to 90 GB in July)
%External I/O: HIPPI 800Mbit/sec (Oct. '91)
This means total disk bandwidth is 32 * 2 MB/sec = 64 MB/sec. This
assumes SCSI-type disks. If they're IPI-3, the bandwidth could be
as high as 100 MB/sec for the entire processor array. That means it
would take about 80 seconds to load the entire array, assuming all the
disks were busy all the time. I'd bet it'll be more like two or
three times that to load up (or dump) all of the memory. Throw
in the two HPPI channels, and you cut the time back down to 30-60
seconds (if you have a 100 MB/sec disk). Total I/O bandwidth is
at most 300 MB/sec for a 30+ GFLOP machine. Amdahl's metric suggests
1 b/sec for each FLOP. The Delta has 12.5 FLOPS per b/sec. Does
this mean that Amdahl's metric no longer holds, or will this machine
need more I/O nodes to be useful?
In any case, it seems like an interesting machine to use, but I'm
interested to see how the (relatively) slow I/O affects its use.
Anyone know if there are plans to attach more/faster I/O nodes?
ethan
=====================================
ethan miller--cs grad student elm@cs.berkeley.edu
#include <std/disclaimer.h> {...}!ucbvax!cs.berkeley.edu!elm
This signature will self-destruct in 5 seconds....