In article <1995Jun21.162255 at ebi.ac.uk>, Roy Omond <Roy.Omond at ebi.ac.uk> wrote:
>In article <DAI3s3.8t6 at murdoch.acc.Virginia.EDU>,
>wrp at dayhoff.med.Virginia.EDU (Bill Pearson) writes:
>>[...snip snip...]
>>> gbmam.seq is about 23Mbytes, so on the alpha, the program was reading
>> and searching >1 Mb/sec, without any degredation by simultaneous
>> searches elsewhere. fgrep on the same file took 33 sec, while cat
>> from the NFS system to the local system took 57 sec.
>>Jeez... 23 Mbytes in 20 seconds real time translates into something
>like 1.15 Mbytes/second which equates to 9.2 Mbits/second (excluding
>packet overhead).
>>Was this over a standard Ethernet ? If so, then that's more than
>92% utilisation...
This was standard ethernet, but in re-testing the searches I learned
that the timings, while accurate, were not informative. I did the
same search three times, and the real elapsed times were: 33, 18.4,
18.7, and 18.4 sec. It seems clear that the Alpha/OSF/1 is buffering
the NFS mounted databases, so that the network was not being used for
the 18 sec runs.
But even at 33 sec for 23 Mbytes, a large fraction of the bandwidth
is being used. (I do these things at night.)
Bill
--
wrp at virginia.EDU
Dept. of Biochemistry #440
U. of Virginia
Charlottesville, VA 22908
