/rev/7856dceaf7b5
changeset: 1331:7856dceaf7b5
user: Martin Geisler <[email protected]>
date: Tue Oct 27 10:13:06 2009 +0100
summary: benchmark: display throughput
When I visited IBM Zurich last year, I talked with Christian Cachin
about VIFF and about our benchmarks. I explained that a multiplication
took about 1 ms when you do many in parallel.
He remarked that this notion of the "average time when doing many in
parallel" is a strange non-standard notion. We should instead report
the *throughput* (the number of sustained operations per second). This
is the standard notion used when measuring the performance of systems.
diffstat:
apps/benchmark_classes.py | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diffs (11 lines):
diff -r 6f02ecfa60e1 -r 7856dceaf7b5 apps/benchmark_classes.py
--- a/apps/benchmark_classes.py Sat Oct 24 18:39:19 2009 +0200
+++ b/apps/benchmark_classes.py Tue Oct 27 10:13:06 2009 +0100
@@ -40,6 +40,7 @@
print
print "Total time used: %.3f sec" % (stop-start)
print "Time per %s operation: %.0f ms" % (what, 1000*(stop-start) / count)
+ print "Throughput: %d per second" % (count / (stop-start))
print "*" * 6
return x
_______________________________________________
viff-commits mailing list
[email protected]
http://lists.viff.dk/listinfo.cgi/viff-commits-viff.dk
