If we run this test as a auto regression test, as we do in jtreg or jprt
job, I don't think the "benchmark"
number matters and the resulting numbers probably will not be checked
(manually), especially 1000
or even 10000 normally might not bring ina reliable benchmark number
anyway, so I guess it probably
not worth the testing circle to iterate more than 1. But not a big deal
for me, if you strongly believe
a 10000 iteration result might help some time, I can check in the
iteration as 10000.
-Sherman
On 11/25/2011 2:51 AM, Alan Bateman wrote:
On 24/11/2011 22:47, Xueming Shen wrote:
Alan, Ulf,
Webrev has been updated accordingly
(1) {@code}
(2) better wording for opening sentence as suggested
(3) added -benhmark option for TimeCheckSum, so only do the benchmark
run
(with 100000 iteration) when this option is manually specified.
Without the
option, it runs as a "normal" regression test with iteration 1.
(4) to use "ratio" of "time of ByteBuffer: time of byte[]" as
http://cr.openjdk.java.net/~sherman/7109837/benchmark
http://cr.openjdk.java.net/~sherman/7109837/webrev
This looks good to me although I think you've over corrected on the
default iteration count in the test. I think it would be fine to have
a default of 1000 or 10000, I was really just concerned about >=100000
which would cause the test to run for a long time on a slow system.
-Alan.
