Chris, what do the numbers represent ?
You say you ran each test for 10 seconds, so I guess the numbers are not
the seconds it took, so what are they ?
I also wonder about the numbers, for example in the first column
(httpd). They seem to grow more or less lineraly as the file size
increases, but they at 512 KB they just level off.
It seems a bit counter-intutive that it would take "the same whatever"
to serve 512KB files and 32MB ones.
André
Christopher Schultz wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
All,
After reading some of your feedback, I've decided to make some changes:
- - Using TC 6.0.18 exclusively instead of 5.5
- - Using tcnative 1.1.16 instead of 1.1.12
- - Using httpd 2.2.11 instead of 2.2.10
- - Running tests for a certain amount of time instead of a
certain number of requests
I have some preliminary results. Just as a smoke-test, I ran my tests
for 10 seconds each (10 seconds per file size, per server config) which
means that I can get a complete set of results in 15 minutes. The
results are borderline useless, but you can already start to see the
different configurations differentiate themselves:
File Size Apache httpd Coyote Coyote APR Coyote APR -sendfile
Coyote
NIO Coyote NIO –sendfile
4KiB 4984.02 3833.73 5674.66 5433.23 3128.34 3247.66
8KiB 8795.03 7468.45 9465.31 10015.06 5616.81 5674.44
16KiB 15913.38 12901.21 16437.40 16426.36
10316.27 10171.56
32KiB 27525.07 21270.07 25361.09 25557.25
17482.09 17803.41
64KiB 47500.61 32990.81 38590.02 37454.34
31113.93 27034.23
128KiB 63920.72 42161.17 58548.64 46011.54 7167.93
31891.99
256KiB 80030.02 51749.21 82274.47 54119.36 256.29
34057.95
512KiB 95386.27 45987.15 89375.52 49531.11 512.10
30722.53
1MiB 105059.69 50127.84 89988.79 50886.45 1020.18
31309.67
2MiB 99790.56 51408.41 95647.38 44390.74 2032.38
32697.59
4MiB 100633.5 51138.52 105273.11 54729.79 4424.93
34088.29
8MiB 99595.03 51523.92 98445.83 56116.61 7936.50
32557.95
16MiB 99126.65 51440.45 98111.82 55406.30
15400.82 32681.28
32MiB 99018.94 52719.74 96605.48 54410.23
28989.75 33275.04
Quick setup: single localhost client (no concurrency), no keepalives, 10
second max samples per file size per server config.
I re-ran the NIO+sendfile tests afterward since the server was busy
(recompiling gcc as well as serving HTTP requests to whomever happens to
be using it right now) and I figured it was a fluke. The second test
showed the same results: NIO looks great until it hits the 128KiB file,
when it experiences a dramatic drop-off in performance. I'll have to
look into that: the NIO connector /without/ sendfile enabled does not
appear to suffer the same drop-off in performance (though it appears to
be the weakest contender in the bunch).
- -chris
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAkoRt2gACgkQ9CaO5/Lv0PCqZgCeN6zXJFY0E5kquz5//CsnaFGN
ZIwAn3b++7waMKoi9iJ2X0pyocKK7d/5
=UWiA
-----END PGP SIGNATURE-----
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
