strange performance differences of 1.2.2 vs 1.2.5
dormando
dormando at rydia.net
Sat May 31 00:35:33 UTC 2008
First I've heard of this :(
Any chance you could do a git bisect on the tree to narrow down the
performance drop?
git clone http://consoleninja.net/code/memcached/memcached.git - stable
tree (sorry, haven't been ready to announce it yet).
-Dormando
Chris Goffinet wrote:
> Hi
>
> I was wondering if anyone here has experience with FreeBSD 6.2? I was
> compiling 1.2.5 for one of the groups at Yahoo and they saw
> dramatic performance decreases with 1.2.5 over 1.2.2. Has there been an
> introduction of code I am forgetting that may cause this?
>
> HP DL140 G3, 2 x Xeon L5320 1.86GHz, 16GB / 16GB 533MHz DDR2
> FreeBSD 6.2 - 64bit
>
>
> Here are some of the benchmarks:
>
> 1) memcached-1.2.2 (no threads)
>
>
> --- dimtest: 20 clients ---
> test count min max average stddev median
> dimtest>found 2268989 0.29 7.66 0.62 0.11 0.61
> dimtest>missed 1517569 0.24 5.66 0.60 0.12 0.59
> dimtest 3786558 0.24 7.66 0.62 - -
> ----
> ops 3786558
> time 120.08
> ops/sec 31534.53
>
> --- dimtest: 21 clients ---
> test count min max average stddev median
> dimtest>found 2371857 0.25 17.72 0.63 0.13 0.61
> dimtest>missed 1569835 0.23 138.64 0.61 0.18 0.59
> dimtest 3941692 0.23 138.64 0.62 - -
> ----
> ops 3941692
> time 120.08
> ops/sec 32825.82
>
> --- dimtest: 22 clients ---
> test count min max average stddev median
> dimtest>found 2458026 0.29 122.33 0.63 0.26 0.62
> dimtest>missed 1637046 0.24 122.28 0.62 0.36 0.60
> dimtest 4095072 0.24 122.33 0.63 - -
> ----
> ops 4095072
> time 120.09
> ops/sec 34099.87
>
> --- dimtest: 23 clients ---
> test count min max average stddev median
> dimtest>found 2545803 0.27 10.53 0.64 0.11 0.62
> dimtest>missed 1707357 0.23 8.01 0.62 0.11 0.61
> dimtest 4253160 0.23 10.53 0.63 - -
> ----
> ops 4253160
> time 120.09
> ops/sec 35417.12
>
> resource utilizations on server machine:
>
> Time user nice sys intr idle
> 05/17-13:20 2.7 0.0 21.8 9.1 66.4
> 05/17-13:25 3.1 0.0 25.5 10.5 60.9
> 05/17-13:30 3.4 0.0 28.7 11.8 56.2
>
> cpu mem ad0 / /home bge0-%util cpueff
> Time %util %util %busy %full %full in out kb/MHz
> 05/17-13:20 33.5 9.7 0.0 31.4 2.2 2.6 17.6 40.50
> 05/17-13:25 39.0 9.7 0.0 31.4 2.2 2.9 19.2 37.96
> 05/17-13:30 43.8 9.7 0.0 31.4 2.2 3.1 20.5 36.20
>
>
>
> 2) memcached-1.2.5 (without threads)
>
> --- dimtest: 20 clients ---
> test count min max average stddev median
> dimtest>found 1337938 0.24 49.31 1.10 1.65 0.62
> dimtest>missed 891894 0.20 46.80 1.01 1.52 0.59
> dimtest 2229832 0.20 49.31 1.06 - -
> ----
> ops 2229832
> time 120.06
> ops/sec 18572.85
>
> --- dimtest: 21 clients ---
> test count min max average stddev median
> dimtest>found 1370150 0.23 43.53 1.12 1.73 0.62
> dimtest>missed 912964 0.20 39.47 1.03 1.60 0.59
> dimtest 2283114 0.20 43.53 1.09 - -
> ----
> ops 2283114
> time 120.06
> ops/sec 19015.86
>
> --- dimtest: 22 clients ---
> test count min max average stddev median
> dimtest>found 1396775 0.24 43.19 1.15 1.80 0.62
> dimtest>missed 929857 0.20 38.02 1.07 1.68 0.59
> dimtest 2326632 0.20 43.19 1.12 - -
> ----
> ops 2326632
> time 120.07
> ops/sec 19377.62
>
> --- dimtest: 23 clients ---
> test count min max average stddev median
> dimtest>found 1367564 0.24 58.26 1.21 1.95 0.62
> dimtest>missed 923226 0.20 44.59 1.16 1.88 0.60
> dimtest 2290790 0.20 58.26 1.19 - -
> ----
> ops 2290790
> time 120.06
> ops/sec 19079.85
>
> resource utilizations:
>
> | CPU Share (%)
> Time user nice sys intr idle
> 05/30-13:30 0.0 0.0 0.1 0.0 99.9
> 05/30-14:00 2.9 0.0 59.6 7.5 30.0
> 05/30-14:30 0.1 0.0 0.1 0.0 99.7
>
> cpu mem ad0 / /home bge0-%util cpueff
> Time %util %util %busy %full %full in out kb/MHz
> 05/30-13:30 0.1 10.2 0.0 31.4 2.2 0.0 0.0 2.31
> 05/30-14:00 70.0 10.2 0.0 31.4 2.2 1.1 7.4 8.17
> 05/30-14:30 0.3 10.3 0.1 31.4 2.2 0.0 0.0 1.05
>
>
>
>
>
> 3) memcached-1.2.5 (threads)
>
>
> --- dimtest: 20 clients ---
> test count min max average stddev median
> dimtest>found 932156 0.28 50.66 1.55 1.88 0.77
> dimtest>missed 619288 0.24 41.38 1.50 1.82 0.74
> dimtest 1551444 0.24 50.66 1.53 - -
> ----
> ops 1551444
> time 120.05
> ops/sec 12923.25
>
> --- dimtest: 21 clients ---
> test count min max average stddev median
> dimtest>found 958508 0.28 36.41 1.59 1.89 0.78
> dimtest>missed 633208 0.23 32.77 1.53 1.83 0.75
> dimtest 1591716 0.23 36.41 1.57 - -
> ----
> ops 1591716
> time 120.05
> ops/sec 13258.27
>
> --- dimtest: 22 clients ---
> test count min max average stddev median
> dimtest>found 970551 0.28 38.60 1.64 1.96 0.80
> dimtest>missed 649273 0.24 36.38 1.58 1.91 0.77
> dimtest 1619824 0.24 38.60 1.61 - -
> ----
> ops 1619824
> time 120.06
> ops/sec 13492.04
>
> --- dimtest: 23 clients ---
> test count min max average stddev median
> dimtest>found 986869 0.26 42.54 1.68 2.02 0.81
> dimtest>missed 663471 0.23 41.46 1.62 1.98 0.78
> dimtest 1650340 0.23 42.54 1.66 - -
> ----
> ops 1650340
> time 120.06
> ops/sec 13746.53
>
> Resource Utilzations:
>
> | CPU Share (%)
> Time user nice sys intr idle
> 05/30-16:00 0.0 0.0 0.1 0.0 99.9
> 05/30-16:30 3.1 0.0 71.9 6.9 18.1
>
> bash-3.2$ more stats_1_2_5_threads_util
> cpu mem ad0 / /home bge0-%util cpueff
> Time %util %util %busy %full %full in out kb/MHz
> 05/30-16:00 0.1 10.2 0.0 31.4 2.2 0.0 0.0 1.95
> 05/30-16:30 74.5 10.2 0.0 31.4 2.2 0.8 5.3 5.45
>
>
> --
> Chris Goffinet
> MyBlogLog Senior Performance Engineer
> *
> *
> Yahoo!
> San Francisco, CA
> United States
>
More information about the memcached
mailing list