转贴请声明原文: http://blog.csdn.net/duck_genuine/article/details/8644412
solrCloud的压力测试:
上一篇文章提过使用solr3.6做的架构优化。 solr 搜索架构优化
测试的时候还没有做更大数据量的并发请求压测试,最近来了新机器,比之前更加好,但没想要把索引放内存,而是放在ssd,内存是满 足的,但使用高于线上2倍多请求的时候,响应就没有之前那么良好,有下降的趋势,即使再加多8个core顶住 ,表现还是差差的。
先看一下非分多结点的测试数据:
4020832 |
#50ms以下 |
2735841 ,68.04% |
#50ms -100ms |
364699 ,9.07% |
#100ms -200ms |
401903 ,10.00% |
#200ms -300ms |
191195 ,4.76% |
#300ms -500ms |
156065 ,3.88% |
#500-1000 |
139782 ,3.48% |
#1000-2000 |
20087 ,0.50% |
#2000-3000 |
11218 ,0.28% |
#3000-5000 |
41 ,0.00% |
#5000-以上 |
1 ,0.00% |
基于3.6.1设计的多结点测试数据:8结点*2 双备负载
响应时间 |
1122万请求 |
1184万 请求数 |
983万请求 |
#50ms以下 |
9032936 ,80.44% |
9859400 ,83.25% |
8492507 ,86.39% |
#50ms -100ms |
730948 ,6.51% |
816640 ,6.90% |
566203 ,5.76% |
#100ms -200ms |
679844 ,6.05% |
751707 ,6.35% |
555547 ,5.65% |
#200ms -300ms |
218665 ,1.95% |
198764 ,1.68% |
129092 ,1.31% |
#300ms -500ms |
166813 ,1.49% |
104998 ,0.89% |
55841 ,0.57% |
#500-1000 |
173188 ,1.54% |
52902 ,0.45% |
21114 ,0.21% |
#1000-2000 |
44532 ,0.40% |
11318 ,0.10% |
268 ,0.00% |
#2000-3000 |
180387 ,1.61% |
42332 ,0.36% |
9754 ,0.10% |
#3000-5000 |
923 ,0.01% |
3565 ,0.03% |
2 ,0.00% |
#5000-以上 |
428 ,0.00% |
597 ,0.01% |
0 ,0.00% |
加双倍压力测试,并且再多加4结点负载。总共是(8*3),负载看一下效果:
响应时间 |
总量 1970万 = |
6985110 + |
6355850 + |
6362254 |
#50ms以下 |
13824099 ,70.16% |
4570760 ,65.44% |
4625922 ,72.78% |
4627417 ,72.73% |
#50ms -100ms |
1397816 ,7.09% |
595075 ,8.52% |
400213 ,6.30% |
402528 ,6.33% |
#100ms -200ms |
1433964 ,7.28% |
711878 ,10.19% |
360096 ,5.67% |
361990 ,5.69% |
#200ms -300ms |
655878 ,3.33% |
318562 ,4.56% |
169003 ,2.66% |
168313 ,2.65% |
#300ms -500ms |
764824 ,3.88% |
298421 ,4.27% |
232987 ,3.67% |
233416 ,3.67% |
#500-1000 |
802516 ,4.07% |
261791 ,3.75% |
269960 ,4.25% |
270765 ,4.26% |
#1000-2000 |
263937 ,1.34% |
148634 ,2.13% |
57885 ,0.91% |
57418 ,0.90% |
#2000-3000 |
538667 ,2.73% |
74715 ,1.07% |
231570 ,3.64% |
232382 ,3.65% |
#3000-5000 |
21222 ,0.11% |
5274 ,0.08% |
8040 ,0.13% |
7908 ,0.12% |
#5000-以上 |
285 ,0.00% |
0 ,0.00% |
170 ,0.00% |
115 ,0.00% |
当然这其中还是可以再做优化,不过再看一下solrCloud请求表现:
同样是双倍的压力请求:但却只有8*2的结点负载。cpu表现稳定,且请求响应分值更佳,暂时只测试跑一会,没跑长时间的数据。取了阶段性的数据,依然保持90%是在100ms以下,而且没有看到几个机器负载彪高的趋势 .
响应时间 |
2563693 |
2606345 |
2898226 |
#50ms以下 |
2168196 ,84.57% |
2207154 ,84.68% |
2449177 ,84.51% |
#50ms -100ms |
173866 ,6.78% |
175126 ,6.72% |
195621 ,6.75% |
#100ms -200ms |
167391 ,6.53% |
168675 ,6.47% |
190621 ,6.58% |
#200ms -300ms |
29169 ,1.14% |
29557 ,1.13% |
33537 ,1.16% |
#300ms -500ms |
10453 ,0.41% |
10755 ,0.41% |
12169 ,0.42% |
#500-1000 |
7348 ,0.29% |
7592 ,0.29% |
8556 ,0.30% |
#1000-2000 |
1887 ,0.07% |
1947 ,0.07% |
2354 ,0.08% |
#2000-3000 |
5297 ,0.21% |
5450 ,0.21% |
6102 ,0.21% |
#3000-5000 |
86 ,0.00% |
89 ,0.00% |
89 ,0.00% |
#5000-以上 |
0 ,0.00% |
0 ,0.00% |
0 ,0.00% |
暂时还没有测试并行读写的时候的性能。
转贴请声明原文: http://blog.csdn.net/duck_genuine/article/details/8644412
作者:duck_genuine 发表于2013-3-6 23:52:51
原文链接