Hi all ,

I just ran the benchmark(TermsFacetSearchBenchmark) from elasticsearch source code, both version 0.20.6 and 0.90.2. I cleared the index data before each benchmark.

The output shows 0.20.6 is much faster than 0.90.2 on terms facet.

------------------ SUMMARY 0.90.2 -------------------------------

name took millis

terms_s 12.4s 62

terms_map_s 54.7s 273

terms_l 38.8s 194

terms_map_l 35.8s 179

terms_sm 51.8s 259

terms_map_sm 6.3m 1914

terms_lm 3.1m 942

terms_map_lm 3.1m 956

terms_stats_s_l 1.4m 441

terms_stats_s_lm 3.2m 984

terms_stats_sm_l 12.5m 3774

------------------ SUMMARY -------------------------------

------------------ SUMMARY 0.20. 6-------------------------------

name took millis

terms_s 9.8s 49

terms_map_s 23.6s 118

terms_l 9.7s 48

terms_map_l 19.3s 96

terms_sm 41.1s 205

terms_map_sm 3.1m 938

terms_lm 39.9s 199

terms_map_lm 2.2m 689

terms_stats_s_l 28.9s 144

terms_stats_s_lm 1.3m 415

terms_stats_sm_l 3.8m 1160

------------------ SUMMARY -------------------------------

And I used Jmeter to do a simple json terms facet benchmark on a 20 nodes cluster with 200gb data to get a average throughput , the result shows 0.20.6 is also better than 0.90.2.

So my question is can I come to a conclusion that 0.20.6 is better than 0.90.2 on terms facet from these two benchmarks? or what I did wrong?

From my side , I think 0.90.2 performance should be better than 0.20.6 because latest lucene version.

Thanks,