BDR Performance
BDR's performance is very impressive. Our own tests using pgbench and (2ndQuadrant benchmarks testing on custom workload with a latency of <2 seconds) reveal BDR's significant advantages:
1. Basic pgbench test: the overhead of BDR is very low and is very close to Hot Standby.
The small number for Londiste and Burcardo on Standby are the result of setting wal_level=minimal.
![](https://www.2ndquadrant.com/media/filer_public_thumbnails/filer_public/13/49/13494779-cb3c-4828-af4a-460fd3648ebf/bdr_p_chart_1_1_display.png__650x547_q85_subsampling-2.jpg)
2. Transaction performance: again BDR's performance is very close to Hot Standby
![](https://www.2ndquadrant.com/media/filer_public_thumbnails/filer_public/47/cb/47cb0380-1744-462a-83d1-46dcfb92a8a9/bdr_p_chart_2_display.png__484x408_q85_subsampling-2.jpg)
3. Latency: BDR avoids trigger processing so achieves a low latency per transaction.
Latency measured in milliseconds
![](https://www.2ndquadrant.com/media/filer_public_thumbnails/filer_public/53/c0/53c07010-14a3-41c1-a752-3b1784afb99a/bdr_p_chart_3_display.png__439x415_q85_subsampling-2.jpg)
4. Custom workload throughput with latency < 2 seconds.
![](https://www.2ndquadrant.com/media/filer_public_thumbnails/filer_public/f3/80/f380c2da-db76-481a-bfc0-9ea7eee58dd7/bdr_p_chart_4_display.png__650x498_q85_subsampling-2.jpg)
Performance setup: x i2.4xlarge. Intel Xeon E5-2670 v2.122GB RAM. 4x800GB SSDs Raid 0. Minimally tuned setup. pgbench scale 50, -c 32 -j 32
Comments
Post a Comment