Comments on: Better Than Linear Scaling https://www.percona.com/blog/better-linear-scaling/ Fri, 16 Jun 2023 19:21:08 +0000 hourly 1 https://wordpress.org/?v=6.5.2 By: Joe Koenig https://www.percona.com/blog/better-linear-scaling/#comment-10968047 Tue, 02 May 2017 19:15:10 +0000 https://www.percona.com/blog/?p=42896#comment-10968047 Does this at all account for joins across tables that exist on different backend servers?

]]> By: Baron Schwartz https://www.percona.com/blog/better-linear-scaling/#comment-10968033 Thu, 27 Apr 2017 14:37:35 +0000 https://www.percona.com/blog/?p=42896#comment-10968033 Neil Gunther, the author of the USL, wrote a paper about superlinear scaling (in Hadoop) and has worked on modeling it for years. The USL does model and explain it as-is.

]]> By: Jouni Järvinen https://www.percona.com/blog/better-linear-scaling/#comment-10968021 Mon, 24 Apr 2017 12:31:00 +0000 https://www.percona.com/blog/?p=42896#comment-10968021 I took your nodes/throughput/… table and added speedups when comparing a higher node count to the previous smaller one. Here are the results:

1 node=NOT APPLICABLE
2 nodes=2.78
3 nodes=2.43525179856115
4 nodes=1.65731166912851
5 nodes=1.23083778966132
6 nodes=1.03837798696597

As you can see, you get decreasing benefit from setting up mode nodes. 2 and 3 nodes are still a big improvement, and sorta the 4th node too, but after 4 nodes the benefit doesn’t increase so much with each added node. That’s likely cuz the per-node allocation ain’t big enough for this load.

]]>