[MXS-1666] Benchmarking ideas Created: 2018-02-13  Updated: 2022-07-18  Resolved: 2022-07-18

Status: Closed
Project: MariaDB MaxScale
Component/s: N/A
Affects Version/s: None
Fix Version/s: N/A

Type: Task Priority: Major
Reporter: Michaël de groot Assignee: Unassigned
Resolution: Won't Do Votes: 0
Labels: None


 Description   

Hi,

As requested, some benchmarking ideas for MaxScale:

  • How much extra latency is aded when MaxScale runs on the application server?
  • And how much when it runs in a separate instance?
  • How much decrease in throughput is there if MaxScale runs on a CPU-based workload?

The last one is a bit vague: My goal with this benchmark is to determine how much cost there is in extra hardware when a big project starts to use MaxScale. If 10.000 instances have 10% less throughput, then the hardware cost would be equal to 1000 instances.

 Comments   
Comment by markus makela [ 2022-07-18 ]

I'll just close this as it's been open for quite a while and nothing has been done for it.

I did do some testing many years ago and the conclusion was that the extra network hop is the thing that costs the most: using UNIX domain sockets to communicate with either the client or the database is what speeds up things the most.

Generated at Thu Feb 08 04:08:30 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.