[SERVER-64465] Investigate and improve the negative scaling at high-concurrency levels for FIFO ticketholder Created: 14/Mar/22  Updated: 03/May/22  Resolved: 03/May/22

Status: Closed
Project: Core Server
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Improvement Priority: Major - P3
Reporter: Jordi Olivares Provencio Assignee: Daniel Gomez Ferro
Resolution: Won't Do Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Problem/Incident
is caused by SERVER-64076 Fix deadlock when waiting for tickets Closed
Sprint: Execution Team 2022-04-04, Execution Team 2022-04-18, Execution Team 2022-05-02, Execution Team 2022-05-16
Participants:

 Description   

With the work on SERVER-64076 we've identified that at high load cases our system doesn't scale as well as with the previous semaphore based implementation. In particular, as soon as we begin enqueueing requests there's a drop in performance.

This ticket's scope is thus to investigate why this is the case and if possible to mitigate it.



 Comments   
Comment by Jordi Olivares Provencio [ 15/Mar/22 ]

A cursory investigation to the drop-off in acquisitions leads to the _queueMutex being extremely contended with a high number of threads. If possible, work should be devoted to investigating how to remove contingency on it or to remove it entirely altogether.

Generated at Thu Feb 08 06:00:24 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.