[SERVER-18675] WiredTiger default configuration should be adjusted for laptop (e.g. default cache size should be <2G) Created: 27/May/15  Updated: 25/Jun/15  Resolved: 14/Jun/15

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

Type: Bug Priority: Major - P3
Reporter: Eitan Klein Assignee: Daniel Pasette (Inactive)
Resolution: Won't Fix Votes: 0
Labels: 32qa, WTAD
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Related
related to SERVER-19135 Tune default cache size settings for ... Closed
Operating System: ALL
Participants:

 Description   

WiredTiger default configuration should be adjusted for laptop (e.g. default cache size should be <2G)

On laptops (mac or windows client) , when users evaluate with WT

1) The current default of ½ RAM size likely to be too greedy and will cause a significant competition on RAM w/ other apps (which leads to high amount of page faults)
2) The situation will be amplified if users will evaluate a replica set as our mongod apps will compete on memory
3) The laptop processor HW is usually lower then server (no L3 cache and etc)
4) Higher cache size longer I/O pressure during checkpoint
5) My initial verification on an empty Mac (OSX) produce the following results

YCSB 50/50 load phase (insert only) measured on 3.1.4-pre-

  1/2 memory e.g 4 G 1G cache %Improvement w/ lower cache
Median Throughput 3984 5860 140%
Avg Throughput 4425 5405 122%

It’s easy to detect the Mac OS or windows client platforms and adjust our default for this platform



 Comments   
Comment by Martin Bligh [ 25/Jun/15 ]

Set to 60% of available - 1GB after discussion
The key difference that came up here was having more pagecache might be useful as it's compressed storage.

Comment by Daniel Pasette (Inactive) [ 14/Jun/15 ]

After further discussion, we may adjust the amount of memory devoted to the WT cache, but since we cannot anticipate the workload nor the "type" of machine (laptop or production server) mongodb is deployed on, we will likely opt to use more memory.

Current thought is to use whichever is greater of (80% of total available - 2GB) or 1GB.

Generated at Thu Feb 08 03:48:25 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.