[SERVER-12428] 2.5.4 seems significantly slower than 2.4.x releases Created: 22/Jan/14  Updated: 11/Jul/16  Resolved: 22/Jan/14

Status: Closed
Project: Core Server
Component/s: Performance
Affects Version/s: 2.5.4
Fix Version/s: 2.5.5

Type: Bug Priority: Major - P3
Reporter: John Greenall Assignee: Unassigned
Resolution: Done Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified
Environment:

linux


Attachments: PNG File m1Large0iopsReadOnly2.4.9.png     PNG File m1Large0iopsReadOnly2.5.4.png    
Operating System: Linux
Steps To Reproduce:

Don't think I'm doing anything special. Just doing a large volume of queries should do it.

Participants:

 Description   

I have started using 2.5.4 for development (as I am using the new $min and $max operators). Noticed performance is much worse than 2.4.x versions.
Apologies if this is already a known error. Have searched for duplicate issue but can't find one.

Criteria:
I am volume testing using pymongo on amazon EC2. Mongod is running on an m1.large instance and I am querying from a separate box within the same availability zone.
Start 100 paralllel processes (each with its own MongoClient). Each process makes 3000 find queries across 3 different collections.
On 2.4.9, I see a throughput of approx 7200 queries / second.

If I switch to the 2.5.4 64bit linux binaries, my throughput drops to around 3500 queries / second.

Performance graphs included. Further info available on request.



 Comments   
Comment by Eliot Horowitz (Inactive) [ 22/Jan/14 ]

2-6 weeks from rc0 to stable depending on what we find...
Wish I could give you a more definitive answer, but it all depends on how many bugs are found.

Comment by John Greenall [ 22/Jan/14 ]

Ah Sorry I just realized that 2.5 is never going to be stable version anyway. I can just continue working with the nightly release in Dev. My concern is when I am likely to see a stable version with the $min and $max operators in. Looks like we're going to get 2.6.0rc on 11/Feb/14. Any idea how long to go from RC to full stable release?
Best,
John

Comment by Eliot Horowitz (Inactive) [ 22/Jan/14 ]

Great.
That's certainly what we're shooting for.
I'm not sure how much money I would put money on Tuesday, but we're trying...

Comment by John Greenall [ 22/Jan/14 ]

Tried nightly build and it actually seems faster than 2.4.9. Getting a throughput a little over 9000 queries / second now.
Am I correct in thinking that this performance improvement will be in the release of 2.5.5 on Tuesday?

Cheers,
John

Comment by Eliot Horowitz (Inactive) [ 22/Jan/14 ]

This is overall known, but spread out through a number of tickets.
Would you mind trying a 2.5 nightly build as there have been lots of changes and our internal numbers are now looking a lot better.
Would love to get some external data points even before 2.5.5 comes out.

Generated at Thu Feb 08 03:28:31 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.