[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[EP-tech] Antwort: IRStats2: Innodb tables; forks

Hi Martin,
The changes in https://github.com/eprints/irstats2/pull/84 seem to help us quite a bit. Out daily stats start run at 06:30, and were still running at 12:00.
Implementing this fix yesterday (and with two days of data to process), the stats completed processing just after 09:00.

Some of the tables (details in https://github.com/eprints/irstats2/issues/83) are re-created each time the stats are processed.
These tables may be using InnoDB as the engine isn't specified in the CREATE TABLE statements (if you're running MySQL >= v5.5.5 the default engine is InnoDB).

The structure of these tables means that the InnoDB engine has to do a lot of work (maintaining the index? moving data?) for each insert or transaction. If the inserts are in a few transactions, there's a lot less work to be done (so it's quicker!).

I'll continue to monitor how quickly our stats process - it looks like it helps us, but it might not help everyone.


From: eprints-tech-bounces at ecs.soton.ac.uk [mailto:eprints-tech-bounces at ecs.soton.ac.uk] On Behalf Of martin.braendle at id.uzh.ch
Sent: 16 December 2016 13:49
To: eprints-tech at ecs.soton.ac.uk
Subject: [EP-tech] Antwort: IRStats2: Innodb tables; forks

Hi John,

we migrated our test system to InnoDB about a year ago, did some tuning using the MySQL performance tuning primer script, and carried out some performance tests with the Apache Benchmark tool on both the test and production system.

Result: For sequential accesses, response times with InnoDB were 930-960 ms, with MyISAM they were 700ms. For concurrent accesses, response times dropped to 250ms on average with InnoDB. But you need many concurrent accesses to achieve this number. Also, when Web browsing, the InnoDB test system felt sluggish compared to the MyISAM production system.
IRStats2 weekly update runs take about 10 hours on the MyISAM system, and 17 hours on the InnoDB system, although there are more updates to process on the production system.  (I see that you have added transactions to IRStats2 on GitHub, we will check whether this helps).

To bring InnoDB to fly, you would need a huge innodb_buffer_pool_size, around the total size of your databases (which in our case is around 30-40 GB), requiring a hefty DB server.

Since we were not sure, in as much the current EPrints version supports InnoDB transactions, we decided to stay with MyISAM on the production system for the moment.



[Inactive hide details for John Salter ---13/12/2016 12:59:06---Hi, Our IRStats2 processing runs slowly, and I suspect there mig]John Salter ---13/12/2016 12:59:06---Hi, Our IRStats2 processing runs slowly, and I suspect there might be some improvements to be made -

Von: John Salter <J.Salter at leeds.ac.uk<mailto:J.Salter at leeds.ac.uk>>
An: "'eprints-tech at ecs.soton.ac.uk'" <eprints-tech at ecs.soton.ac.uk<mailto:eprints-tech at ecs.soton.ac.uk>>
Datum: 13/12/2016 12:59
Betreff: [EP-tech] IRStats2: Innodb tables; forks
Gesendet von: eprints-tech-bounces at ecs.soton.ac.uk<mailto:eprints-tech-bounces at ecs.soton.ac.uk>


Our IRStats2 processing runs slowly, and I suspect there might be some improvements to be made - especially as we're using InnoDB tables.
Before I start trying to work out how to solve this, has anyone done any work in this area?

Also, there are two forks of IRStats2 that might need to be merged:

(Most other forks come from the head of eprints/irstats2).

Comparing these two forks:
it looks like there has been some re-arrangement of files (these seem sensible to me), and a couple of fixes.

Does anyone have any thoughts on whether these should be merged, so we have one main master again? Will doing this cause problems for people who have installed the package using the previous file-layout?

John*** Options: http://mailman.ecs.soton.ac.uk/mailman/listinfo/eprints-tech
*** Archive: http://www.eprints.org/tech.php/
*** EPrints community wiki: http://wiki.eprints.org/
*** EPrints developers Forum: http://forum.eprints.org/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ecs.soton.ac.uk/pipermail/eprints-tech/attachments/20161216/ac91c4d5/attachment-0001.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.gif
Type: image/gif
Size: 105 bytes
Desc: image001.gif
Url : http://mailman.ecs.soton.ac.uk/pipermail/eprints-tech/attachments/20161216/ac91c4d5/attachment-0001.gif