Talk:Wikimedia cache strategy evolution during 2003
Wikipedia has become consistently very slow due to overwhelming traffic. Attention should be paid to upgrading the proccessing power and available bandwidth of Wikipedia, improving software performance, and increasing cacheability. Is there a distributed database architecture? Would this be a reasonable and workable solution? [anon]
Wikipedia Implementation and performance
During the last few weeks I have been thrilled to discover Wikipedia. However, it is not perfect, and there appear to be very significant performance issues with the software/hardware.
Is there information about how WP is implemented, and is there any form of discussion going on about how to improve it?
What sort of loading does WP server out - 10 page requests/second, 100?, 1000?
Also, what proportion of the load is due to edits - it should be quite low - but maybe it's not. Is it possible to prioritise edits over searches? This would be useful, as quite often these days waiting for an edit to Load takes so long - it really is quite a nuisance.
Would performance be enhanced if WP were distributed over several servers? Is this feasible? If not, why not?
Who is looking at these issues? --en:User:David Martland
- Wikipedia is open source software. See the SourceForge page for details. See en:Wikipedia:Statistics for detailed access statistics.
- The performance issues are discussed on the wikitech-l mailing list. The server is reasonably huge and not suffering from the load. The main problem appear to be locking issues (we use MySQL's MyISAM tables, which only support table-level locking) and unoptimized queries (some columns do not have indexes where they should have etc.). These issues are slowly being worked on by Wikipedia developers, feel free to participate. See How to become a Wikipedia hacker for a growing tutorial on the code. --Eloquence 12:22 Dec 3, 2002 (UTC)
- Regarding the number of page views per second: On Special:Statistics. In the past 4 1/2 months (since 20 July) there have been about 20,000,000 page views and 400,000 to 450,000 edits. This is equal to slightly under 2 page views per second. As we are still growing dayly, the number will be higher at the moment, put it will probably be well under 10 per second. Andre Engels 14:29 Dec 3, 2002 (UTC)
Thanks for the quick answers. I suppose it's possible the poor performance I get sometimes is due to network problems - besides the locking problems. Another possibility is that a mean rate of 2 queries/second still permits a much higher peak rate. It quite often takes a minute or so to have an update during editing from my locations in the UK - maybe that's the locking problem? David Martland 20:19 Dec 4, 2002 (UTC)
- Minor version mixmatch between files, sorry. Should be fixed now. --Brion 00:47 Dec 4, 2002 (UTC)
In the phrase,
"Suda has 2GB available for caching, with another megabyte unusable",
should the word "megabyte" be "gigabyte"?