Seriously, in the past two months, whenever I’ve visited Wikipedia, I feelmlike it’s running on an Atari 520ST connected to the world with a 1200 baud modem. It’s just slow, slow, slow, and eight times out of ten clicking on a link just gets me the following:
Wikipedia is an incredibly busy site; recently we peaked at 6000 page requests per second through the squid farm. Wikipedia has more traffic than the New York Times, the Chicago Tribune, the Los Angeles Times, and the San Francisco Chronicle, combined. The Wikimedia Foundation has over 100 servers running to make all this work.
The following discussion is my understanding based on discussions with other administrators and a couple of developers. Some of this information may be mistaken, although I am trying to be as accurate as possible.
The main reason for the slowness of late has been overloads at the media server. Until just recently, we’ve had a single media server for all of Wikipedia. This server has been increasingly overloaded of late, and has been slowing all sorts of things down. The nonresponsiveness of this server has caused the frontend squid caches to experience thread stalls and other problems, which make the site overall slower. In order to remediate this problem the developers have implemented a media cluster, spreading the load for serving images over several servers. This should hopefully reduce contention for media content and reduce the frequency of squid stalls.
The other underlying problem is more difficult because it cannot be solved by throwing more hardware at it. We’re reaching the technological limits of the MySQL database software to process and replicate transactions (which are caused by edits). We’ve already thrown just about as much hardware at the master server as we can. We need a solution that allows for multimaster replication, but there isn’t a solution in the realm of free software that offers that. Getting past this scaling point is not going to be easy.
I just went and checked it out and it seems to be running a good deal faster than it has for the last week or so. Not perfect, but a definite improvement.
I imagine saving every edited version of every page has something to do with the overall problem. I think that is part of what KellyM was talking about in the earlier post.
There has been a notice on the front page for the last week or so saying they were undergoing an upgrade.
I think the fear is that if donated Oracle is used then the price will be that every page will have to have a “brought to you by Oracle!” on it and that is not acceptable. I think they can’t afford to buy the needed licenses. Also there is a strong preference for open source software by many there.
As for Oracle, well, the software that runs Wikipedia is open source, and distributed to anyone that wants to start a wiki of their own. Most Web hosts offer MySQL databases, but none that I know of offer Oracle. Moving to Oracle would make the software far less useful to others.