![[info]](http://l-stat.livejournal.com/img/userinfo.gif?v=92.1){kind=small}
![[info]](http://l-stat.livejournal.com/img/community.gif?v=92.1){kind=small}
There seems to be quite a few misconceptions about why the servers are so slow and/or why there are so many database errors at the moment. First, the site is not slow because of:
- A full disk. There was a full disk, it's now resolved. We aren't running out of disk space (yet). Tim's new history compression mechanism will help immensely here.
- Lack of servers. Although this has been a problem in the past, and surely will be in the future, it's not the current issue.
- Lack of bandwidth. However the forthcoming gigabit uplink will be nice.
- A crawler using all the resources. Yes, there was one, it's now gone.
What is making the site slow is mainly configuration and software issues. The primitive load balancing in use doesn't make good use of the hardware - some apaches are at 100% CPU while others are nearly idle; the 512MB apaches start swapping, while the 1GB ones are fine, but reducing the number of connections means that the 1GB ones run at half capacity. A proper solution to distribute requests based on load is needed. Yes, it's annoying. Sorry. It's being worked on. If anyone wants to help, please do.
The "deadlock found while trying to get lock" errors are transient - they're annoying, but reloading a few times should fix it. The ideal solution here is that the software should do it for you, but this is somewhat non-trivial to fix and hasn't been done yet.
I have a database error / my new article doesn't show up / history is missing
Wait a while and try again. It will fix itself, eventually. Please don't recreate the same article 50 times.
January 14 2005, 00:04:41 UTC 7 years ago
More servers, separate pipes
The server admins like to focus on the proximal cause of sloth, which is great. But a more general cause is that we *don't* have enough servers or enough alternate bandwidth pipes, which makes us less robust to a sudden onslaught by crawlers, a spike, a key machine going down.It would make sense for us to have twice as many machines as we do now, and to actively prevent any of them from every getting close to 100% load. I hear LJ keep twice as much hardware as it "needs" at all times, for load balancing and hot spares...
+sj+
January 14 2005, 00:11:41 UTC 7 years ago
Re: More servers, separate pipes
Yes, but LJ is a profit-making company and can afford hardware :) While the WMF continues to make money only from donations, we're unlikely to ever have enough money to run the site properly. It's probably time to start looking at grants/hardware donations ASAP - I keep hearing about these but nothing seems to materialise.September 18 2005, 15:58:06 UTC 6 years ago
slow
Two months ago, I had not heard of wiki. Now, not only do I know about wiki, so does everyone else I know. Success is killing wiki, because of the intollerable load times. If amazon had had load times like this, somebody else would be selling me my books today. I'm glad to be reassured that someone is working on the problem, but meanwhile a simple solution would be to require people to log in. The log-in procedure is already in place. It's not much trouble, doesn't cost anything. Anyone seriously interested in wiki would log in. And the people who are using wiki to post dirty words might go away.October 30 2005, 03:57:16 UTC 6 years ago
Re: slow
the problem is that this is very un-wiki in nature. The whole premise of wiki, wikipedia, and the WMF is that anyone can contribute regardless of account status. Accounts have their own privelages but contribution (and especially reading) privelages is not one of them.