I notice that the WP site is very slow. I think WP is a victim of its growth.
Have you thought of cooperating with some service like Coral? Or, you could just build your own version of it.
Dr. Stallman in his proposal for a universal learning resource explained why this scheme is preferred.
Sorry I forgot to include a link to Coral which is: http://www.scs.cs.nyu.edu/coral/
My guess is that you wrote this during last night's little incident.
Wikipedia is almost never quite that slow, we were having some problems where all but one of the database servers were offline for a bit, thus the really sluggishly slowness.
--node
On Fri, 22 Oct 2004 20:23:30 +0300, NSK nsk2@wikinerds.org wrote:
Sorry I forgot to include a link to Coral which is: http://www.scs.cs.nyu.edu/coral/
-- NSK Admin of http://portal.wikinerds.org Project Manager of http://www.nerdypc.org Project Manager of http://www.adapedia.org Project Manager of http://maatworks.wikinerds.org _______________________________________________ Wikipedia-l mailing list Wikipedia-l@Wikimedia.org http://mail.wikipedia.org/mailman/listinfo/wikipedia-l
On Friday 22 October 2004 20:40, Mark Williamson wrote:
Wikipedia is almost never quite that slow, we were having some problems where all but one of the database servers were offline for a bit, thus the really sluggishly slowness.
You are right that WP is not always slow, but I had noticed that lately its performance was depressing. Today it's better but not perfect, it is still much slower than other websites.
In any case, I think the implementation of a distributed scheme is of great importance.
On Fri, 22 Oct 2004 19:59:40 +0300, NSK nsk2@wikinerds.org wrote:
I notice that the WP site is very slow. I think WP is a victim of its growth.
Have you thought of cooperating with some service like Coral? Or, you could just build your own version of it.
Various ways of distributing the server load are indeed always being considered. If you look at the diagram on http://meta.wikimedia.org/wiki/Wikimedia_servers you will see that while not *geographically* distributed, the current system is not exactly what you'd call "flat". :D
I believe further proposed enancements include the purchase of "squid" proxy servers in different countries (I tihnk somebody was in the process of trying this in France), with some kind of geographical server selection. There was also talk of using something called "Super Sparrow", or a similar solution, which is also related to this issue.
The biggest problem, though, and why we can't simply piggy-back a general-use replication network like Coral seems to be, is that by its very nature, Wikipedia content is not static, and it is imperative that we can keep track of what version is being served to whom. In the current setup, the software is able to flush out-of-date pages from the various 'squid' caches when they are edited, so that things like removal of vandalism won't take an arbitrary amount of time to filter through to normal users. Any wider distribution system needs to retain this per-article control.
Plus, a lot of the slow-down is from people using inherently un-distributable features like editing pages, or refreshing their watchlists. It *might* be possible to create some kind of wikiP2Pedia[tm], where the database itself was more widely distributed, but it would be no simple feat of software engineering...
On Friday 22 October 2004 21:43, Rowan Collins wrote:
Your system is impressive, indeed.
general-use replication network like Coral
Perhaps you could cooperate with them to share algorithms/code et cetera.
wikiP2Pedia[tm]
Why not distribute editing, too?
I imagine a system like this: 1. A collection of network nodes acting as catalogs of articles and a list of other nodes serving these articles. 2. A collection network nodes acting as editing servers and storage servers for the articles. 3. Users query the catalog server to find an article. 4. Users can edit the local article stored in any given server. The server reports the change to all known catalog servers. 5. Catalog servers ask all known storage servers whether they wish to update their content. 6. et cetera...
NSK wrote:
I notice that the WP site is very slow. I think WP is a victim of its growth.
According to Alexa we're in the 49th percentile for speed, so that's not that bad. And performance has been steadily improving over the last three years. This is because the average cost of providing each hit has been decreasing with increasing traffic due to better software, especially better caching. Our resources have been growing roughly in proportion to traffic.
Have you thought of cooperating with some service like Coral? Or, you could just build your own version of it.
Various mailing-list inhabitants have espoused the P2P pipe-dream, generally because they failed to consider the crippling problems of synchronisation and development time. There are technical reasons why a central database cluster is convenient, if you want to know about them I suggest you read the previous thread on this subject. There are various viable schemes for improving geographical distribution, which we have discussed in the past. However geographical distribution of hardware remains a touchy political subject. There's currently some Celerons in France which will be turned into squids as soon as someone takes them to a colocation centre and plugs them in.
Dr. Stallman in his proposal for a universal learning resource explained why this scheme is preferred.
According to a credible source, Dr. Stallman is our love slave.
-- Tim Starling
On Saturday 23 October 2004 19:05, Tim Starling wrote:
they failed to consider the crippling problems of synchronisation and development time.
Syncing is indeed difficult.
wikipedia-l@lists.wikimedia.org