Talk:Wikimedia servers/archive1

From Meta, a Wikimedia project coordination wiki
Jump to navigation Jump to search

DNS problem[edit]

Not a DNS problem actually, just showing the wrong error message. This was resolved a week ago. --Brion VIBBER 02:06, 20 Feb 2004 (UTC)

Server names[edit]

Is our server was named from en:Pliny the Younger or en:Pliny the Elder. Yann 21:17, 24 Feb 2004 (UTC)

The machine has been through a couple generations of upgrades, so I'd say both. ;) --Brion VIBBER 03:33, 25 Feb 2004 (UTC)

Plan for expected growth?[edit]

In the recent announcements, there were growth figures, including the prediction that Wikipedia will reach 1 million articles on September 29th. Is there a plan to keep the hardware in pace with the growth? What is the capacity of the current hardware in Views per day/Edits per day/Total Articles? Nroose 01:57, 18 Mar 2004 (UTC)

There's a fair amount of money left in the Wikimedia foundation account, and the architecture is now really scalable, so it should be mostly a matter of buying new Apache and Squid servers as needed. The one place where there is currently a bottleneck is on the database server, but it is currently fairly lightly loaded. This bottleneck can also be made more scalable later by any of a number of strategies, including:

  • upgrading to a fast multiprocessor machine with lots of disks and RAM (in the short term): RAM is 10,000 times faster than disk...
  • using database clustering and replication techniques (in the longer term)

-- Anon.

I hate to be alarmist[edit]

But the last few days have been grim for server performance with "database server unavailable" coming up all the time. Is someone planning a long way ahead? I know it takes money and all that? Given how much it must be affecting users we should have some fairly upfront explanation in place --AndrewCates 13:54, 20 May 2004 (UTC)(talk)

Database server unavailable messages are normally a sign that the Apache web server machines have hit the connection limit on the database server (usually because it's temporarily overloaded). This preserves reasonable performance for the majority of requests. Without it, the total number of connections would just build up and every request would be slow. If you do get the error, a 30 second wait and a reload will usually succeed. This isn't ideal. The new database server should largely prevent this from happening. Being phased in now, starting with reads to prove that it's reliable under load. It'll probably be at least a month of that before it's trusted enough to become primary. The rest of the Hardware order May 2004 equipment is in use now and should be helping with this problem - one more Squid cache server and two more web servers, plus one not in use because of CPU overheating - I think that one has been sent back to the vendor to be fixed. Jamesday 04:01, 22 Jun 2004 (UTC)

RAID 10?[edit]

Would it be a good idea to move suda over to RAID 10 to increase performance?

Which kernel version and GNU/Linux distribution are installed on your servers?

210.49.113.112 12:51, 14 Jun 2004 (UTC)

Yes, RAID 10 on Suda would probably be good and I'd like to see another 146GB drive to make four and allow that. It's not likely to happen until the new database server is in full use. We've already had too much unavailable time in the last month and it would take the whole site down to change, probably (the controller might be clever enough to switch from RAID 5 to RAID 10 without losing anything while still running but I don't know if it can). The ganglia cluster stats have options which show the operating system name and version. I see that's been added to the server table here as well. Jamesday 04:01, 22 Jun 2004 (UTC)

A drive for this has been ordered. The controller can't change from RAID 5 to RAID 10 without discarding the data but that doesn't matter now that Ariel has been in full use as a database server for a month. Jamesday 21:50, 20 Aug 2004 (UTC)
The drive was installed on 10 September and is now in use on Suda. Jamesday 17:22, 13 Sep 2004 (UTC)

Names suggestions used so far[edit]

what about naming one of the servers after Diderot?

done on 20 August 2004 . Jamesday 21:54, 20 Aug 2004 (UTC)

The next server we get should be named Albert, in honor of Saint Albert, "a Dominican friar who became famous for his universal knowledge". Raul654 03:13, 10 Jun 2004 (UTC)

Albert was used today. Jamesday 20:50, 10 Sep 2004 (UTC)

What about "Bacon"? In honor of en:Francis Bacon is an adequate name for an encyclopedia server. Bacon was an important precursor of the historical Enlightment Period and an outstanding scientist in his historical period. --80.130.28.26 22:13, 20 Aug 2004 (UTC)

Bacon was used today. Jamesday 20:50, 10 Sep 2004 (UTC)

Name suggestions[edit]

Four names are wanted as soon as possible, two more (one emphasising vast storage and one easy availability of information) are desired. All MUST be historical encyclopedists, non-English are preferred. See Wikimedia servers for the current list. Jamesday 21:54, 20 Aug 2004 (UTC)

I propose en:Averroes, an Andalusi philosopher and physician, a master of philosophy and Islamic law, mathematics and medicine. He was born in Cordoba, Spain, and died in Marrakesh, Morocco. Med 22:07, 20 Aug 2004 (UTC)

There is an german etymological dictionary called Kluge, dont know who that is/was, but its very ambiguous becouse klug is "clever, wise". --141.53.194.251 22:12, 20 Aug 2004 (UTC) The first name seems to be Friedrich

on de:Geschichte_und_Entwicklung_der_Enzyklopädie you'll find many more proposals for naming issues :)


Please consider en:Hypatia of Alexandria, who, besides being an accomplished scholar, was also known as an adept compiler and transcriber of knowledge. -C.Lovett

Red Hat/Fedora and security[edit]

Hi,

I note that the servers are currently running a mix of RH/Fedora servers. Red Hat 9 has dropped out of security support period, so this is worrying. Fedora Core 1 is about to do the same.

Could you guys not get a support contract from RH for free (good publicity for them, etc). Failing that, maybe Debian Stable would be a better distro to use as it's supported for longer.

RH9 has several unpatched exploits available. Be careful.


RH9/Fedora[edit]

I was just wondering a simmilar thing; why does Wikipedia use RedHat/Fedora? I don't mean this to sound so incendiary, but what do you get out of paying for the distribution? I haven't used Fedora, and I haven't touched Redhat since 8.x, so I can't say for sure, but last I knew, administering a RH server wasn't as easy as an equivalent Debian system.

We don't pay for either Red Hat or Fedora Core; they're both free (gratis) downloads. Raul654 07:13, 17 Oct 2004 (UTC)
Mainly because it's what comes with the servers we buy, it supports both i386 and x86-64, everyone knows how to use it, it's well supported, and we're too apathetic to change it. We (well, I) have just installed SUSE 9.1 on one server to see how it goes (personally I don't really like Fedora...). And yes, we don't pay for it, except for the cost of writing a CD. —Kate | Talk 10:06, 2004 Oct 28 (UTC)

Has anyone tried Fedora Core 3? It's supposed to have better performance than 2 (so I have heard). exeryone is too scared for Gentoo then? --Haggis 13:50, 17 Dec 2004 (UTC)

FC3 is one at least one of the new database servers, perhaps also on the 5 new Apache/memcached/squid machines. Jamesday 20:49, 25 Dec 2004 (UTC)

Apache and Apaches[edit]

The section about 'Apache Servers' is hard to read and misleading. Why do you use the term 'Apache' both for the web serving software and the machines? Why not name the machine after their function?

Name suggestions[edit]

If you would like to suggest a name, please enter it and link to the relevant Wikipedia article. This list is no longer being used to name servers.

Encyclopediasts or similiar (with articles)[edit]

  • Aristotle - Aristotle, wrote so many varied works that together they comprise a kind of encyclopedia of all Classical Greek knowledge (good name for a really important database server).
  • Mills - John Mills
  • Franklin - Ben Franklin (author of Poor Richard's Almanac). (or/and w:en:Rosalind Franklin ?)
  • panckoucke - see w:en:Encyclopédie Méthodique or w:de:Charles Joseph Panckoucke (editor of the « Encyclopédie Méthodique »)
  • ashurbanipal - w:en:Ashurbanipal IV (maybe not an encyclopedist per se, but the first known to try and bring together all knowledge. Sooner or later there must be a server function that such an ancient name would be appropriate for.)
  • ephraim or chambers - w:de:Ephraim Chambers/en:Ephraim Chambers
  • johann - for w:de:Johann Georg Krünitz, w:de:Johann Samuel Ersch (en),w:de:Johann Gottfried Gruber (en), w:de:Johann Jacob Leu
  • Judah - Judah haNasi was the editor of the Mishna, a compendium of Jewish religious law, on which the Talmud and the Shulkhan Arukh were based.
  • Mortimer - Mortimer Adler - Author/director of planning for Britannica
  • Asimov - Isaac Asimov
  • Johnson - Samuel Johnson - wrote A Dictionary of the English Language
  • Insert name - "Beautiful it may be, but Dale Hoiberg, the editor of Encyclopedia Britannica, thinks there is such a thing as too much information. Wikipedia reminds him of a Jorge Borges story in which the cartographers of an empire make a map so large it's as big as the empire itself and rather useless." Find the name of the fictional cartographer it's a guild of cartographers
  • McHenry - Robert McHenry, former Editor-in-Chief of the Encyclopædia Britannica and current Wikipedia critic (good name for a machine that performs some obscure menial task).
  • amane - 西 周 (NISHI, Amane), wrote 百學連環 (Hyakugaku-Renkan), the oldest encyclopedia in Japan.
  • Maimonides - Maimonides was a Jewish rabbi, physician, and philosopher who is best known for codifiying the Jewish religious law in 14 volumes known as the Mishneh Torah.
  • Seldon - w:Hari Seldon, fictional encyclopediast from w:Issac Asimov's famous "Foundation" serious. It's his idea to produce and disseminate w:Encyclopedia Galactica, to contain and preserve all human knowledge, in a long dark age after the Galactic Empire's collapse.
  • Jorge - w:Jorge Borges, the argentinian writer, author of "The Library of Babel".
  • Duden - w:Konrad Duden. German philologist. Duden's dictionary was the officially recognized standard of German spelling through much of the 20th century.
  • Roget - w:Peter Roget, creator of the first thesaurus of the English language.
  • Grimm - w:Brothers Grimm, German professors who were best known for publishing collections of authentic folk tales and fairy tales, and for their work in linguistics
  • Rousseau - w:Jean-Jacques Rousseau, great Swiss encyclopedist (he wrote the music pages in Diderot's encyclopedy), philosopher and writer
  • Capella - w:Martianus Capella, author of what is considered the first "encyclopedia" of ancient times.

Others (with articles)[edit]

People without articles[edit]

  • Al-Munjid (almunjid?) is a famous old arabic encyclopedia, mentioned by Khalid
  • chagas - Manuel Pinheiro Chagas
  • mellado - Francisco de Paula Mellado
  • wassili - Wassili Nikititsch Tatischtschew
  • pavel - Pavel Bujnák
  • jacob & carl - Jacob Johann Ankarström & Carl Christophersson Gjörwell
  • banerjea - K. M. Banerjea
  • francis - Francis Lieber
  • Black - Adam & Charles Black - Founders of Britannica
  • kiyonori - 小中村 清矩 (KONAKAMURA, Kiyonori), edited 古事類苑 (Koji-Ruien), a Japanese encyclopedia consisting of 1,000 volumes, created by the government during the Meiji Period.
  • fr:utilisateur:Treanna - a regretted great french wikipedian and historian

Ancient Library originated names[edit]

May I Suggest that if names are needed, Ancient Libraries be honored? I think this keeps with the theme. Toastysoul 10:14, 3 October 2006 (UTC)


Early computers & SciFi names.[edit]

Enigma, Leo, Univac, Multivac are a few that come to mind. Napier deserves a mention as well.

HAL9000

Where are browne and coronelli?[edit]

browne and coronelli are listed as Squids on the page, but they are not listed in [1].

Browne is not sending ganglia data for an unknown reason. Coronelli is down. —Kate | Talk 14:08, 2004 Dec 11 (UTC)

Why usage statistic doesn't work?[edit]

Statistic for wikipedia on this page deosn't work from october. Why?

Because it places too much load on the servers. They will be back at some point when a better way to do it. —Kate | Talk 14:08, 2004 Dec 11 (UTC)
How about copying the raw files to another computer and running webalizer against them to generate the stats? Any news on the progress of getting fresh stats would be appreciated.

Content page edits[edit]

User:Kate what are you doing?--213.66.119.190 22:13, 5 Jan 2005 (UTC)

There was a few changes up to [2] that User:Kate reverted and then made some kind of change, what's going on User:Kate?--213.66.119.190 11:22, 7 Jan 2005 (UTC)

Since nobody else would, I'll revert to the mentioned edit and let Kate respond.--213.66.119.190 11:34, 7 Jan 2005 (UTC)

I think she updated the name links. They were just reedited.--213.66.119.190 22:16, 14 Jan 2005 (UTC)

Servers in *.fr[edit]

Is there an estimate when the squids in Europe will start helping wikipedia? Is there a good possibility to balance the load between EU/US-squids through round-robin-dns? Would be great to see some more servers helping to cope with the load :-) -- Arno 84.128.35.174 12:06, 6 Jan 2005 (UTC)

They are currently active and serving en: and fr: for users in most of Europe excluding the UK and Ireland. See statistics. Kate.
They seem to be serving the UK as well now. -- Anon, 17 Feb 2005.

discrepentcy: Is "ariel" the master database server or just the standby?[edit]

In the table it says "ariel" is the standby and "suda" is the master database server. In the thrid bullet point below it states that "ariel" is the master and "suda" is the standby. Which is correct?

For a while, Ariel was the database master. Recently, Suda was elevated to master status, because of its more powerful IO array.

However Ariel is now the master again, temporarily (explanation). Suda is still the preferred master ATM because although ariel is faster, it has less disk space. Kate

Can I ask if this is still valid after all the fun around the 22nd of Feb? When I was lurking on IRC it seemed that you may have switched again. Thanks. --Haggis 11:11, 23 Feb 2005 (UTC)

ariel is still the master. after some compression and rebuilding of ariel's innodb files, the disk space issue is mostly solved for the moment. after the new hardware order arrives and is burned in, we'll probably change to using that for the new master. kate.

So how do you rate Squid?[edit]

I know you folks are working hard and that you have almost unparalleled traffic growth to contend with, but from the user's vantage point I haven't seen a whole lot of site speed improvement in the English Wikipedia since the installation of Squid servers. Some, but not much. Is it a case of solid hardware/software/networking improvements being offset by crazy traffic increase, or have the improvements (particularly Squid) proved disappointing? JDG 05:57, 18 Feb 2005 (UTC)

Before we used Squid we had only two servers, one of which was sorely underpowered, and believe me it was a lot worse. :)
We set up squid well over a year ago; roughly 75% of HTTP requests are cache hits, leaving only 25% for the Apache+PHP cluster to deal with. In the last twelve months our traffic in average hits/sec has increased eightfold; the cache hit ratio has stayed about the same, so in raw numbers cache misses requiring service have also increased eightfold. (That's a doubling every four months.) In the same time we've gone from about 8 servers to about 40, which as you can see doesn't quite match the eightfold increase though it's hard to compare a number like that.
See the Squid stats since the end of February last year: http://wikimedia.org/stats/live/ and our bandwidth usage in the Florida cluster: http://65.59.189.201/www.bomis-total/www.bomis-total.html
Squid's been wonderful: without it we'd have been dead in the water long since (or had to do a lot of redundant work on alternate caching systems to reduce the cache hit service time to be equivalent to what Squid would have given us). Having a front end cache keeps cache hit returns fast, helps free up the apaches' queues for real work, and vastly reduces the impact of spiders and 'flash-mob' hits. --brion 06:21, 18 Feb 2005 (UTC)
Ok, I guess you would have been dead in the water. It's not easy to grasp but I suppose when you reach a certain traffic level you're doing well just to stay even. Google and Yahoo are the undisputed champs for fast response under super heavy usage, but apparently they manage it mostly through brute force: dozens and dozens of regional server farms... I was going to suggest more resources go into load balancing, which I think is more key than caching. But I'm not in much of a position to pontificate. My admin experience was with busy webservers, but way back in `00. I had great results breaking up the load between 6 boxes using simple DNS round robin, but my average concurrent users were at most 1/4 your average... Oh well keep up the good work and I hope the latest fundraiser allows you to actually gain on the problem rather than just keeping pace. JDG 06:41, 20 Feb 2005 (UTC)
Our current load balancing is pretty awful, although it has been getting better recently... there are so many things needing work that it tends to be quite low down on the list of priorities, though. Maybe some day... -Kate. (incidentally, i'm not sure it's more important than caching - if you removed squid, all apaches would be at 100% CPU no matter how you distributed the load :)

Can we please move that list?[edit]

The list of servers has become ridiculously lengthy and should be moved to a page of its own. When I did that, however, I was quickly reverted. As the explanation given for that revert seems not quite satisfactory, I'd like to suggest to re-revert (unless I am missing something and there is a sound reason for Alterego's action). Kosebamse 18:56, 10 August 2005 (UTC)

Questions about statistics[edit]

I have some questions about the statistics at http://www2.knams.wikimedia.org/stats:

  • In what timeframe and how often is the data collected?
  • Is it possible to get the data for pages that are not part of the 100 most visited pages?
  • Why are the numbers in the count column of the referer and the user agent stistics very low compared to the URL statistic?

Thank you for the answers. -- 84.167.85.10 20:42, 12 August 2005 (UTC)

Ambiguous albert :)[edit]

The standalone list says SuSE Linux 9.1, whereas the albert entry @ meta-wiki says Fedora Core. What is true, then? ;) 80.129.108.48 12:04, 20 August 2005 (UTC) -andy

this one. i fixed the other one. kate

Is there any particular reason[edit]

that two of the database servers have public ips? Plugwash 23:32, 22 August 2005 (UTC) likewise it seems even more odd that one of the squids has a private ip, aren't the squids the systems that actually take the requests from the internet. Plugwash 23:35, 22 August 2005 (UTC)

Bad link[edit]

http://wikimedia.org/stats/live/ is 404 for me right now. -- Beland 01:30, 23 August 2005 (UTC)

Moving overview to the top[edit]

I guess that the overall system architecture is more interesting to most people than the detailed (long) listing of every single server. What do you think about swapping their positions in the article? --84.56.228.90 12:55, 23 August 2005 (UTC)

connection refused[edit]

"If you are getting a "connection refused" error, that is a squid problem. Determine which IP address you are trying to connect to...". Please could you add either a simple explanation of how to do this or a link to a simple explanation of how to do this (preferably an external link, as internal links will give "connection refused" messages). I had this problem last night, and after a bit of searching on google found a reverse DNS, which gave me a list of ~20 IP addresses, all of which result in "this wiki doesn't exist" pages. I don't have IRC set up on this machine, and cannot access IRC at work when the problem happens there. Thryduulf 07:33, 13 September 2005 (UTC)

Actual?[edit]

Who can insert the Talk:Hardware ordered August 30, 2005? Thanks.80.185.50.73

Why no advertising? I think a modest amount of targeted advertising would be ok, as long as it only took up a very small amount of space. The money collected should be more than enough to pay the costs of running Wikimedia.org.

Paid accounts could be offered that are free of advertising.

Now actual?[edit]

What is with the Hardware ordered September 14, 2005? 80.185.38.93 19:11, 16 November 2005 (UTC) Status: instaled problem is the new image server ordered november 05.

Amane CPU's[edit]

What CPU's does Amane has? According to Intel there's no 3.4 GHZ Xeon MP and no 3.4 GHZ Dual-Core Xeon, only normal ones with HT. Is that system a multiprocessor system without multiprocessor CPU's? --212.204.66.66 10:24, 24 November 2005 (UTC)

Statistics[edit]

I would like some statistics like cpu-, disk-, network-, ram-usage. This could be monitored with mrtg and snmp for example. what about this idea to add a monitoring suite to each server? -- 24.18.207.109 17:44, 25 December 2005 (UTC)

Just have a look at ganglia, it will probably tell you about everything you want to know, if you click yourself through to single server level. There hast been some talk on the tech-mailinglist about switching to another Monitoring System (Nagios, I believe), but untill then Ganglia is what you get :-). Happy Christmas, as I am European and I can call it anyway I want ;-), --Mdangers 22:27, 25 December 2005 (UTC)

Updating?[edit]

Is it that this page hasn't been able to be updated, or do we really have over 20 servers that were ordered in Nov and haven't been installed yet? I'm really not asking this from a compliaint standpoint, which we have all too much of around here, but from a what needs to be done standpoint. Is it a lack of manpower, and if so is it because we can't find people we can trust or does there just need to be more done to attract people that can help. In the same vein, there's nothing new in Wikimedia servers/hardware orders. Is it already known what is needed to be bought next or is that just delayed until there's time to install the current hardware? Anyway thanks for your time and efforts. I don't have the technical admin skills to help, but maybe I can contribute in other ways that are helpful. - Taxman 20:05, 25 January 2006 (UTC)

I concur - Ravedave

How many run the GNU/Linux operating system?[edit]

The article states "all but two of them run the GNU/Linux operating system".
But the table lists isidore as running "FreeBSD5.3-REL" and zedler as running "Solaris 10".
Something's wrong. What? --David Edgar 17:53, 3 February 2006 (UTC)

This page is wrong and should be phased out in favor of our offsite cluster management documentation at https://wikitech.leuksman.com/. isidore is running Linux these days. --brion 19:12, 3 February 2006 (UTC)

I have a Network Forensic System to Donate[edit]

Solera Networks

Where do I send it and to whom? Please name the server "Sequoyah" after the inventor of the Cherokee Alphabet.

Jeff V. Merk

Kennisnet clusters[edit]

Hello, there are 13 Kennisnet clusters, not 12. --80.136.226.130 06:35, 24 April 2007 (UTC)

actually there's 1 Kennisnet cluster, with (by my count) 28 servers. this page is hopelessly out of date. Kate

I'm missing Yarrow in the list. --62.154.250.10 07:12, 11 October 2007 (UTC)

Why don't use blade server[edit]

Why don't use blade server for Wikimedia Projects? It occupy smaller rack space and easier to maintaince.--Ellery 06:18, 16 May 2007 (UTC)

Server diagram[edit]

Wikimedia servers

Some time ago de:Benutzer:Kolossos challenged us at de:WP:Grafikwerkstatt to draw a new server diagram "if someone is bored". I'm not admitting anything, but here is a draft. I am a bit weary to do something like this without being involved in the matter, especially since there is so little documentation. So, please comment and tell me what all I got completely wrong. --Hk kng 21:44, 11 November 2008 (UTC)

Nice work! A small note about search: theres is no lucene and conventional search, it's both mw:Extension:lucene-search, but some are version 2.0, and others 2.1. --Rainman 09:15, 12 November 2008 (UTC)
So what are the machines marked "LS2 indexer" (srv56) and "LS2 test" (srv77/79/80)? --Hk kng 14:02, 12 November 2008 (UTC)
All hosts that begin with search (e.g. searchidx1,search1,search2...) are lucene-search 2.1, and all others are lucene-search 2.0, with exception of srv77/79/80 which are currently unused (they used to host ls2.wikimedia.org which is a test environment). --Rainman 14:33, 12 November 2008 (UTC)
So is srv56 also part of this test enviroment, or are there two indexers working? If so, should I depicture them as sharing the load, or are they adressing different machines? --Hk kng 18:45, 12 November 2008 (UTC)
They index different wikis, so they are sharing the load ATM. But eventually srv56 will become obsolete. --Rainman 18:17, 14 November 2008 (UTC)

I think it looks good enought to change it against the old image from 2006. It's a wiki and it's an editable SVG, so it is not so that a small bug would be a real big problem. --Kolossos 13:51, 14 November 2008 (UTC)

403 on Conf files[edit]

http://noc.wikimedia.org/conf/ is linking to pages that return a 403. For instance, the httpd.conf and PHP ini files are forbidden.