Linux Tuning Repository

Owain Vaughan writes "Mindcraft's excuse for their results when comparing highly tuned NT versus vanilla Linux was that there was no central source of Linux Tuning information. Well there is now. Please submit all the Linux tuning articles you can get hold of to root@vaughan.com"

Comments

Link to some tuning resources.

[Anonymous Coward]

There are some OS tuning resources in the "Optimization and Tuning of the OS" section of Sys.Admin. page of Gary's Encyclopedia".

The original effort

[Anonymous Coward]

Check out the community-based effort at:

http://www.nl.linux.org/linuxperf/

Re:The original effort

[Mithrandir]

We're writing as quickly as we can. There are actually a whole heap of documents there, but the guy that did the main page hasn't linked them in yet. You have to read the CVS update messages to know where to find things :) (I remember that at least there is a sendmail page there)

Rik has just moved the list over to majordomo (message came through all of 5 mikes ago). You can join the list by sending mail to

• majordomo@nl.linux.org

with the body

• subscribe linuxperf

and then come join the fun.

Note that we are taking a completely different approach to tunelinux. We're breaking it down into functional areas rather than specific applications. For example, to tune a, oh.. let's say samba server, you need to tune SAMBA, the kernel and a number of other things to get the best from your box. If you take the tunelinux approach, you'll only end up with a 50% solution because you'll just tune SAMBA, and not the kernel. The Mindcraft report is a classic for that. Yes, tune samba, but hey, the kernel still only uses 960MB of RAM. The "tuning samba" doc approach wouldn't cover that sort of information all in the one spot. Effectively you end up with the same problem as we have now - lots of documentation but so fragmented that you can't do anything useful with it.

How does this relate to linuxperf?

[Mithrandir]

Rik van Riel has already started a linux tuning site over at nl.linux.org. Already there are some 50+ folks on the alias with documents starting to fly around the CVS system. Nobody has mentioned this site to us, so what's the deal?

Re:How does this relate to linuxperf?

[luge]

Yeah. I hate to shoot this guy down, but there are some very serious folks doing some very serious work at this URL. It was mentioned a couple of days ago at linuxtoday. I'll try to find the URL for the actual announcemnt there. Anyway, the gist of is that they are not just doing a repository (which is what this appears to be.) Rather, they are trying to coordinate (through CVS) the writing of actual documentation, not just a collection of tips. This is sorely needed- if you have serious time and writing skills, go help them out. I know I will as soon as exams are over...

resolves to..

[zempf]

For those of you who can't resolve the domain name yet or something, www.tunelinux.com is 193.243.238.236.

-mike kania

Re:Question

[peterb]

TCP Tuning information for many operating systems (including Linux, BSD, and Microsoft's offerings) may be found at the Pittsburgh Supercomputing Center and NLANR Engineering Services

Re:Using root e-mail?

[BeBoxer]

Seems like a bad idea to me too. Aside from the numerous security risks that come from using root as a regular account, it implies that the person running this machine is too lazy to even bother creating a non-super account. Or too ignorant to know any better. Either way, it presents a really bad image.

spreading ourselves thin (part 2)

[Jae]

(lets try this again since I can't type today and keep smacking the wrong keys)

Note : I use "we" as a general pronoun here

What are we really accomplishing by having all these "helper" sites come online - especially when they have little or no content to them?

Wouldn't it make more sense (and maybe this is being done, but it needs to be publicized (sp) a bit better) to have all these ppl who are creating these sites pool their efforts and work on one big site?

Their intentions are definately good - but it's not acheiving the desired goal.

I know if I was a newbie user, I'd want to be able to look one place for all the hints, docs, etc that I needed, rather then having to visit numerous sites to find out one thing.

That's why sunsite was so great - b/c you could find a lot of what you were looking for in one place. But - it's not like that any more (at least I think)

We're trying to pursuade people to move over to linux as their OS of choice - but how easy are we making it for them with these "helper" sites that come on-line, but have no "help" in them.

I think that if everyone invested their time into one unified project, we'd have one helluva documentation/hints site on our hands.

Find a group of people to administer it, find a group to write the scripts (I'd help w/ that part myself), find people for "formalize" all the hints and documents that are submitted, and then put them online.

We're moving forward w/ providing documentation, but we're moving backwards as well.

Note : Again - remember that I use "we" as a general pronoun, and it doesn't necessarily refer to people who read this site - it's more of a linux population as a whole

It seems a bit sparse on content

[Agnomen]

tunelinux seems a bit sparse on content. It seems more like a template for a new site, than a ready to launch site. The links I followed all seemed to point to simple descriptions of the various daemons and architectures. They could have seeded it with at least the Apache Performance notes or the more specific OS performance notes already published on the main Apache site. Not to mention the fact that they could have perused the contents of the many howtos listed at Linuxberg among many other places. It looks like a good starting point though, and I'm sure it will become a useful resource once more content is provided. It is a nice design, but "Content is King."

Re:It seems a bit sparse on content

[maw]

But I guess it's easy to set up a blank site and advertise for people to fill it with content, which is why so many sites start up this way ...

I've been thinking about this (fairly common) phenomenon lately. It's a small part of something larger; lately, we've seen a lot of people making big announcements about open source projects, registering a domain, setting up a flashy web page and a mailing list, etc., but it turns out that practically nothing has been coded: main()'s defined and stdio.h #included.

It sounds like good ol' vaporware, but I think it's something more than just that.

I've come to the tentative conclusion that this is some of the negative backlash from The Cathedral and the Bazaar. People end up with a poor understanding of the document (Slashdot is probably guilty causing a lot of this :/); often they never read the whole paper.

I am not sure which is worse: having read but poorly understood Cathedral and the Bazaar or not having read it at all. I'm wondering if it might be the former.

Think twice before you make an announcement like this.

Re:Slight problem, however.

[dillon_rinker]

I am dissapointed that this site is Live and yet doesn't contain any useful information what so ever.
By publicizing this site in this way, the author is accumulating quite a few pointers to useful information. The author demonstrates some level of commitment and credibility by investing time and $$$ to create this site (though admittedly not a WHOLE lot).

How is this going to co-ordinate itself with the kernel-doc effort being placed by riel@humbolt.nl.linux.org?
Ask the site maintainer - how is anyone else supposed to know the answer to this question? Oh wait - you didn't mean to be helpful, did you? You just wanted to criticize.

Mindcraft/Microsoft will be laughing at this site. And it will fuel their FUD that everything under Linux is Alpha/Beta software in the public domain.
What this author is proposing will not spring into view fully formed. It will take a lot of work and effort by a lot of people because no one has all the answers. The author clearly felt that the best way to grow his site was to go public as soon as he could. With a lot of work and a little luck, it will grow up.

I think it should be taken off until the author knows at least some links to Kernel tuning sites - the author could at least put up some "Under Construction" signs...
Once again, you demonstrate your desire not to help but to criticize. What are those links? Do you know where such information is? Why aren't you sharing it? The site's author is trying to be helpful; I see no evidence of that from you.

I, too, was disappointed at the lack of information on this site, but I could see that it was a work in progress. It takes a real jackass to say that because something isn't finished it isn't worth spending time on.

Re:No Vaporware for Linux please

[GC]

At least the www.nl.linux.org/linuxperf site states that it is still under construction.

After 5 seconds consideration I'm come to the conclusion that www.linuxtuning.com is slow and unprofessional.

www.nl.linux.org/linuxperf has to my knowledge 50 volunteers and has been properly consulted through the Linux mailing lists - which seems the right way to get people who know what they are talking about involved.

There has been a delay with linuxperf because they didn't expect to get so many volunteers and what seemed to be a small project at first looks as if it will require more management than earlier envisaged. Thankfully, however, this will probably make for a higher quality and more complete documentation project over the long term.

Thanks for listening. :)