Linus And Alan Settle On A New VM System

stylewagon writes: "ZDNet are reporting that Linus Torvalds and Alan Cox have finally agreed on which Virtual Memory manager to include in future kernel releases. Both have agreed to use the newer VM, written by Andrea Arcangeli, from kernel version 2.4.10 onwards. Read more in the article."

Good news.

[rmadmin]

Regardless of what the article almost implied (that Cox and Linus were at dispute), this is good news for the kernel. From the sounds of it this new VM will make quite a difference from a performance aspect. I could almost care what people are fighting about. As long as new features get implimented, or the system is revamped to improve performance/stability, I'll be happy. And thats what the point is here... A new VM is going to be implimented, and its supposed to kick butt. So enjoy it and quit squabling about weather or not Cox and Linus are fighting!

Details please

[lythari]

how does this new VM manager compare to the old one? I assume it's better, but exactly how does it improve over the old one.

And how does it compare to VM manager in other 'nixs out there, especially FreeBSD.

Media perception of Linux

[Count]

Its so funny how everything that goes on in the Linux community that is reported on, even by "tech-sauvy" news orginizations, has to do with the doom and gloom of Linux. It seems that almost weekly Linux is narrowly "saved" from total destruction. I think the media wants people to believe that Linux is so unstable and that the smallest dissagreement will seperate the whole community and send it into complete chaos.

..."The accord also ends speculation that a fragmented Linux community would be doomed in the face of Windows."...

It seems that news about Linux is not intresting enough unless it is a struggle against Microsoft or has some doomesday issue that could cause it to "fall".

just an observation ... I would just like to see a news article that didn't mention the "unknown future" of Linux

Differences between LINUX and FreeBSD VM design?

[Anonymous Coward]

Can somebody explain the differences between the Andrea Arcangelis and FreeBSD VM desing ?.

I heard a lot of times that the *BSD desing is a lot better than the Linux, is this true ?.

Thanks for comments.

Re:Was there really any "dispute"?

[nyteroot]

yeah, plus the gross inaccuracy saying alan was going to maintain the 2.4 kernel.. didnt they *just* announce that it would be marcelo? oh the joys of clueless linux reporting..

Incorrect memory usage

[Virtex]

Has anybody else noticed that, since 2.4.10, the reported memory usage appears to be wrong? I noticed in the change logs that this was supposedly fixed in 2.4.13-pre1, but I still see the problem. Running "free" shows that I'm using up 245MB of RAM on the "-/+ buffers/cache" line (I'd paste it here, except Slashdot is rejecting the post due to "lameness filter encountered. *sigh*). Now I know I'm not using 245 MB of RAM (after subtracting out the buffers and cache), and I can prove it by running a program which allocates about 350MB of ram then frees it. When I do that, my memory usage, including swap, drops to about 70-80 MB. Is anybody else seeing this?

Re:Summary as I see it...

[hey!]

1. Everyone knew the Rik VM was poor


Even that may be a bit strong. I'd say more that it was late than poor. I'm running a recent ac kernel on one of my production servers (long story, but suffice to say that I needed a bunch of things and this turned out to be the simplest way). I was initially concerned about the Rik VM, but so far it has worked fine. Of course this is anecdotal, but I almost certainly would have had serious problems if I had installed one of the earlier 2.4 kernels, which shows progress.

Rik is right that you make things right before you make them fast; on the other hand the VM really needs to be right AND fast to be truly ready for prime time (and early releases in the 2.4.x series weren't all that "right" to begin with). The Rik VM is supposedly more advanced and featureful, but it may have simply been a bridge too far, at least given the time and resources he had to prove his ideas.

I feel sorry for Rik, but thems the breaks. He may have been right but you don't get forever to prove your ideas. At some point the clock runs out on you, even if you are really close to pulling it off.

The AA VM "beats the pants off" Rik's VM???

[TopherC]

"I took the older VM plus fixes Rik felt would solve all the problems. Linus took Andrea's work. Right now, as of 2.4.14pre, Andrea's VM seems to be beating the pants off Rik's VM. All the current numbers suggest it's the better path."

Hey, I thought slashdot cited [slashdot.org] a comparison [nks.net] of the (fixed) Rik VM and the AA VM, and came to the conclusion that they performed about the same! They were both MUCH better than the 2.2 (old Rik) VM. What's Alan's evidence that the AA VM is "beating the pants off Rik'v VM?" If they really do perform about the same, I would have to side with Alan's original decision to just patch the old VM.

Re:NUMA?!

[psamuels]

But the real question is - if RvR's VM was so broken, how did it get in there in the first place ?

Something had to get there. Let me explain.

Back in 2.3.7, Linus merged a huge change that moved most file I/O from a buffer cache (caching of device blocks) to a page cache (caching based on virtual memory mappings). The VM was severely affected by this, and it never quite recovered.

So the Riel VM was not something wholly new, although it has some bits Rik put in late in the 2.3 cycle. -- But to answer your question, the "new" part of the 2.4 Riel VM was only accepted because the 2.3.7 VM was even worse.

The real question is, why did Linus stop merging Rik's VM patches back in early 2.4? At least according to Rik, Linus's VM between 2.4.5 and 2.4.9 stayed the same even though Rik was still tweaking it and submitting patches.

Re:NUMA?!

[Anonymous Coward]

I mean, how many people have access to NUMA machines, let alone own one?

All multi-processor AMD Hammer machines are / will be NUMA, so we may see a lot of mainstream users getting NUMA machines in the next few years.

That having been said, there are more linux users than just mainstream users .. there are a lot of universities playing with NUMA machines, and some businesses too. Linux is working hard at being all things to everyone, but if it drops niche users whenever it's convenient its credibility will suffer in those niches. You may say "devil may care", but this is more or less what happened with gcc -- a lot of fringe groups got abandoned, and one of the reasons egcs was a success was because it picked up these fringe groups again (GNAT, et al) and made'em all one big happy family. If nothing else, the current maintainers need to avoid ostracizing people if they want to avoid having the mountain moved out from under them, like Cygnus did when they took over gcc maintanence from RMS. (OTOH, if Linus + friends don't care, then this is moot.)