If IBM Is Serious About Linux, What Do WE Want?

bfree asks: "Robert LeBlanc, Vice President, Software Strategy, Software Solutions Division says both that IBM would open source any part of AIX and that we would be better off taking bits and pieces and the expertise that IBM bring along with it. IBM's AIX Web site lauds Linux compatibility of AIX and the new AIXL only just slightly behind their statements such as 'A robust, scalable UNIX platform for critical applications.' It's clear IBM wants to be involved with Linux, and I feel that we should want that also. What should we ask them to do for us in return for their involvement? Networking scalability and redundancy, optimization and facilities for database systems (as the jfs has started) or systems management applications? It seems to me we have the offer on the plate from IBM to create a new joined project to bring Linux up another level if we can find a way from AIX. Surely we must take them up on this?"

Comments

IBM has BIG plans for Linux

[Anonymous Coward]

Disclaimer: I'm an IBMer.

IBM has big plans for Linux, as I'm sure everyone knows. There is a Linux Technology Center in Austin, Texas. Many good people down there doing nothing but working on Linux. IBM also has many open source projects, most under the IBM Public License, at the developerworks open source site. And I'm sure there are many other things IBM is doing that I don't know about (I'm just an employee).

However, IBM is not a complete friend of Open Source, just yet. It took me over 4 months of constant effort to get approval to attempt to contribute to the Linux kernel, and even then I only have approval to modify 4 files!!! Let's not forget that IBM is, by *far*, the world leader in patents. There are many parts of IBM which use Linux and open source, but are very hesitant to contribute to open source.

That said, there are many good people in IBM who really do want to do good things. It's just not that easy to get paid by IBM to do them. But it is possible.

So, back to the question of what Linux can do for IBM. I know exactly what the community can do for IBM. Accept us. I have received a lot of resistance to any help I offer on mailing lists. Patches and comments are ignored. My initial contact with a certain kernel developement mailing list was met with...quite a bit of hesitation, to say the least. I think a lot of kernel developers are very skeptical of IBM employees. Maybe they think we're only sending patches or commenting/discussing so we can sell some product, and in some cases, they may be right. But please, don't assume someone should not be trusted just beacuse they have @us.ibm.com in their sig. Just because I get paid by IBM doesn't mean I don't want to help.

A better question:

[rjh]

First, I think the Free Software community needs to do some serious thinking about what technologies IBM can bring to the table. However, I'm not going to try and answer that question, because I think there's another question that needs to be asked in almost the same breath.

If IBM is willing to cater to our needs, then how can we reciprocate?

The foundation of free software is sharing resources, ideas and solutions--helping the other guy get his job done--building a community out of relationships, not between programs but between people.

If IBM is serious about joining the community (and if any IBMers are reading this, let me say that I really hope you are), then we need to reciprocate their seriousness with seriousness of our own.

We need to seriously ask, what can we give IBM?

I think the more IBM sees a willingness from us to work with them, the more willing IBM will be to work with us.

Respect

[heikkile]

I think we, the Linux community, have clearly proved that we do not need anything from IBM. They can throw developers at Linux, so can we. They can write databases, and device drivers, so can we. There are two things IBM has that we do not: Lots of experience in the computer business, and a respectable marketing organization.

In this light, I would only ask this of IBM: Sell a hundred million copies of Linux, especially to the big businesses. Show the commercial world that there are good alternatives to Windows. Do what ever it takes do this. Adding software will be a small part of the project, which you may not even have to write yourself, if you discuss your needs with us. Build your own distribution, sell commercial software on the side, build a world-wide support network, print manuals, what ever it takes. Make sure that when Linux reaches total world domination, a good part of it is IBM Linux.

Important [read: tough] things IBM could help with

[sanemind]

Linux is absolutely amazing for typical workstation class systems, but the one place that it seriously lacks in comparision to the other serious [and propriatary] unix'es [unices?] is in support for really big systems.

What, you say? Isn't linux used on many supercomputer class systems, and heck, IBM even ported it to the S390! Well, yes... and no.

On clusters, linux is an excellent choice, due to it's efficiency and perhaps especially due to it's lack of licensing costs/hassels. But the key thing with clusters is that they are just that, clusters of workstation class machines. The `big'ness of them comes from the parallelism of many small machines working together, which it turns out is quite usefull for certain classes of a problems.

The S390 linux port is the same sort of thing. A massive big iron system, running hundreds of different simultanious copies of linux, each in it's own virtualized address space. Linux couldn't manage the resources and power of the entire machine, but as a 'process' in it, it does just fine; and this too is usefull, for say, consolidating what would otherwise have required rooms of rackmounts, with their own upkeep hassle, into one box [plus, cluster type applications can take advantage of the much faster [then ethernet] internal system bus for message passing, thus achieving comparatively much better performance then `real' clusters.

So, to make a long story short, what does linux really need? Big iron stuff: quality NUMA integrated into the memory management code [as a compile-time option, of course]. True support for high order SMP [say, 64 or 256 processors] and all that that requires [such as distributed kernel threads across multiple CPU's].

One of the problems, of course, is that adding this extra architecture into the kernel would slow it down on good old fashioned workstations, which is where linux is [currently] almost exclusively used [although that S390 thing rocks!]. ...Now, if only this could all somehow be a compile time option, that would be spectacular. The problem, though, is that we are talking about significant differences in basic architecture and of what the kernel has to worry about, these are significant design decisions. I don't think it's as easy as putting a few #ifdefs in the source.


---
man sig