Older blog entries for Zaitcev (starting at number 121)

So, the United Linux thing.

One well known community leader said (IIRC): It'll be curious to see Andrea and Rik to work on the United kernel 8-)

I had an opportunity to poke my curious nose into the BSD universe. I prepared an RPM of a program called mptable, which prints APIC related control blocks on x86. My collegue Doug Ledford added a printout of other control block, called $PIRQ, and I had a fancy to contribute that code. Doug put it under BSD license. So, I went to freebsd.org and read the guidelines for contributors. Instead of posting patches to a list They request all patches to be filed as bugs (called "PR" in their bug system), and so I did. I am intrigued, what happens next. I would not be surprised if all such bugs get ignored until patches rot and cease to apply, then someone closes them, by the SOP. I observed that Dillon simply posts his changes to freebsd-hackers, but hey... I am not Dillon, I cannot just barge into their social fabric :)

I have no idea why alan got a second account. I think it was an experiment with case sensitivity of Advogato accounts.

I do not have stats, but I suspect that Advogato readership is growing still, because I get requests to update diary more often. I am not sure it would be a great idea. Side by side with raph's profound entries my ones seem like a contribution to poor S/N. For instance, do not expect any explantions about Red Hat patent policy from me (Mark Webbink has this job :) Now, Infiniband may be interesting to the community, so I'll post updates about it now and then.

If I think of something groundbreaking, I'll let the world know, not to worry.

Mental note: create a vanity domain.

I KNOW that Intel cancelled their Infiniband hardware. Now, please stop e-mailing me about it :)

I went for a stroll in the wilderness and saw two small mountain lion cubs crossing my path. Scared living shit out of myself, because obviously their mom was nearby. Also, I the climb was steep and I left my rifle locked in the trunk to ease my burden.

On Sunday I went to Watsonville airshow and was glad to find the "big small show" still hanging there. The attendance is a little thinner this year, but it's not a total bust as might been expected. The Salinas was simply cancelled last year.

Some small observations from the show.

• I was told the liability insurance is 50% of an airplane price these days, and it's easy to believe when I see Husky AH-1B selling for $186K sticker. I did not dare to check how much they asked for a Cirrus.
• Escape Helicopters moved from Frisco to Sacto area, because VFR flying over Frisco is now prohibited.
• Announcer still used "Confederate Air Force". I cannot believe he did not know how CAF wimpered out, so I think he pretented to forget about the renaming to remind them about their sins.

Infiniband

The Infiniband project moves forward on both the Toy Stack and the Intel's SourceForge fronts. Yesterday I made a milestone release for the Toy Stack, so while people poke about it I am turning to Intel's stack for a while. The objective is to prepare it for a possible inclusion into the kernel, the only little problem is that the stack is 200K lines, is an NT port, and contains a lot of such code which I would NEVER accept if I was Linus. Intel people do not seem inclined to see what is wrong with it, but instead they think that replacing WORD StudlyCaps with __u32 studly_caps ought to do the trick of smuggling it past Linus.

Yesterday I received my first hardware from Intel. It looks very cute. Comparing it with a picture of the IBM 4x board from IBM websie, the following may be observed.

• Intel uses decent connectors, and IBM uses DB9's. Go, Intel.
• Intel's form factor is 3/4, while IBM is full. Go, Intel.
• Intel is 4 ports, IBM is 2 ports. Hmm. Why in the world would anyone want 4 ports on an add-on card?
• Intel uses external retimers, IBM uses internal ones. Go, IBM combined semi process!
• Intel has a fan, IBM does not. Go, IBM.
• Intel is 1x, IBM is 4x. Go, IBM.

Never mind the stack, Intel do know a thing or two about hardware. Or some parts of Intel know, anyway. I still remember Intel 82586 and its inter-packet gap problem. Man, some people do like to hold grudges :)

And don't forget, IBM's drivers are not open source, while Intel's are (they are dual licensed under GPL and BSD-no-ad-clause with patent grant). Nice pictures aside, IBM seem to be a little behind in respect to Linux. Isn't it a little strange for a company which loves Linux so much?

rms spreads falsehood

In an article in Petreley's Linuxworld at http://www.linuxworld.com/site-stories/2002/0520.rms.html, RMS writes:

The use of Bitkeeper for the Linux sources has a grave effect on the free software community, because anyone who wants to closely track patches to Linux can only do it by installing that non-free program.

This is such an outrageous bullshit, and from a such influential person that I cannot let it slip. Anyone who wants to track Linux patches has only one choice now: use kernel.org FTP site. I installed Bitkeeper too, and found that:

• Identification of a patch level or revision is nearly impossible, because a proper tagging of changesets was not done. Tracking of changeset numbers requires a Sherlock Holmes' skill and minute attention.
• Mirrors are not run properly and lag by various amounts.
• Bitkeeper generated patches are officially discouraged by Linus (I have his e-mail to that effect, it has something to do with the graph of trees ordering and its conflict with patch application in Bitkeeper).
• Bitkeeper is buggy.
• And the most important - Linus discarded -pre and provides development kernels in the proper way at small intervals, thus removing the whole reason to use Bitkeeper.

So, gregkh uses Bitkeeper. It only shows that Greg is a mazochist. Remember that he took on USB maintainership voluntarily.

Al Viro does not use Bitkeeper, and is perfectly functional without it. And I happen to do the same.

Intel's Infiniband Stack is GPLed

Yo rejoice, Intel published their stack, which they nursed since the NGIO days, at http://infiniband.sourceforge.net/. Bob Woodruff is to be congratulated for pushing it all, and it is a pity he was unable to cut the red tape before.

The beast is 220,000 lines long according to wc. It's got the the driver for the hardware, beginings of IP-over-IB, and VIA emulator of sorts. What it has NOT got is a Subnet Manager, so even if you manage to build the thing, and if you have the hardware, you still need a binary from some extortion startup.

My current plan is to finish the very minimal Subnet Manager (SM) running with my Toy stack, then to connect two stacks together using the HCA emulator. This requires a port of HCA emulator to Intel's stack. Once there, take the SM and port it over. It would essentially merge the Toy stack and Intel's stack, minus all nice design ideas in the Toy stack. Those have to battle it out with Intel's articles in a contest of merit.

Life of driver monkey

Speaking of SM, it has to discover and periodically sweep the subnet, thus providing an opportunity for me to program a graph walking and a couple of index lookups. Simple pleasures of life. I cannot stand filling out stupid little fields in packet headers anymore. My brain is all getting crusty.

OLS

Bob and I scheduled an OLS BOF. When I did it last year, it was very lame, because everything was NDAed and nobody doing related work showed up (except Greg K-H, IIRC). This time it's going to be much different, with explanations about real code (well, I hope for good questions - come over and ask them). It is very likely that we'll have code bases not merged yet, so I'll answer questions from Toy stack perspective, and Bob and Ashok will answer questions about Intel's stack.

kir:

I you did not read "Monday", do it. When Naina Kiyevna pests the hero to give her a lift to the Bald Hill, he decides to end it by asking 50 rubles. She looks at him with respect and bugs off.

tk:

Once upon a time I went a little further with PDP-11, by necessity. I aquired a DVK-3 CPU (a Soviet clone) with 256KB of memory and a little microcode bug or a feature in MTPS instruction, which precluded regular images from booting. I found a portable assembler, and hacked K&R cc to run on a sparc (big endian, and 32 bit words). Then, I recompiled bootstraps, wrote them on a floppy, and was able to install the system. That source is privately available for people who need it.

UNIX v7 did not have a shutdown command. My operator's manual called for issuing "sync" command TWICE, then WATCHING the disk controller lights until they stop flashing, then halting the CPU with a key, halting and powering down disks, then powering the CPU off.

Infiniband, this time for real.

I forgot to mention in previous entries that this week was a red week for my Infiniband project: it was published at <http://www.fenrus.com/infiniband/>. The license is GPL. I may re-issue it as BSD if someone asks. I am not religions about these things as long as good code gets written.

Currently, the stack is only a toy, but I cosider the release a big deal. One very big corporation kept promising to open their existing and fully functioning stack for about a year (since I started to work on my stack). This waiting while they were goofing off did an untold damage to the developemnt of Infiniband support in Linux. I am very happy that the waiting is over.

I guess this means that I am joining the counless ranks of wanna be project leaders who look for helping hands.