I decide who lives and dies

whargarrrbl

Mon, 21 Sep 2009 05:42:14 GMT

Today was princessdammitt 's surprise birthday party. damiondead rented a caboose on a steam train for a whole group of us, after which there was a picnic. Somehow he managed to keep the whole thing secret from her, which apparently is an impressive feat as Damion is not known as the king of surprises. Everyone was dressed up in victorian/steampunk attire, much the amusement of the other passengers and the railroad staff.

This past week, I finally conquered the P4080. This is the multicore network accelerator doohickey that I've been pounding on at work for the past couple of months. Originally all the work was being on using a simulator as real silicon wasn't available yet, but we finally got a couple of actual boards about 10 days ago. Naturally, the same code that worked on the simulator didn't work on the real hardware. It took a few days of poking around to figure out why, but it's finally done. There are a few minor issues left to smooth out, but all the really hard stuff is behind me. This is probably the most complicated device I've had to program so far in my career, and I'm kind of surprised I ever managed to make it work. (The reference manual is over 3000 pages long, and it isn't even the final version.)

The saga of the guy with the radio next door continues. After a couple of days of quiet, it started up again. Although it was coming from the same room, I'm not certain that it's the same guy: I went over there on Friday after I got home from work (and heard the radio blaring again) to ask one more time if he would keep it down. The guy I saw didn't look like the same one I'd seen previously (the haircut was different). His neighbor from across the hall tried to tell me he was a paranoid schizophrenic, and his arms shook a lot and his speech seemed a little slurred. I told him how he had the radio on all night the previous night, and he tried to tell me to just let him know when it was too loud and he'd turn it down. (This of course does me no good; I can't go over there every time, especially if it's in the middle of the night, and if I yell out my window it'll just wake other people up.)

So I finally decided to go for the nuclear option. I ordered one of these:

http://shop.christmasincruces.com/product.sc?productId=2&categoryId=2

It's a very low power transmitter (25 milliwatts), but given his close proximity, I think it ought to be strong enough to overwhelm whatever station he's listening to. And if not, I can probably build a amplifier to boost the signal a couple of watts. If I wanted to be really evil, I could hook a microphone to it and do the whole "Kent, this is god speaking" routine, but that probably wouldn't help his schizophrenia much.

I've been toying with this idea for a while. I'd prefer not to do it, lack of sleep turns me into an ever bigger stress monkey than I am normally, and that's just not good for business. Anyway, the thing should be here in a couple of days. We'll see how it goes.

Share and enjoy...

Tue, 08 Sep 2009 20:00:34 GMT

This morning, slashdot posted a link to this:

http://seclists.org/fulldisclosure/2009/Sep/0039.html

No, this is not a repeat from Windows 95. Well, ok, maybe it is. In any case, the exploit code is in Python. I don't like Python (or Perl), so I rewrote it in C:

http://people.freebsd.org/~wpaul/w00t/smb-bsod.c

Compile with:

% cc -o smb-bsod smb-bsod.c

Run with:

% ./smb-bsod [IP address of machine to clobber]

Supposedly this will clobber Win7, Vista, and possibly Win Server 2008. Have fun, at least until next Patch Tuesday.

Well poo

Fri, 31 Jul 2009 03:16:07 GMT

There's a new apartment building at 1 Polk, about two blocks from me, called the Argenta (http://argenta.riverstoneres.com/). Originally it was supposed to be all condos, but that plan apparently didn't work out, so they're now all rentals. I figured at the very least it would be worth a look, so I signed up for a tour.

Sadly, the prices they're asking are higher than anticipated. For whatever the reason, units are more expensive the higher up you go, and the least expensive unit in the layout I was interested in was on the 10th floor. On the plus side, it's a high-rise (I can look down from on high, etc...), it's bigger than my current place, it's quiet, it comes with a washer/dryer in the unit, the view is pretty nice (it faces Civic Center) and there's a 24-hour fitness center. On the minus side, they're asking $2431/month. Add to that about $40/month more for water/sewer/trash. Also, they want you to take out at least $50,000 in renter's insurance. The only incentives they're offering are about two week's free rent.

Oh, and all of that isn't counting whatever money it costs me to actually move.

I also noticed a couple of strange things. One was that I could not seem to get cell phone reception on my Blackberry anywhere in the building. Another was a strange water dripping sound in the washer/dryer unit.

So basically, I'd be looking at an increase of about $1200/month in expenses (guessing, dunno how much renter's insurance is at this point). I was hoping for maybe $700-$800 at most. It's not beyond my ability to afford, but it's enough to dull my excitement about the idea. I am conflicted. There's still some temptation there, but my fiscal responsibility subroutines are sending out strong warnings.

Blah.

Two nights in a row...

Wed, 15 Jul 2009 23:33:06 GMT

...I've actually been able to sleep. Either the guy next door with the radio finally got the point, or he moved out, because the last two nights have been blissfully quiet. Not only that, I've actually slept the whole night through. It's glorious.

In other news, I ordered, and just received, one of these:

http://www.soekris.com/net4826.htm

It's an upgrade for this one that I already have:

http://www.soekris.com/net4521.htm

I've been using the 1000mW 802.11b/g card with the existing one, and by itself it works fine: I have a rock solid signal to the "Free the Net" node one block over. But I also want to run a second wireless interface that I can use to create a wifi network in my apartment. That part has issues: I have an atheros-based cardbus card (Linksys) that I can plug into one of the cardbus slots, and it works, but if I load the thing down with a lot of traffic, the board will spontaneously reboot. No warning from the OS, no nothing: just *pewf*, restart.

I suspect a power supply issue as the same setup works if I use one of my older, lower power cards in place of the 1000mW one. I don't know for sure if it will work any better in the new board, but I think it's worth a try. In nothing else, having a faster processor won't hurt. (The old board runs at 133Mhz. Yes, 133Mhz.)

I also ordered some new antenna cabling. I got a replacement 'pigtail' connector to attach directly to the 1000mW card, via MMCX port instead of the teeny, tiny U.FL port. The new cable also has a different panel-mount connector at the other end which fits exactly into the pre-drilled holes in the case of the Soekris board. This helped to reduce the insertion loss I was getting with the old setup. I probably only gailed a couple dB worth of signal, but the new setup is much sturdier.

Yes, make me grumpier: what could possibly go wrong?

Mon, 13 Jul 2009 08:40:18 GMT

Although I've tried to conceal it, for the benefit of those around me who are not at fault, I've been a seething little ball of angry lately. It bubbles beneath the surface, this angry does, just waiting for a chance at release.

I came home tonight (at around 12:30) to find that once again, someone in the SRO next door has the radio on. I can't fall asleep with a radio or TV on. The sound of passing cars I can stand, but voices and music drive me bonkers when I'm trying to doze off. I just can't tune them out, and my brain keeps focused on them, which makes next to impossible for me to fall asleep.

In this case, the radio was sitting on the sill of an open window, about 5 feet away from my window. Even with my window shut, I could still hear it. I looked over and saw the lights were out.

Luckily, the people who run the place seem to be sympathetic to my plight. I went over and walked upstairs to the offending room, and could hear the music out in the hallway. I knocked repeatedly, but there was no answer. The manager came up with the key, and we discovered that the room's occupant was not even home. Apparently he'd just turned the radio on and left.

I unplugged the radio and stormed back to my apartment.

These days, about the only good things I have to look forward to are spending time with friends, eating food that's not good for me, and getting a good night's sleep. Apparently that's too much to ask, which is a shame, because not sleeping is only going to make me even more irritable than I already am, which frankly I don't need.

I'm starting to feel like my life is just a giant cluster-fuck, infrequently interrupted by all too brief moments of sensibility, instead of the other way around. When I see how I've ended up, all I can think is that I don't want to be here. I'm not just talking about where I live, either, though I think that's a big part of it. I would like to live somewhere where it's quiet at night. Where I'm not constantly assaulted by the smell of pot in the hallways, or worse, outside my own window. Where they don't dump trash outside my building. Where water doesn't sometime leak through the ceiling from my upstairs neighbor's bathroom when he takes a shower. Where the garbage men don't leave a trail of smelly trash in the basement when they take out the garbage at night.

None of that crap really matters though. I could find the perfect home for the perfect price, and it would still suck, because at the end of the day, it would be just me living in it. Some days, coming home to an empty apartment just kills me. So by the time I discover that some schmuck has decided to keep me awake with his radio all night, I'm already plenty aggravated. The noise is just icing on the cake. A rage cake with violence frosting. And I'm just dying to serve it up.

Bleh

Sun, 12 Jul 2009 01:22:55 GMT

I just realized that the Marquis Fetish Ball is happening next Saturday in SF, and of course I have no date.

*grump*

Free the internets

Sat, 04 Jul 2009 06:50:09 GMT

I discovered recently that a "Free the Net" wifi node had appeared within range of my building in SF. This was fortunate as the access point that I had previously been leeching from now was no longer unsecured. (Took them long enough...) The "Free the Net" network is a basically the closest thing to free community wifi in SF. There's info about it here:

http://sf.meraki.com

It took a bit of fiddling with my cantenna, but I finally found the right direction (straight out the window) and polarization (horizontal) to get a reasonably stable connection to the access point. It was still a little marginal though, so started looking into various options to improve the situation, hopefully in a way that would not require me to mount an antenna on the roof of the building. (I can get the roof, but I'm on the second floor of a four story building, and at 2.4Ghz, a cable run that long would probably introduce unacceptable signal attenuation.) My first thought was a higher gain antenna, though most of good ones seemed to be quite large and unwieldy. But then someone showed me this:

http://www.metrix.net/valemount-networks-kxs30sg-p-438.html

An 802.11b/g wifi card with 1 watt of TX power. I didn't think they made such things, or that you can buy one in the US, but apparently they do and you can. I ordered one and it arrived today. According to the client survey tool on the remote node, my signal strength more than doubled. (I had been using an Intel 2200BG card previously.) Unscientific web tests show I'm getting about 1.2Mbps download speed, which ain't bad for free.

Oh, and it's based on the Atheros 5414 chipset. It works with FreeBSD out of the box, and since it's Atheros there are all sorts of funky things you can do with it.

I am suitably impressed.

Down time during down time

Thu, 02 Jul 2009 03:15:40 GMT

I'm off from work this week, taking my 3rd and final furlough week of the year. Technically, I'm taking advantage of a loophole in the furlough process: we're required to take a complete week off, but this Friday counts as a paid holiday, since the 4th of July falls on a Saturday this year. That means they're really only allowed to charge me for 4 vacation days rather than 5. A bunch of other WR people are taking this week off as well (though some are taking next week off too, in some cases because of pre-existing vacation plans).

The only problem with having the time off is that while it gives me the chance to catch up on lost sleep, it also means I get to spend a lot of time doing nothing, which is bad because when I have nothing to do, I think, and when I think, I get depressed.

The same day that the Intel acquisition of Wind River was announced, right before the start of my previous furlough week, things thoroughly fell apart with Tabby. I don't feel like going into detail: it just didn't work out. (I'm starting to think that will be my epitaph.) Since then, I've been stuck partway between angry and sad. There are times when I feel as if I could launch into a long, drawn out and overwrought introspective of the whole thing, but nothing much ever comes of that feeling because I know it would just be a waste of perfectly good time and electrons. And besides, that's been done to death.

That said, I am, quite frankly, at a loss. I keep thinking there's something I'm supposed to do next, but I can't get a handle on what it is. There is an overwhelming sense of malaise that I just can't quite see past. I don't have the answers, and I hate that.

Just when I thought it was safe...

Fri, 26 Jun 2009 03:58:03 GMT

Most of the day at work was occupied by various discussions related to the sudden death of Michael Jackson today. For some, it's practically a national day of mourning. For others, it's an excuse to trot out every internet meme that's ever existed.

By the time I headed out of my office to meet up with the carpool group, I thought I'd seen it all. Then, as I get in the car, the following conversation with one of my decidedly mundane cow-orkers ensues:

Cow-orker: "You heard about Michael Jackson right?"
Me: "Yeah, I've been hearing about it all day."
Cow-orker: "Well I heard a rumor that Harrison Ford and Jeff Goldblum died too."
Me: *brief pause, followed by peals of maniacal laughter*

I had to explain the joke to her. I still don't think she quite got it.

The trip home took a little longer than usual due to some kind of traffic accident in the Posey Tube in Almeda, which required us to take the long way off the island. When I arrived home, I found a noticed posted outside main entrance to my building. It was a notice of violation from the city, citing the building owners for failing to provide heat in the building. I had actually noticed the lack of heat after I returned from my trip to LA, but I didn't really mind it because it happened to coincide with pretty nice weather here in the city. The problem with the heat is that when it's on, it's impossible to avoid: the steam radiators in the apartments can only be controlled by a shutoff valve, but the valves are so old that they've been ground down to practically nothing, making it impossible to really halt the flow of steam. I also suspect that the boiler only has a timer on it rather than a thermostat, given that it seems to come on at the same time every day, even during the summer.

According to the notice, a follow-up inspection is scheduled for next week, and each failed inspection will cost the landlord $170. This doesn't strike me as a terribly stiff penalty.

I noticed that Pidgin started having trouble with Y! Messenger again recently. It seems that Yahoo has once again jiggered their authentication servers, possibly in an attempt to force people to upgrade their clients. Luckily, there's already a new version of Pidgin available to fix this. I upgraded my setup at work an it's happy now, but I'm going to have to rebuild it on my home machine and my laptop too. Joy.

Last weekend was Gothnic 2009. Much fun, food and sun (booooo! hisssssss!) was had by all. I saw many people that I hadn't seen in a long while, which was nice even though it made me feel even more like an old fart. I even managed to drag Ceren out of her hermitage. Damion should consider this an organizational success, even though he may insist that he has all the organization skills of a pack of retarded gerbils. Luckily, with a group like us, who basically _are_ a pack of retarded gerbils, you don't really need much more.

Work has been eating my brains more vigorously than usual. The Intel merger/acquisition hasn't been finalized yet, so things are still mostly business as usual, though there's definitely an undercurrent of anxiety. I've been too busy to really think about it, but I know that in other departments there's been no shortage of gossip. For one thing, since we will now be part of Intel's accounting process, there's some question as to what will happen to our own finance and accounting people. As for me, I had just enough time after getting back from my first furlough week to finish up my latest project before having to take another one next week. I'm still not sure what I'm going to do with the time, but I did decide that I don't want to go to to Toorcamp. (Hey, Graziela, are you doing anything next week? *poke* *poke*)

My trip to LA was fairly relaxing. I stayed with Pixel again, and he hosted a party for his friend Amber the Saturday after I arrived. There were balloons, toys and snacks (the latter of which produced a photo labeled "two girls, one cupcake"). I also met this really cute girl who does some kind of database programming who, given my luck, I'll probably never see again. (And if I do, I'm sure she'll introduce me to her boyfriend.) I also got a chance to futz around with Pixel's brand new HP 2140 Netbook. Since he was going to trash the factory Windows install put Linux on it, he let me try it out with FreeBSD 7.2, since I happened to have the CD with me. The 2140 has Broadcom wifi and Marvell ethernet. Neither one worked exactly out of the box, but did work eventually with some coercion. The Broadcom wireless required Project Evil. For the ethernet, I had to grab the latest msk(4) driver from FreeBSD-current. It has Intel graphics, which did work out of the box with Xorg. So did the sound and bluetooth. I didn't do anything with the built-in camera. Unfortunately, video is one of FreeBSD's perennial weaknesses: Linux has a framework for it, but FreeBSD doesn't, so all camera drivers are pretty much ad-hoc.

The one thing I was really curious about was whether or not the Intel Atom processor in the 2140 is actually faster than the Celeron M in my existing laptop. The Atom has a higher clock speed, but from my experiments there didn't seem to be a huge difference in response and performance.

Oh, we also ran into some of HP's stupidity. Pixel wanted to replace the internal Broadcom wifi adapter with an Atheros one, for various sundry reasons. First of all, HP designed the laptop so that you have to take it completely apart in order to remove the wifi module. And I mean _completely_. Second, after going to all that trouble and reassembling the whole thing, we turned it on only to be greeted by a BIOS error telling us that an unsupported piece of hardware was installed, and that we had to remove the offending device in order to successfully boot. It turns out HP rigged the BIOS to check the PCI vendor/device ID on the wifi card, and it won't start the system up if you don't have a "supported" one. So we had to disassemble the laptop all over again to put it back the way it was.

I hope a lot of people complain to HP about this, though something tells me they probably won't, which is a shame. I'm still on the fence about getting an Atom-based netbook. I kinda want one, but can't really justify it while my existing laptop is still fully functional. Ah well, they'll still be out there if I change my mind.

This just in...

Thu, 04 Jun 2009 17:09:20 GMT

Intel accidentally my whole company.

http://money.cnn.com/news/newsfeeds/articles/reuters/MTFH42071_2009-06-04_16-03-37_BNG177267.htm

I wonder if they'll change the sign out front again.

I have seen the future, and it looks just like the present, only with more special effects

Wed, 27 May 2009 22:26:26 GMT

I had Memorial Day off from work, so I went to see Terminator: Salvation with damiondead . The reviews that I'd read beforehand were pretty much correct: lots of special effects, lots of action, very little plot. The one high point in the movie for me was the appearance of a computer-generated Ah-nuld, who, ironically, seemed to be capable of the same emotional range as the real one.

Before heading to the movie, damiondead dropped off his iMac at the SF Apple store. He's apparently been having a lot of strange problems with it which the "geniuses" at the other Apple stores have been unable to fix. (The last time they apparently returned the machine with a diagnostic routine of some kind still installed on it.) The problems he describes (the machine sometimes freezes after an iPod is plugged into a USB port, sometimes it freezes during heavy graphics activity (i.e. playing WoW), sometimes audio will stutter and then it will freeze) all sound like classic Heisenbugs. Given the non-deterministic nature of the failures, I'm not inclined to think it's a software problem, especially since he upgraded the OS to 10.5.7 just recently, and that didn't help. The RAM has been replaced, but the problems occured both before and after the RAM upgrade, so faulty RAM itself is probably not the issue either. Trying to think of potential sources of trouble, I came up with the following:

- Machines fear Damion (though, clearly, they don't fear him enough)
- Cat fur-induced thermal problems
- Dante feeding the computer a ham sandwich while Damion's back was turned
- A transient system board fault (bad solder joint?)
- Some bizarre form of electromagnetic interference
- A problem with the power supply in the computer
- A problem with the electrical system at Casa Dammitt itself
- Canada

Normally I'd be totally in favor of just blaming Canada, but that won't help Damion, and I kinda want to. Evidence suggests that the last batch of Apple geniuses couldn't find anything wrong with the machine, though that could be because they're all a bunch of hipster douchebags. But if it really didn't fail for them, and if the tests they ran were really worth a damn, then that means the problem is situational. I'm really tempted to say that the problem is not the computer, but something at Casa Dammitt itself. The only trouble with that theory is that there's a bunch of other electronic doodads there which work fine (including Tina's computer, which is also a Mac).

I guess it is possible for this particular machine to be especially picky about input voltage variations or line noise. Finding a way to prove this though is the tricky part. One possible test would be for Damion to bring the machine into work and run it there for a while, though a) I'm not sure how picky his boss(es) would be about him bringing a non-work system into the office and b) I get the feeling that these days the only thing he wants to bring to work with him is a club with a spike in it. I guess we'll have to see what the geniuses from the SF Apple store have to say.

FreeBSD upgrade follies

Mon, 18 May 2009 23:17:46 GMT

I finally decided to upgrade the OS on my laptop. I've been running FreeBSD 6.0 on it for a few years now, and while it still works ok, putting newer software packages has become a bit of a chore, and there are a few bugs which have long since been fixed in newer versions. Since FreeBSD 7.2 was just released, I thought I'd give it a try.

Most things worked ok out of the box, with two exceptions: wifi and accelerated graphics. Actually, that's a bit of a fib: FreeBSD has a native driver for RaLink wifi, and it did work out of the box with my wifi card. But I wanted to use the Windows NDIS driver with Project Evil (the NDIS emulator) so that I could use WPA2 Enterprise security (which I need for work). That gave me a lot of trouble: the Windows driver would crash my laptop. It turns out that this was not a FreeBSD problem though: it was a bug in the Windows driver. The same buggy driver would cause Windows XP to blue-screen too. Fortunately I was able to find a slightly newer version that works correctly. Everything including WPA2 Enterprise security worked great after that.

The graphics problems were not so easy to resolve. The main problem here is that the bulk of the Xorg development is being done on Linux these days, and some of the Xorg drivers need to work in conjunction with additional drivers in the kernel, at least if you want accelerated graphics. But all of the latest Xorg code expects all of the latest Linux driver features, which haven't yet been ported to FreeBSD. My laptop has Intel graphics, and the Intel driver has been modified to use some special kernel feature called "GEM" which FreeBSD doesn't have. That makes life interesting.

I could go on for several pages about what I had to do to sort all this out, but to make a long story short, I had to patch the Intel i810 AGP driver a little and install an older version of the Intel xorg video driver that wasn't quite so insistent about using GEM before I finally got everything to work the way I wanted. Now that I have, it's actually pretty spiffy.

I also decided to give KDE4 a try. It's shiny, but it's a CPU hog by default, which is less than desireable on my laptop's 1Ghz celeron M processor. For one thing, the "DirWatch" feature in kded4 scans your files every couple of seconds. You have to edit a configuration file to increase the polling interval to something more reasonable. Also, KDE4 comes with something called "nepomuk services" enabled by default:

http://en.wikipedia.org/wiki/NEPOMUK_(framework)

According to that web page, it cost 17 million euros to develop. I'm not sure what it does other than chew up valuable CPU cycles, so one of the first things I did was to turn it off.

While it was more trouble to get graphics support set up that it really should have been, everything else seems to be working pretty well. The wifi works nicely with wpa_supplicant, and I was able to get tethering to work with my Blackberry via bluetooth very easily. And DPMS suspend works on my laptop now, which it never did before (the backlight wouldn't shut off). Sadly, ACPI suspend doesn't work, but then it never did.

I can't wait to find out what horrors await me in FreeBSD 8.

Weird crap day

Thu, 23 Apr 2009 04:13:14 GMT

Today has been more than a little strange, even for me.

As I walked out of my building this morning and headed up the block towards Oak street for ye olde carpool pick-up, I was confronted by the sight of a what I'm forced to assume was a homeless man walking towards me, urinating on the sidewalk. Now, I don't mean he was standing on the sidewalk peeing against a wall or something: I mean he was walking along, pretty as you please, with his dick out, as if he was watering the lawn. This was just too outrageous to ignore. I charged at him, firmly intent on drop kicking him into next week, but thought better of it when I decided I didn't want pee all over me. I screamed at him and threatened to call the cops. He was unperturbed by this, but I kept on yelling at him and berating him for behaving like a miserable excuse for a human being. I must have made an impression because he apologized. Whether he meant it or not I can't say.

Later on, I was working in my office when I heard a hissing sound, and I suddenly felt moisture spraying onto my arm. I looked behind me and the only culprit that caught my eye eight away was the fire sprinkler in the ceiling. I stuck my chair against the desk behind me and climbed up to get a closer look, and that's when I realized what had really happened.

Before coming to Wind River, I worked for BSDi. Actually, I worked for what used to be Walnut Creek CD-ROM, which had been bought by BSDi. When Wind River acquired BSDi and we left the old WC CD-ROM offices, I ended up inheriting various unclaimed office tchotchkes that nobody really wanted, but nobody had the heart to throw away either. (Sadly, my favorite of these, a tangled bundle of magtape with a post-it on it that read "FREEBSD BACKUP ARCHIVE," has disappeared.) One of these items was a can of Dole pineapple chunks. Supposedly this can once belonged to Mike Smith, and there was a story of some kind associated with it which I can no longer remember. The can has been sitting on top of a wall-mounted storage cabinet in my office ever since I started working at Wind River. It's been there so long I pretty much phased it out of my conscious mind.

Today, it burst.

When I climbed up on the chair, I saw that the can had bulged and sprung a leak at the bottom. Luckily, most of the fluid that came out ended up on one of the open cabinet doors rather than on the various target boards sitting on the table below it. Some time was spent cleaning up the mess and tossing the now-defunct artifact in the trash. (The "best if used by" date on the can was in 2002.)

In the afternoon, everyone received an e-mail from our illustrious CEO. Apparently the forced one-week furlough last quarter, a cost-cutting measure designed to force people to burn up vacation time, worked so well that they want to do it again next quarter. It seems to me there's something fundamentally wrong somewhere when a company can make money by not having its employees work.

So how was _your_ day?

Work pays off and stuff

Mon, 20 Apr 2009 00:25:27 GMT

Intel sent us a sample of their 82598 dual port 10Gbase-CX4 adapter, which means I was finally able to run some decent benchmarks on the 82598 driver. I decided to put the card in the Nehalem board and run it against the Myricom 10GbE card in the Bear Lake board. The results were pretty good: I was able to achieve 9.8Gbps in the TCP stream tests, which basically means I maxed out the link.

I had to run VxWorks in SMP mode to do it though. Running uniprocessor, the same test maxes out at about 5.6Gbps. I'm not sure if this is because we're actually getting some useful parallelism in SMP, or if it's due to other optimizations that come into play in the SMP build. I suppose I could test it by building an SMP image that only activates one CPU core. Maybe I'll have time to do that next week.

I got a letter this week informing me that I still have money in a retirement plan account from when I worked at Columbia University. It never occurred to me to check my retirement plan when I left, and Columbia lost track of me since I never kept them my apprised of my new addresses. The account has still been accruing funds, though it pretty much flattened out last year (which isn't terribly surprising). What I don't know is if I should leave the money where it is, or cash it out and dump it in my savings. I don't know what penalties I might incur if I terminate the account and cash out now, or if there's a risk that the account will lose money.

'Tis a puzzlement.

The thing is done

Sun, 05 Apr 2009 20:08:43 GMT

Well, it took a bit longer than I expected, but I conquered the Myricom 10G card.

mye0 Link type:Ethernet HWaddr 00:60:dd:47:7d:75 Queue:none
capabilities: TXCSUM TX6CSUM
inet 10.0.99.1 mask 255.255.255.0 broadcast 10.0.99.255
UP RUNNING SIMPLEX BROADCAST MULTICAST
MTU:9000 metric:1 VR:0 ifindex:3
RX packets:110936923 mcast:0 errors:0 dropped:1
TX packets:602410047 mcast:0 errors:0
collisions:0 unsupported proto:0
RX bytes:1206M TX bytes:2311M

Getting the TX logic just right proved to be the biggest challenge. I had a race condition in there for a while that kept causing the transmitter to stall in certain tests. I have it running pretty solidly now though. (Made it all the way through 3 or 4 different netperf runs without any issues, and the TX torture test passes.)

The performance isn't too bad, though the max TX frame rate isn't as good as that of the Intel card. Using an SMP VxWorks image on a 2.4Ghz Q6600 Core 2 Duo processor, I was able to achieve just over 9Gbps in the netperf TCP and UDP streaming tests.

That's a lot of CPU power though. The 2.0Ghz T2500 Core Duo test machine could only reach 5.5Gbps. FreeBSD running on the same hardware is a bit faster in the same test. This could be due to additional overhead in the TX path of the VxWorks TCP stack.

I'm hoping to acquire an Intel 82598 CX4 copper card soon so I can finally collect some meaningful test data from the Intel driver too. I have a sample Nehalem board (CoreI7) across the hall from me that I'd love to run some benchmarks on, but sadly the Myricom cards seem incompatible with it in some way (the boards reboot partway through BIOS initialization when the card is plugged in).

FreeBSD 7.0 wouldn't boot up on it either, which was a big disappointment. I'm not sure what the deal is there. I only tried the GENERIC kernel though, which has SMP on by default. At some point when I'm not too lazy, I need to try a UP image.

I also found out there's another rev of the Intel 10GbE silicon coming out soon. This one has 128 RX and 128 TX queues, and hardwre crypto stuff. It remains to be seen when we get samples for it. We still don't have a decent 10GbE testbed.

I'm off from work next week. They're making everyone take a week off to burn up accrued vacation time as a cost saving measure. The timing is kind of crummy, since there are things that need to get done. But at least they picked a time when the weather seems nice.

I think I'm going to go out and read this week's funny books now.

Ok, I get it, but...

Sat, 21 Mar 2009 19:20:15 GMT

That's what I kept thinking while I was working on the driver for the Myricom 10GbE cards. We bought two of the CX4 adapters and a cable for them a while ago and they've been sitting in my to-do pile. Myricom claims some pretty impressive numbers for them, but, at least from my perspective, Intel has them a little outclassed with the 82598.

The Myricom cards are entirely firmware driven. To program the Intel cards, you have to tweak a fair number of registers. This may sound burdensome, but it's actually pretty normal. With the Myricom cards, you do pretty much everything by issuing commands to the firmware. Want to bring the link up? Issue a firmware command. Want to set the MAC address in the RX filter? Issue a firmware command. Want to join a multicast group? Issue a firmware command. Want to load a new firmware image? Issue a firmware command.

No, I'm not kidding about that last one. Each NIC comes with firmware already pre-loaded in its on-board NVRAM, but Myricom also issues firmware updates periodically, which you can download from their web site. Usually the updates are just for bug fixes or performance improvements, but sometimes they also add features. There also appears to be a separate set of images with RSS support. The driver has the option of using the firmware already on the card, or it can load a new image into the NIC's SRAM instead. (You never actually overwrite the firmware in NVRAM, so there's no danger of permanently bricking the device.)

The fact that the process for loading new firmware involves a firmware command is, at least in my opinion, not such a great idea. If you, say, try to load a corrupt image, then the NIC hangs. Unfortunately, the process for resetting the card via software also requires -- wait for it -- yes, you guess it: a firmware command. There is no dedicated hardware-only register for restting the NIC's on-board CPU. There really ought to be.

Also, the firmware images are fairly large (~100K compressed), and if you compile them into the driver it bloats the object code size significantly.

The CPU in the Myricom cards (whatever it is) is big endian. PCI/PCIe devices are typically little endian, which means to write a portable driver you need to handle byte-swapping on some architectures: on little endian systems (x86, x86-64), things "just work," but on big endian (PPC, SPARC), you have to byte swap register or descriptor values so that the NIC can interpret them properly. With this card, it's backwards: you have to coerce values into big endian instead. I've never seen anything like that before. It's... well, it's interesting.

The Myricom uses two RX RMA rings, one for small buffers and one for large buffers. Packets that are small enough to fit into the buffers in the small ring are put there automatically by the NIC, while everything larger goes into the large buffer ring. The idea here is to make efficient use of memory (putting a 64 byte frame into a 9000 byte buffer is pretty wasteful). The driver can program the buffer sizes how it wants. Well, almost. There are some tricky limitations. Buffer addresses must start on a 32-bit boundary. Small buffers can be at most 4K, and large buffers must be at least 4K. Small buffers must also reside entirely within a 4K page (i.e. they can't straddle a page boundary). I found that if you don't obey these rules, the NIC will freeze up after receiving a certain number of frames, and you have to power cycle to reset it.

There is one TX DMA ring. I haven't gotten the transmitter running yet, but it looks like transmissions must be done in chunks of 4K. That is, if you try to transmit a 9K jumbo frame, you have to set up a scatter/gather DMA transfer with three decriptors, each of which transfers at most 4K (In some cases, you may be limited to 2K.) This will make jumbo frame transmission in VxWorks a little tricky. I can't really punt on it, since you sort of need jumbo frames for 10GbE. I'm not looking forward to it.

Both the RX and TX DMA rings have a maximum size limit which is enforced by the NIC firmware (and probably governed by the size of the on-board SRAM). The rings exist inside the NIC's SRAM rather than in host memory. There is an RX completion ring as well, where the NIC writes the RX status for each received frame. That ring does live in host memory.

I'm also a little suspicious of Myricom's claims that you can to IP and TCP/UDP hardware checksumming on both RX and TX. For the receive path, you get a 16 bit checksum value from the NIC. That's it. It's up to the driver software to figure out how to use this in conjunction with the OS's checksum offload support APIs. I know it's possible to do this, but it wastes precious host CPU cycles for 10GbE. And it doesn't look like it supports VLAN tag insertion and stripping.

Myricom's design reminds me a little of the Alteon Tigon II 1Gbps device (which was acquired by Broadcom and ultimately became the BCM57xx NetXtreme family). It too was firmware driven. It too used different sized RX buffer supply rings and a single completion ring, though Alteon's design had three ring sizes (small, normal, jumbo), and in later iterations you had the option of putting all of the rings in host RAM instead of on the NIC. Alteon's design was a little smarter though in that the RX completion descriptors actually told you which ring a given completion event refered to, and it provided more in-depth checksum offload info.

Oh yeah, one more thing. That FreeBSD companion system is running FreeBSD 7.0 on the Intel Cappel Valley eval board (basically a 945 chipset with 2Ghz Core Duo CPU). When I tried loading the mxge(8) driver module, it issued errors about the DMA tests timing out and the interface wouldn't start. I was able to fix this by tweaking the driver code a little to increase the timeout and rebuilding the module, so it works fine now, but I was a little surprised that it botched like that right out of the box.

I'm hoping I'll get far enough to take some performance measurements next week.

Another one bites the dust

Sat, 14 Mar 2009 22:40:03 GMT

I have conquered the Intel 10GbE card.

tei0 Link type:Ethernet HWaddr 00:1b:21:19:39:58 Queue:none
capabilities: TXCSUM TX6CSUM
inet 10.0.99.2 mask 255.255.255.0 broadcast 10.0.99.255
UP RUNNING SIMPLEX BROADCAST MULTICAST
MTU:1500 metric:1 VR:0 ifindex:3
RX packets:17 mcast:0 errors:0 dropped:0
TX packets:5484184 mcast:0 errors:0
collisions:0 unsupported proto:0
RX bytes:2608 TX bytes:3664M

Sadly, the TCP performance tests I've run max out at around 680Mbps, which is a far cry from the 10Gbps theoretical limit. Raw UDP tests (which are transmit only) produce somewhat better results, reaching aroung 3.5GBps. I haven't tried jumbo frames yet.

One of the limiting factors here is a lack of decent test gear. The only other system I have with a 10GbE fiber port is a MIPS target which can run either VxWorks or Linux. The netperf TCP results seem equally bad with either one. I don't have any way of doing forwarding tests, since the Smartbits box we have doesn't have 10GbE interfaces. *sigh*

More stupid work crap

Wed, 11 Mar 2009 08:22:47 GMT

I had a support issue from Broadcom foisted upon me last week. Unfortunately, it was one of those things where I couldn't reproduce their problem on our own hardware (the test that they said crashed always worked for me). Finally they sent one of their targets to us, and I saw the problem for myself and had a chance to run it down. It turned out to be a bug in their own code. (Well, ok. The major problem was a bug in their code. There were a few minor problems due to bugs in our code.)

We had a conference call this morning to wrap up all the details. I was the only WR person on the call though, because everyone else was attending the all-hands meeting. Yes, we had another one of those. I never go to them anymore because their either boring or depressing. This one turned out to be a bit of the latter.

Apparently, they're doing away with 401k matching. My understanding is that they match 401k contributions with stock, so it's debatable whether or not this is any great loss. And I don't participate in the 401k plan anyway.

But that's not all.

They're also giving everyone a forced one week furlough in April, to force people to burn vacation time. I happen to have plenty of banked vacation time, so this won't make much of a difference to me either, aside from the fact that I'll almost certainly be interrupted in the middle of an important project when it happens.

Lastly, there was also supposedly words to the effect that if these cost cutting measures didn't... well, cut it, that there might be salary cuts too.

They better hope not, because for me, that would be the last straw. It's bad enough putting a hold on bonuses and raises for a while, but if they try to cut my salary too, I'm very likely going to tell them to go climb a tree. I strongly suspect that they're doing a lot of these things not because we're desperate for money, but because all the other cool kids are doing it too. (And by cool kids, I mean idiots.)

Having said all that, the wire transfer from Cirosec reimbursing me for my trip to Germany finally came through today. Whee.

Lights are on, no one's home

Tue, 03 Mar 2009 06:24:39 GMT

The power outage I wrote about last time mercifully lasted only a couple of hours. I find it annoying that when my building's grid goes out, the buildings right next door stay lit.

Things have been quiet lately. Perhaps a little too quiet. The night I flew back from Germany, Tabby called and asked if I wanted to meet up. I did, but I was wiped out from the trip. We've been trying to get together since then, but she keeps canceling, for various different reasons. We were going to try again for tonight, but it seems now like that's not going to happen. I'm not sure what this means. It may not be deliberate, but it irks me none the less.

What also irks me is that I've been a little too anti-social of late. In spite of wanting to Go Places (tm) and Do Stuff (tm), I just can't seem to get myself motivated. Of course the weather isn't helping, but that's not the only thing holding me back. I don't really know what is holding me back, exactly. Maybe I've just fallen out of the groove.

Hopefully there will be Watchmen-ness this weekend, assuming that Mr. and Mrs. Dammitt and the amazing Burrito Boy recover from their respective bouts with the plague. Truly, I envy them not.

I found out that there was one other development that happened at work while I was away: on top of the mini-layoff and the retirement of the sabbatical policy, there was also a salary freeze. I shouldn't complain since my salary is far from sucky, but still. In spite of this, we also just blew a few million dollars on buying yet another company, this time to get our hands on some embedded graphics and GUI builder technology.

I finally got Intel to send me an 82576 card so that I could add support for it to the PRO/1000 driver. Getting it to work was pretty much a no-brainer. This means we're no longer lagging pitifully behind in our Intel gigabit support. (The driver now supports 80 different devices. Man what a pain.)

Of course, as soon as I got that done, the Intel rep started asking about 10 gigabit support again. So I finally got out the Intel 82598 fiber-optic 10GbE card I've had sitting in my to-do pile and started working on it. Technically I haven't been assigned to work on this yet, but I've found that the best way to get support for these devices done is to just do it when nobody's looking. I swear there are times when it takes more time for marketing/management to fuss over whether or not to commit to supporting a given NIC than it takes me to just write the driver. I figure this way I save some time.

I'm still thick in the tedium of register transcription, sadly. But the device is looking pretty straightforward to program, now that I've had my nose in the documentation for the past few days. There's some weirdness in setting up the interrupt vectors, but I think I can deal with it. I'm roughly at the point where I can start getting the thing to receive packets. The one major complication here is that I only have one of these Intel cards. Why? Because the damn things cost $1500 each. I've only got one other system with fiber 10GbE ports, and it only runs VxWorks. We don't even have 10GbE ports on our Smartbits traffic generator.

What this means is that I've got a good enough setup that I can get the driver completed, but I can't really do any decent performance tests with it, and I just know people will ask me for performance numbers once I'm finished. I guess they'll just have to lump it.

Sat, 21 Feb 2009 05:35:07 GMT

The power went out in my building about an hour ago. No idea when it'll be fixed. At least now I don't have to listen to my upstairs neighbor's TV/music/whatsis.

Thanks a lot, PG&E.

Fear and loathing on a plane

Tue, 17 Feb 2009 20:34:39 GMT

I woke up Monday morning in Potsdam and was rather alarmed to find it was snowing. It wasn't a heavy blizzard or anything mind you, but snow of any kind is never good for air travel. Sure enough, when I got to Berlin Tegel airport, I found that my flight to Munich was running an hour late. This was a problem, as I only had about an hour leeway between that flight's arrival in Munich and my connecting flight's departure for SFO. I made it with about 10 minutes to spare. Of course, the gate where my first flight arrived was way at the other end of the airport from the gate where the other flight was supposed to leave. The boarding pass said gate H02. After following the signs (and portents) to terminal H, I finally came to one pointing me to gates H1-H24. Starting with gate H24. Counting backwards. Bastards.

Also, remember the biblical plagues. The plague of boils? Plage of locusts? Plague of frogs and blood, and so on? I think I've identified a new one: the plague of babies. Every goddamn flight I was on was chock full of families travelling with their infant children. And infant children know only one way to express themselves, which is to cry their surprisingly powerful lungs out. And when there's three of four of them, it's even better because they can go in shifts. (And when I say 'better' I really mean 'DIEDIEDIE.') I'm beginning to hope there is a special hell for parents who think it's a good idea to bring their precious little snowflakes with them on trans-atlantic flights, and that it's a very stabby kind of hell.

When I got out of the Civic Center BART, it was damp but the rain had temporarily abated. It was cold, but nothing compared to the cold in Potsdam/Berlin. I laughed at the pitiful excuse for inclement weather and trudged home. Then I had some food and half a bottle of gatorade and collapsed into bed. I was out like a light and had a full night's sleep for the first time in about a week. It was glorious. Then I got up and dragged my ass to work.

Apparently, interesting things happened at the office while I was gone, and I mean that in the chinese sense of the word. There was another layoff. (No, I was not one of the people affected.) About 50 people got the chop, though only one that I knew personally. They also did away with the sabbatical policy. I imagine I'll hear a little more about it later at the staff meeting.

Joy.

Ich bien ein berliner

Sat, 14 Feb 2009 23:55:10 GMT

Or something.

My stomach has finally settled down, and I have something like an appetite again. (It is a twisted appetite, but that's a journal entry for another day.) As a result, I finally made it to Berlin. I saw the parliament building and the WWII monument next to it, as well as one of the spots where the Berlin wall used to stand. (We drove over it.) I also visited the CCC, or as I have come to think of it, the Club Mate warehouse with hacker space attached. I doubt I'll ever understand the hacker obsession with Club Mate, but while not understanding I will at least also be sleeping once in a while. I also got to experience Berlin's idea of a chicken burrito. While I'm sure germans are generally skilled in many arts, the art of the burrito is not one of them.

We got back from the CCC last night around 3AM. Jake's flight to Bangalore was at 9AM. By the time he arrives in India I think he'll be a zombie. I went back to the CCC in Berlin again today to spend some time with Kartsen, and had my first brush with Germany's public transit rail system. It seems to work rather efficiently (all the trains were dead on time), but god damn does it get cold after dark. The trains are of course all heated. But the platforms are not. (This one aspect at least they have in common with the NY subway.)

I have one more day tomorrow during which I could do some additional sightseeing, albeit by myself. (Karsten had offered to accompany me at one point, but he seems to have become bogged down with work.) I had considered trying to buy something nice for Tabby, but I'm not sure what I'll be able to discover just wandering around by my own, what with Berlin being pretty big and all. My flight back leaves on Monday at 1PM. I had hoped to return to work the next day, but I think I may have been overly ambitious. We shall see.

That's the signpost up ahead, your next stop: the TMI zone

Thu, 12 Feb 2009 04:30:12 GMT

The nice people running the conference were able to find me some Powerade, so that's one crisis averted. Unfortunately, now I have the runs. :( And my appetite is shot to hell. The rotten thing is that I don't know what set me off: I figure it has to be something I ate or drank, but not knowing what it was makes it hard to know to avoid it in the future.

On a ligher note, our talk went well, and the iPod demo went off without a hitch. One disappointing thing is that the talks being given by german presenters are being done in german. That's fine for the majority of attendees since they're also german, but since I don't speak the language it means many of the talks are of little value to me. (The slides, at least, are in english.)

I really hope my guts sort themselves out soon so I can enjoy at least a little time here. They're doing guided tours of the Castle Sanssouci tomorrow evening, but I'm not sure I'll be up to attending.

Wed, 11 Feb 2009 04:03:08 GMT

So, I'm in Potsdam, with Jake. We're supposed to do our talk at 2PM tomorrow. It's ohmygod o'clock right now, and I'm still awake. This is because a) my body thinks it's still early afternoon, and b) I think I'm dehydrated again. My heart is beating too fast, and it's preventing me from dozing off. I think I managed to sleep for a couple of hours, but I kept having the most bizarre dreams that kept waking me up.

I think the trip over is partly to blame. The flight from SF to Frankfurt started out ok, but the landing was terribly jumpy and bumpy. I got motion sickness and had to make a dash for the air-sick bag. I'm pretty sure this is the first time I've ever thrown up on a plane. (Luckily I didn't puke very much.)

Then our connecting flight from Frankfurt to Berlin was canceled due to bad weather. (The same bad weather that made the landing so bad.) Apparently the veather has been terrible all over Europe. Luckily there are hourly flights from Frankfurt to Berlin, so they moved us to the next one.

Finally, once in Berlin, we had to takr a 45 minute cab ride to the hotel. The hotel and the area are very nice, but it's really quite far from Berlin proper. And it's persistently rainy, as Germany tends to be, apparently.

What I really need right now is Gatorade, but something tells me that won't be easy to find in Germany. I already drank all the orange and apple juice in the minibar in my room, and I've been trying to drink water too, but my heart just won't slow down.

I would give anything for some sleep right now. :(

All systems are go

Wed, 04 Feb 2009 17:52:48 GMT

After having a great deal of trouble sleeping Saturday evening, and still not being free of the fever, I went to the hospital on Sunday. (Superbowl? What superbowl?) Apparently, in spite of trying really hard to drink as much as I could, I'd become dehydrated and my heart rate was elevated. They put two bags of IV fluids into me which slowed my heart down, and there was much rejoicing.

And peeing.

Lots and lots of peeing.

Anyway, I'm still a little behind on sleep, but otherwise I feel like my cranky old self again.