A Penguin in Purple Clothing

Thursday February 10, 2005

Late night hacking at openib workshop

A whole bunch of guys from Trilabs came to the openib workshop in Sonoma. In order to make the conference more productive, they brought some hardware. Well, with a little power, brains, and beer, anything can happen. They hijacked the bar in the hotel and grouped around for a all night hackathon. Did they ever get LinuxBIOS to initialize the VGA head? Well, sorta: if you forget to hook up the power to the card, X does come up, which it shouldn't. In any event, a fun night was had by all, including the HPC Microsoft guy who happened upon us and shared a beer.

(2005-02-10 18:21:42.0) Permalink Comments [1]

Wednesday February 09, 2005

openib.org conference

I am sitting at the openib developers conference in Sonoma, CA. A large collection of vendors and open source contributers are here talking about what has been done and what needs to be done to make openib world class. It has been a very helpful conference and there are some very good things coming. SDP will be checked into the tree probably Thursday. [uk]DAPL will be ready, license-wise, in about a month. Both iSER and SRP are in the works.

(2005-02-09 09:05:41.0) Permalink Comments [0]

Tuesday February 01, 2005

17 Bennington St.

My fiancee, Aubrey, and I recently bought a house. It still hasn't closed, but it has been a long road to where we are today.

We first saw the house back in November druing a public open house. 17 Bennington St. had been on the market since September, a long time in the bay area. When we went in, Aubrey fell in love with it immediately. Although I love Victorians, I was a bit more reserved having noticed the plethora of issues including water and termite damage that I would have to fix. The house had great curb appeal -- bright yellow with beautiful detail -- but the rest of it is old, rotten, and in need of much work. When Aubrey pulled me aside and said she wanted it, I gulped and made sure she was thinking clearly. She was.

I went over to the dejected listing agent and talked to him a bit about the property. He was a brutish Russian fellow who was very forthcoming with the situation: the house has been on the market a long time; it is a probate sale; there have been five offers all of them rejected; the seller is old and a bit nuts. In any event, we took down his name and number and left.

That was on Sunday, November 21. I called the listing agent on Monday and asked if the house was still available. He said it was, but that another offer was going to be presented on Tuesday, the next day. At that point, I called Aubrey and asked what she thought. She wanted to put an offer in as well. I spent that day calling banks and loan officers trying to get the funds ready for the ten percent needed as a cash offer (this is necessary in a probate sale). I also got a real estate agent of my own to represent us.

That was a mistake. Apparently, the listing agent had inferred by our conversations that he was representing both the sellers and us. This caused a bunch of consternation, but was eventually worked out. Why can't this all just be done on computers (like buying gas) without human intervention?

The other (sixth) offer was rejected on Tuesday before we even got ours in. But, we were higher than that one, and the seller's agent assured me it was going to be enough. Then, we got the call on Wednesday before Thanksgiving: they had accepted the offer and would sign it next Monday. We celebrated.

Next Monday rolled around and I didn't hear back from anybody. On that Tuesday, I called. There had been a screw up. The sellers hadn't signed in ink. There was some sort of language issue. They thought the seller had verbally agreed, but apparently not. The seller was an old Filipino woman who didn't speak much English. She wanted full price. Well, Aub and I were devastated. We had offered as much as we could afford and couldn't go any higher.

A week passed and the selling agent called my agent: if we could raise our offer just ten thousand dollars, the seller would take it. And the listing agent would reduce his commission that same amount so it would be a wash to us. That sounded great so we agreed to put in another offer. Once again, we went back to the agency to sign pages and pages of paperwork.

Then, we got another call: they rejected that offer. What?!?! In any event, we couldn't do any more, so we said goodbye to the whole situation.

After that, I did a little online digging and found out that the probate sale is only for 25% of the property. Plus, there were close to eight other people CC'ed on all the paperwork for this sale. I called up the agents to ask what the deal was and finally I learned the sordid truth: It turns out the man who died left the house to his wife, but he had lived a separate, secret life with another wife and set of kids. Everybody had a stake in this sale and it was a total mess.

Aubrey and I reluctantly let go of the house. We moped our way through xmas break.

When we got back, a few days into the new year, I got a call from my agent. Apparently the sellers had reconsidered. If both the seller and buying agents would reduce their commission, they would sell for only 5K over our original offer. Not to let 5K get in the way of our dream house, we agreed. The paperwork was drawn up, offer submitted, and it was finally accepted in writing.

But alas, this was a probate sale subject to court confirmation. We had to present our offer in front of a judge. And once our offer was tentatively accepted by the seller, it went into the public record. That meant, anybody who wanted to could bid just 5% over our offer and then come to the court and snipe the house from under us. We had to wait til January 24 for the court date.

On the 24th, I went down to the court house in downtown San Francisco and presented our offer. Luckily, there were no other offers on the table, and the judge approved the sale.

So, now we are in the 30 days between approval and closing. Let's hope there are no other glitches in ths already hectic process. And without further ado, here is a slide show of the money pit we bought in San Francisco.

(2005-02-01 19:39:28.0) Permalink Comments [0]

Thursday January 06, 2005

Scott McNealy poked fun of at CES
Here Conan O'Brien pokes fun of Scott McNealy at CES in Vegas last night

(2005-01-06 08:59:35.0) Permalink Comments [1]

Tuesday January 04, 2005

OpenIB in Linus's development tree
Great News! Dave Miller took the OpenIB rev 5 patches posted to netdev and lkml and put them into his netdev tree. Subsequently, Andrew Morton and Linus Torvalds pulled the patches into their trees. So, now OpenIB will be in 2.6.11 and is already in 2.6.10-mm1.

If you want the cutting edge, go grab a copy of Linus's bk tree or download this patch on top of 2.6.10.
(2005-01-04 10:44:05.0) Permalink Comments [0]

Thursday December 16, 2004

OpenIB ready for inclusion

The OpenIB base codebase has been submitted for inclusion into the 2.6.11 branch of the Linux kernel.

This is the basic gen2 tree that has been developed over the past year or so through an open source process involving national labs and many vendors including TopSpin, Voltaire, Intel, IBM, Infinicon, Mellanox, and Sun.

The gen2 tree includes:

• mthca driver (InfiniBand Host Channel Adapter)
  • Mellanox tavor support
  • Mellanox arbel support (in tavor compat mode)
• middle layer software (abstraction of VAPI)
• One upper layer protocol: IPoIB

A discussion is underway both on the Linux Kernel Mailing List as well as on the OpenIB mailing list. If you have anything to say, please speak up. (2004-12-16 08:13:46.0) Permalink

Thursday November 18, 2004

OpenIB news

Two good news items on the OpenIB front:

1. OpenIB gets funding from the Department of Energy. http://news.zdnet.com/2100-9593_22-5446887.html
2. The openib gen2 codebase is cleaned up and ready to be submitted to the Linux kernel mailing list next Monday. Here is Roland's post on this matter.

Tuesday November 02, 2004

VOTE!

I know it is probably close to closing time at your polling place, but if you haven't already voted, please go out and VOTE!!

Here are some pictures from my polling place in S.F. (2004-11-02 17:19:55.0) Permalink Comments [1]

Thursday October 07, 2004

Got the LSI card working

The problem with the card turned out to the be the firmware of the SAS drive from Fujitsu. Once that was updated, the drive loaded and I was able to play with it on Linux.

In other news, I went to the Palo Alto office of Sun to help bringup an Opteron board in anticipation of our own efforts. They had never tested 64-bit Linux on the system (only 32 bit Linux and 32 bit Windoze). Got that up and mostly working. Unfortunately, the link to the Palo Alto office is only over a T1, so getting the software loaded was taking forever. I will be going back next week to finish what I started.

I wanted to go today, but they didn't have a board for me to work on. It seems that as soon as they get a board working and debugged, it is snatched up for somebody in the company and leaves their office.

I am off tomorrow with my finace to go to Massachusettes and Maine to scope out a place to potentially get married next summer. I will be back on Tuesday. (2004-10-07 13:21:37.0) Permalink

Thursday September 30, 2004

No luck on the LSI SAS 1064

Yesterday, I wanted to test the status of hotplug under Linux. I had a beta board from LSI that is a Serial Access SCSI card. This is a PCI 3.3 volt board, so it didn't go into the 5 volt boards from Nvidia. I reinstalled an old Tyan system that was laying in pieces in the lab that has PCI-X 3.3 slots. I slapped Fedora Core 2 on there, latest kernel updates, but it didn't recognize the card. I figured it wouldn't because this is still early access silicon. So, I checked in the kernel for which driver might be close to work on this card.

The mpt driver is from LSI and is in drivers/message/fusion. I tried to load that driver, but it did not attach to the card. I figured I may need an even more up to date kernel. So, I compiled 2.6.9-rc2 and put that on there. No luck.

I decided I might have to dive into the driver, so I started to look at the code. Lo and behold, it was not attaching to PCI id 0x0050 which was the LSI card. I added in that, plus another line to identifiy the 1064 card in the kernel messages. Compiled, installed, and the driver attached! But, it did not see any drives.

This morning, I got an email from the LSI representative with the latest version of their driver, 3.02.04. I was using 3.01.16 which is the one in 2.6.9-rc2. I compiled this new driver and installed it. It attached without issue to the LSI card, but I *still* don't see any drives. Arg.

So, I still don't have the ability to test hotplug on this SAS card. Maybe I can get a better dialog going with the LSI engineers.

(2004-09-30 11:14:44.0) Permalink

Thursday September 23, 2004

C'mon, Try Firefox

If you haven't already switched from using Internut Exploder, you should. Give Firefox a try. This is a kickass browswer. Tabbed browsing, ad blocking, popup blocking, and much much more. Seriously. Don't be scared to install new things. Iz gooooood.

(2004-09-23 13:05:37.0) Permalink Comments [2]

Tuesday September 21, 2004

Debugging sucks (sometimes)

I spent all day trying to figure out this problem. I was trying to redo the threading in the OpenIB IPoIB code to use kthreads instead of a Topspin home grown method. And in fact, I think I almost have it. It is just that there seems to be some weirdness still going on. If my client cannot join the multicast group, it behaves fine. It backs off according to the exponential algorithm and dies properly when the interface is brought down.

My problem seems to stem from when the interface actually works ok. If I try to bring it up and it can join the mulitcast group, it does it in a two stage process. The first stage works fine, it can come up and join the group. It is when it goes to the second part that it fails. It locks up the console. I can log in via ssh on another screen, but the console locks hard. I don't have a fresh enough brain to figure this out tonight. Maybe tomorrow with more coffee, I will be albe to diagnose the problem.

So, goodnight folks, I am s00per frustrated.

(2004-09-21 22:09:50.0) Permalink

Monday September 20, 2004

IPoIB Works!

Woot!!!

[root@nisus ~]# uname -a
SunOS nisus.SFBay.Sun.COM 5.10 s10_64 sun4u sparc SUNW,Sun-Fire-280R
[root@nisus ~]# netstat -nr
 
Routing Table: IPv4
  Destination           Gateway           Flags  Ref   Use   Interface
-------------------- -------------------- ----- ----- ------ ---------
192.168.0.0          192.168.0.78         U         1      6  ibd1
10.0.0.0             10.6.98.78           U         1  87102  eri0
224.0.0.0            10.6.98.78           U         1      0  eri0
default              10.6.98.1            UG        1      1  eri0
127.0.0.1            127.0.0.1            UH        8 292360  lo0
[root@nisus ~]# ping 192.168.0.233
192.168.0.233 is alive

[root@sins-stinger-10 root]# uname -a
Linux sins-stinger-10 2.6.9-rc2openib #3 SMP Mon Sep 13 10:40:38 PDT 2004 x86_64 x86_64 x86_64 GNU/Linux
[root@sins-stinger-10 root]# ping 192.168.0.78
PING 192.168.0.78 (192.168.0.78) 56(84) bytes of data.
64 bytes from 192.168.0.78: icmp_seq=0 ttl=255 time=1.21 ms
64 bytes from 192.168.0.78: icmp_seq=1 ttl=255 time=0.130 ms
64 bytes from 192.168.0.78: icmp_seq=2 ttl=255 time=0.125 ms
 
--- 192.168.0.78 ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 2002ms
rtt min/avg/max/mdev = 0.125/0.488/1.210/0.510 ms, pipe 2
[root@sins-stinger-10 root]# netstat -nr
Kernel IP routing table
Destination     Gateway         Genmask         Flags   MSS Window  irtt Iface
10.6.98.0       0.0.0.0         255.255.255.0   U         0 0          0 eth0
192.168.0.0     0.0.0.0         255.255.255.0   U         0 0          0 ib0.8001
169.254.0.0     0.0.0.0         255.255.0.0     U         0 0          0 eth0
0.0.0.0         10.6.98.1       0.0.0.0         UG        0 0          0 eth0

So, what does all this mean? Well, it means that Linux/OpenIB and Solaris are configured on the same IB network and are able to talk to each other vi IPoIB. That is pretty cool. Thanks to Roland and Jeremy for all the help to make this first important compatibility step happen.

(2004-09-20 14:46:40.0) Permalink Comments [1]

Friday September 17, 2004

I won a DVD player

Ha! Who would have thunk it. I don't think I have won any thing ever in my life. So, on Thursday, Xerox came to the Sun cafeteria to show off their warez and services. They had a little drawing that you could put your name into for a DVD player. And guess what? I won it. Cool. Although getting a $30 Apex AD-1110WM isn't the most exciting thing in the world, I did want to get a DVD player for my downstairs guest bedroom. So, now I have one. Coolio.

(2004-09-17 14:00:00.0) Permalink

Thursday September 16, 2004

IBSRM

Today, I fixed a bug with the OpenIB IPoIB implementation (this is the module that allows you to run an IP network on top of an Infiniband network). There was an issue with /proc file creation that was b0rking unloading of the module. Roland has been also actively fixing up ipoib and it is getting pretty good.

Also, I got a couple of Arbel cards in finally after a mixup with the first order. Arbel is a PCI Express Infiniband card. Since I needed to bring down my brinup system to install the card, it gave me a good opportunity to upgrade the Opteron's from 244s to 248s bringing them both from 1.8Ghz to 2.2Ghz. Q had already put the faster 3200 memory in there, so now I have a rocking machine for sure. With big, honking heat sings and fans on their, it looks pretty cool too. I should take a picture and put it up here...

The picture here is a graphical representation of the IB network. The big blue dot is a sleipner IB switch. The red ones are host nodes. Even though there is nothing indicating what is what, the one with the SM on it is a Sparc64/Solaris box running IBSRM (Sun's subnet manager). The other red dots are Linux systems with openib on them. One is an Opteron system. The other a Sparc64/Linux box as mentioned previously.

(2004-09-16 16:25:50.0) Permalink