geek - brad's life — LiveJournal

geek - brad's life — LiveJournal

Brad Fitzpatrick

[	website	\|	bradfitz.com	]
[	userinfo	\|	livejournal userinfo	]
[	archive	\|	journal archive	]

geek	[Feb. 4th, 2004\|10:37 am] Brad Fitzpatrick
Got access to an IA64 system for testing. (Also soon an Opteron system). memcached runs on it no problem... cool. Address space so big! New mono 0.30 is out, including my async socket fixes! Look at these props to me from the release notes: Roughly 60 developers following developers contributed to this release: .... [not me] ..... This list is not complete, it is missing contributions that were sent to the list, as it was very hard to track the one million incremental line patch since the last release. I'm some of those millions! :P sniffle I feel so proud. Building a PBX at work with Asterisk. Learned about FXO vs. FXS signalling. Bought a 1-port FXO card and a 4-port FXS card. The FXS card needs 12V power from the power supply, but these old Dells don't have free cables dangling around like most home-built computers. Need to wait for my local parts store to open to buy an extender so I can get power to the PCI card. Since we only have 4 analog phones currently, we'll stick with this 4-port card and get IP phones for everybody else, which are only $65 and well-supported by Asterisk. Then everybody will be able to dial 9 to get the outside line (local calls only) or dial 8 or something to get an outside long-distance line, using VoicePulse Connect, which is pretty much the coolest commercial service ever. Been trying to figure out a bug we're seeing in the Linux e1000 driver. The newer intel networking cards do TCP Segmentation Offloading, but the driver is racy when it runs out of Tx descriptors, I read? There are options to turn it off or increase Tx resources, so I have to tackle that sometime here. Any "fix" is just a workaround it seems. The driver should be fixed. :-(
Link	Reply

Comments:

From:

taral
2004-02-04 11:38 am (UTC)

So fix the driver?

(Reply) (Thread)

From:

brad
2004-02-04 11:41 am (UTC)

Re:

Because I know what I'm doing, or what the problem even is?

It seems that if anybody knew what the problem was, the maintainers would fix it.

(Reply) (Parent) (Thread)

From:

taral
2004-02-04 12:35 pm (UTC)

Re:

Oh, that's not necessarily the case, trust me. :)

(Reply) (Parent) (Thread)

From:

mulix
2004-02-04 11:42 am (UTC)

You might want to try akpm's -mm tree, which has IIRC updated e1000 drivers from Intel. Or just grab them directly from Intel and give them a spin. I can hunt down the link if you want.

The PBX thingy sounds pretty cool, too.

(Reply) (Thread)

From:

brad
2004-02-04 11:59 am (UTC)

Re:

I diffed the intel version last night with the latest 2.6 (vanilla) and saw nothing notable. I'll check out -mm, though, thanks....

(Reply) (Parent) (Thread)

From:

mulix
2004-02-04 12:32 pm (UTC)

Re:

ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc3/2.6.2-rc3-mm1/broken-out/bk-netdev.patch, look for the e1000 changes.

(Reply) (Parent) (Thread)

From:

brad
2004-02-04 12:12 pm (UTC)

Re:

Sure enough!

* o Bug fix: hang under heavy Tx stress when running out of Tx
* descriptors; wasn't clearing context descriptor when backing
* out of send because of no-resource condition.

2.6.2-rc3-mm1/broken-out/bk-netdev.patch.

Thanks!

(Reply) (Parent) (Thread)

From:

mulix
2004-02-04 12:34 pm (UTC)

Re:

sure, my pleasure :-)
I jsut saw on the commits mailing list that Linus applied some e1000 patches, so you could grab tonight's bk snapshot if you don't want to run -mm (which has tons of not-quite-stable goodies, e.g. 4/4 split). Or just update drivers/net/e1000/* by hand.

(Reply) (Parent) (Thread)

From:

krow
2004-02-04 12:30 pm (UTC)

Which IP phones are you thinking about getting?

(Reply) (Thread)

From:

jzawodn
2004-02-14 06:16 pm (UTC)

Operton all the way...

Seriously. AMD got it right. They're wickedly fast. :-)

(Reply) (Thread)

Log in

Re:

Re:

Re:

Re:

Re:

Re:

Operton all the way...