Older blog entries for berend (starting at number 358)

Source upgrade from FreeBSD 6.4-STABLE to FreeBSD 8.1-STABLE completed. Only one issue though:

ad2: 152627MB <WDC WD1600BB-00GUA0 08.02D08> at
ata1-master UDMA100
GEOM: ad2: geometry does not match label (81h,63s !=
16h,63s).

This appears to be a harmless message and can supposedly be fixed with:

bsdlabel -e -A /dev/ad2

And it did. After manually fixing the heads, I could mount it. Ran an fsck:

fsck /dev/ad2e

to make sure nothing was screwed, and it wasn't.

Just finished reading a very beautiful article in CACM July 1010: x86-TSO: A Rigorous and Usable Programmer's Model for x86 Multiprocessors.

Very useful, very clear semantics.

Very weird issue. I was on the Vodafone NZ Red network and had set my MTU to 1492 as that supposedly is the optimal size for pppoa. I had to switch to the Vodafone wholesale network in order to get ownership of my phone number so I could transfer it to a voip system.

After the transfer I had very weird connection issues, for example https would only work up to a point, i.e. I could make the ssl connection, send the HTTP request, but the response would never arrive. Or ssh to a server, but when I typed ls the reply would never come back.

Ultimately my issues are described here. MTU Path Discovery is broken. I suspect Vodafone's wholesale network might be blocking things, but don't know for sure. Could be my firewall for all I know. When I change the MTU on the clients to 1492 everything works. But that's annoying, so set the MTU on the modem back to 1500.

Another weird thing on Linux is that I can't do MTU Path discovery. For example:

ping -s 10000 www.dslreports.com

works fine. Setting the DF bit:

ping -M do -s 1472 www.dslreports.com

gives me the biggest packet I can sent. Which I don't really understand because as I understood it the biggest packet on ppoa is slightly smaller, i.e. 1478. But I suppose with ATM cells they just carry the MTU and don't care what the real limit is.

Anyway, drove me crazy for hours.

Installing pfSense on a Gigabit motherboard. Core dump with ohci_add_done. Seems to have been present since 7.0 and this is FreeBSD 7.2. Not nice. You have to unplug the USB keyboard and then FreeBSD will boot.

Had an issue with OpenOffice 3 not wanting to open files on an nfs mounted drive. It appeared that my FreeBSD NFS server wasn't running the lock and stat daemons. After that it was perfect.

26 May 2010 (updated 26 May 2010 at 20:29 UTC) »

Still having trouble with my new firewall rules. Sometimes web pages take a long time to load (up to a minute), just looks like connections open very slowly or packets travel slowly.

Not sure what this is. Have no increased my network buffers and max sockets, which did help with my network speed. I can now transfer 20MB/s on my Gigabyte lan. Far short from the supposed 100MB/s I should get, I know. But currently don't have the time to figure out why I don't get that speed.

Also played with turning off net.inet.ip.dummynet.io_fast, that might have effect as well. Currently it's on again, so let's see how it behaves. Those intermittent lags are quite noticeable: you click on a link, and nothing happens, so let's see if I get that today or not.

Maybe I should upgrade to FreeBSD 8 as I'm still on FreeBSD 6.4. Could do in-kernel nat as well.

Our school also needs better internet, so I'm thinking of deploying a pfsense box with a cheap ADSL modem. I could do traffic shaping with lowest bandwidth for the students and higher priority for staff and voip. And of course have Squid, so browsing experience should become much better.

15 May 2010 (updated 15 May 2010 at 07:24 UTC) »

Spent an entire day trying to get traffic shaping working with FreeBSD. Had to recompile kernel to allow traffic shaping. Next good experiments in actually limiting rates.

But hit a road block to see how I could limit only outgoing packets. Started to play with my firewall settings and got nowhere. In the end I decided to start afresh with the one from the FreeBSD manual as I understood I might have issues between the dynamic rules and nat. Anyway, now back to exactly where I was (I hope I didn't break anything), and my firewall script has become a bit leaner, and a bit clearer.

Installed the 2talk voip client on Windows, which worked out of the box. On Ubuntu Empathy and Ekiga got me nowhere. Installed a STUN daemon, but that didn't help.

Installed twinkle and that worked out of the box.

Basically a day where I only learned what doesn't work.

To continue with my VOIP musings, one of the things I'm looking into is using FreeBSD's traffic shaping. Which is ridiculously easy.

Tomorrow my throttle comes off, so I will be trying Ekiga.

I have an ISP that gives me only 40GB. While the rest of the world rolls over laughing, this is pretty much the norm in NZ. However, this provider won't give you anymore, but throttles you at 64kbps once the limit is reached.

That's mad especially when you get hit by a small issue eating 4GB in 4 hours...

So it's clear to me I need to have another ISP. And need to get my phone line unbundled so I'm free to swap ISPs when needed. Which means ditch the land line and go for VOIP. That's a bewildering landscape it appears.

After some checking out it appears I need to go with 2talk. After I had made that selection I talked to another company who were already with 2talk, unknown to me, although I knew they had VOIP.

So need to step through these things slowly. Would like to keep my existing DECT phones. In order to do that I need an adapter, which I selected as the Linksys SPA2102 and 2talk actually recommended that in a later email.

The award for the most confusing message I've seen in a quite a while. Added 2GB more to my Dell D820 laptop, so it now has 4GB. Boot. Then it says:

The amount of system memory has changed. If you did not change your memory... To resolve this issue, try to reseat the memory.

Hmmm. Take out the memory. Reboot. Message comes back. So what now? System screwed?

But Google is my friend.

349 older entries...

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!