thm 2005-08-15 Summary

Note: This summary only contains what I thought to be significant, interesting and/or funny. -- Hyperion

Original agenda:

9P RELIABILITY PROJECT STATUS -- ERICVH

  • Gorka has completed his initial port of Russ' recover application to the current distribution. It allows one to recover a session that is lost due to network error, server reboot, etc.
  • It also has some functionality to support failing over to alternate servers (assuming some sort of mirroring of resources) - but we haven't tested that much.
  • It works under both Plan 9 native and Plan 9 Ports. We are currently putting the finishing touches on a paper describing how it works and then we'll release to sources and p9p. Downside is it incurs about a 2x performance penalty on loopback under Plan 9 and 8x penalty under p9p on Linux. This overhead is likely less visible on higher latency connections than loopback. The only way around the overhead would be to push the recovery code inside the kernel instead of providing it as a user application.
  • Axel asked if the paper discusses the resason of the performance penalty? Eric said it should and the reason is copy overhead. The extra overhead on p9p under Linux is inefficient threads.
  • Eric said it's actually an interposer for 9p. So instead of
srv sources.cs.bell-labs.com
you
recover sources.cs.bell-labs.com
mount /srv/sources.cs.bell-labs.com
  • Uriel asked what was required for having a fallback server. Eric said that for static file systems it could be some sort of synchronous mirror - Plan B has some facilities for this, but we haven't tried using them yet.

PROPAGANDA EFFORTS AND DOCUMENTATION ISSUES -- URIEL

  • Would be good to have some input on what direction to take with the docs. So far I was trying to put as much stuff on the wiki as I can, so at least nothing is lost.
  • Vester added that openness on best practices would be good too.
  • Uriel would like to setup a mirror of all papers, ideally on sources. He also said that having a reliable sources web interface would be nice, not being sure if linking to 9grid.de from the main wiki is a great idea, mirrors are good, but not as reliable and uptodate as sources itself. If sources was exported over the web from it's main location it would be great.
  • Zimba asked if there were a list of who is responsible for what. Uriel explained that there isn't and that everyone is free to work on what they like, coordination has been traditionally thur 9fans, if you want to work on something, just do it, don't wait for anyone to give you 'permision'.

JIM GETTY'S SNAP VISION -- ERICVH

  • Jim Getty's is pushing his vision of the way the universe should look, particularly in the home network environment. If you look at his OLS paper http://google.com/search?q=getty's+snap, you'll see it looks somewhat familiar - very much like the Inferno/Plan 9 vision, at least on the high level. Jim seemed to be very open to Plan 9 style ideas, so it may not be a bad carrier to push Plan 9 ideas. Plan 9 technology seems to provide a lot more of what he wants than the alternatives he mentions in his paper(s). Getty's was pushing "lots of devices talking to eachother".
  • Charles Forsyth remarked: He is Jim Gettys, so that to get him interested you need to show him something he thinks he can't do or prove to him you can do it better.

PLAN B -- GORKA

  • We are probably packaging an image with everything for September. So people can look at it an maybe merge whatever they like into Plan 9. Our idea was to get the good ideas in it into sources. It is running on a modified version of the kernel, but there have been many strategy changes. The most changes which make it to sources, the less we have to maintain.

FUTURE 9CONS -- URIEL

ANNOUNCE AND ADOPT AN OPEN SOURCE MODEL WITH AN OPEN CORE TEAM -- VT3

  • Vester asked what do we want to do as a community?
  • Hyperion thought we'd all like to see such a model, but it depended on the Bell Labs developers.
  • Uriel warned that many people are happy with the status quo; Including Andrey, Eric, Axel, Gorka.
  • 20h remarked we have a half open source model, as you can take out the distribution and publish it on your own and that noone complained about the differences of the 9grid.de .isos.
  • Charles said the best thing to show Plan9 is not-dead is to apply it more, esp. to do interesting things (or boring things in an interesting way). That's really why Linux surpassed BSD: hardly because it was better, but programmers wrote for Linux. And to be most accurate, universities supported it.
  • Axel wondered if it's about marketing. Vester and XTL agreed it is.
  • 20h thought about having something like User Mode Linux for Plan 9.
  • Vester suggested if we want to stand out from other OSes then add user auditing features. This is what corporations do to standard Windows and *ix machines.

PLAN 9 PORT TO THE MAC MINI -- DAVID ECKHARDT (CMU)

  • What the students have runs and has run on a mini. There are N pieces of 1-person-sized work. There is a relevance issue, though, given that G5's don't have BATs and that Apple is ditching PPC.

INSTALLER PLANS/WORK -- URIEL

  • Uriel thinks that there is need to unify the install/bootabe-system/floppy to make testing easier, right now there are four slightly different kernel images in one part or another of the install system, and depending which one you end up with some things work or not.
  • 20h said one of the problem is that the kernel configuration specifies where to boot from.
  • Vester said network installs don't work correctly.

AFTER-CHAOS

PS

  • 20h's tip:
htfilefs -m /tmp http://www.9grid.de/plan9/plan9.iso; 9660srv; mount /srv/9660 /n/cdrom /tmp/plan9.iso;

UNIX/LINUX VERSION OF AUTH SERVER -- TONY_T

  • Skipped becuase tony_t were absent