BSDCon 2003 FreeBSD Developer Summit

BSDCon 2003 FreeBSD Developer Summit

The fifth FreeBSD Developer Summit was held on September 13, 2003, at Vicor's offices in Richmond, CA, immediately following BSDCon 2003 in San Mateo, CA. The FreeBSD Developer Summit was sponsored by DARPA, NAI Labs, and Vicor. Notes were taken by Bruce Mah and were edited by Murray Stokely.

Meeting began at 10:00am, ended at 5:00pm.

Agenda

Introduction / Summary of FreeBSD BoF / Growth of Project
Release Engineering, RELENG_4
Release Engineering, HEAD

Attending:

In person:

Doug Barton
David O'Brien
Matthew Dillon
Julian Elischer
Poul-Henning Kamp
Greg Lehey
Scott Long
Warner Losh
Bruce Mah
Jun Kuriyama
Jonathon Mini
Masafumi Nakane
Jeff Roberson
Hiroki Sato
Gregory Shapiro
Murray Stokely
Jacques Vidrine
Robert Watson

On The Phone:

The meeting followed a format where each section was led by an individual and then a discussion ensued.

Opening Remarks - Robert Watson

Welcome to the fifth FreeBSD Developer Summit at BSDCon 2003. This event is sponsored by:

Defense Advanced Research Projects Agency (DARPA)
NAI Labs, the Security Research Division of Network Associates
Vicor

The sponsors covered the cost of the room, food, telephone access, etc.

Growth of the Project (from BoF)

Positive Netcraft numbers

HEAD activity maintained even when RELENG_3 came out.

As soon as RELENG_4 came out, RELENG_3 activity dropped dramatically.

Large spike in commits just before RELENG_5_0_0_RELEASE.

About half the number of committers are working on RELENG_4 compared to the HEAD. Implies that most commits to RELENG_4 really do go through HEAD first.

Slightly less than half of commits go into sys/

Perforce submission rates are about half that of rate in the main src/ CVS branch.

"RaidFrame fell victim to a disk error." --scottl

(Tentative schedule reviewed)

Release Engineering, RELENG_4 - Murray Stokely

Complaints about PAE stability. Definitely need to tweak 4.9 schedule. Need to get people access to hardware to check stability under high load. Really need stability for non-PAE case.

Seems like a 2-week delay. sam is holding changes for MFC, waiting for things to settle down. dg wants the PAE stuff ripped out. phk thinks we're going to delay more like 6 weeks before we have confidence in this.

Discussion

PHK : Could we put this in P4?

IMP : Problem is that non-PAE users (who are having problems) won't have motivation to use it if it's not in CVS.

rwatson : What's minimalist backout patch? We need to get more info. Are PAE developers going to have time to work on this? If not, we have to kick it out.

Two-week timeline to get more information for fixing it. Need to be able to reproduce it in a developer environment...need information on workload.

peter : Things that he and ps would run into this problem pretty quickly. (Want to get silby in touch with him.)

"I'll just tell Paul that the next snapshot he makes has an unknown bug causing instability and he'll just go crazy trying to fix it." --peter.

Release Engineering, HEAD - Scott Long

"Somebody get him some tissues so he can cry without shorting out his laptop." --peter.

scottl : amd64 ia64 sparc64 support.

Discussion

marcel : New uart(4) driver supports serial console on the HP boxes in the cluster. HP boxes with no ISA bus emulation now supported from CVS. Should now be able to release for all ia64 boxes from CVS.

imp : uart(4) driver encapsulates lots of different UART chips. puc(4) driver is for aggregating many things at one attachment point.

Could we replace sio(4) with uart(4)? sio(4) is more about squeezing out performance, uart(4) is more about flexibility. Low-end performance would suck, but that's probably not the target platform for 6.0.

phk wants minimum hardware specs for releases. grog says that the people in this room are not typical users. Much debate about this. rwatson: Defer this for 6.0 session.

5.X efforts need to concentrate on stability and performance.

imp : Need to tie up loose ends in supported hardware, etc.