Matt Dillon explained what he wants the port-creation experience to be for a port maintainer, using VFS, quoted below:
Continue reading “VFS and ports for maintainers”
Hiten Pandya mentioned that Robert Watson has set up the FXR website (FreeBSD/Linux Cross Reference) for BSD code. It’s a personal domain, so please don’t tax it at present.
Greg Lehey put up a paper and slides from his Singapore presentation, describing the FreeBSD-5 SMPng work, both in terms of events and code. It’s an interesting read, since this is the path avoided in DragonFly. (Greg Lehey has a 2001 USENIX paper on SMPng, too, which mentions much of the same content.)
Jeroen Ruigrok has fixed
On the freebsd-hackers mailing list, a slight flamewar erupted over discussion of checkpointing code from DragonFly. Amidst the dumbness that normally ensues in a flamewar, there’s some interesting descriptions on what work has been done/will be done on DragonFly. Check out the archive, mostly in the “FreeBSD mail list etiquette” thread. Matt Dillon’s posts here, here, and here are all info-packed.
Matt Dillon’s posted his slides from his talk at UCB Berkley, in OpenOffice format.
As part of another discussion, it’s been noted that trying to boot FreeBSD 5.1 and DragonFly from the same disk currently may not work if it’s UFS1, and definitely won’t if it’s UFS2.
Jeroen Ruigrok posted some links to papers about checkpointing and threads:
A User-level Checkpointing Library for POSIX Threads Programs:
Checkpointing and its applications:
The main DragonFly site has been fixed up, in part with some changes I’ve submitted. Whee! I contributed!
Kip Macy brought up the idea of moving FreeBSD’s ataNG into DragonFly; the consensus so far is to bring it in separately from the existing ata support, since ataNG’s not yet completely stable.
Thursday October 23, 306 Soda Hall 6 – 8 PM. Matt Dillon will be talking to whomever shows up to the CSUA meeting.
Hiten Pandya brought up futexes for possible inclusion in DragonFly, pointing at this page for reference.
Bill Huey brought up http://people.redhat.com/drepper/futex.pdf, with the caveat that futexes are used for 1:1 threading models, not M:N models such as DragonFly is using/will be using.
Jeroen Ruigrok posted a whole pile of links to routing information for anyone wanting to work on it:
TCP Vegas: End to End Congestion Avoidance on a Global Internet
Evaluation of TCP Vegas: Emulation and Experiment
TCP and Successive Fast Retransmits
Congestion Avoidance and Control
Performance Problems in BSD4.4 TCP
Comparison of Tahoe, Reno, and SACK
TCP Vegas: New Techniques for Congestion Detection and Avoidance
A Tree-based Packet Routing Table for Berkeley Unix
TCP Vegas Revisited
Analysis on TCP Vegas and TCP Reno
Scalable Timers for Soft State Protocols
End-to-End Internet Packet Dynamics
End-to-End Routing Behavior in the Internet
Routing with a Clue
IP Lookups using Multiway and Multicolumn Search
Fast and Scalable Layer Four Switching
High-Speed Policy-based Packet Forwarding Using Efficient
Multi-dimensional Range Matching
Fast Address Lookups using Controlled Prefix Expansion
Matt Dillon’s made a diary update.
The memory slab allocator is now on by default, and the old system is gone.
Man pages for pim(4) and multicast(4) for “Protocol Independent Multicasting”, are in.
Sysinstall has been modified with changes for NFS; you can pick NFSv3 (default) and TCP protocol (not default), now, along with some internal changes.
The SysV IPC regression test suite has been added, in
David P. Reese posted his 9th step into syscall separation, and noted that there remains:
- “37 more uses of stackgap allocations in the linux emulation code”
- “26 more 43bsd syscalls that require relocation”
Kip Macy’s checkpointing code has been committed; I’m pasting Matt Dillon’s post about it as there’s a lot of issues to consider.
For those of you late to the party, checkpointing allows you to “freeze” a copy of an application so that, in theory, you can restore the program to that running state at a later point in time. Useful, for instance, if you have a program that takes a long time to complete and you don’t want to have to restart from the beginning if there’s an interruption.
Continue reading “Checkpointing is in”
David Cuthbert brought up the idea of Doxygen headers for the source; Matt Dillon didn’t care for the idea, pointing at this for an example. Hiten Pandya noted that he’d like to have a separate by-hand handbook, for which Eirik Nygaard posted a possible example. Matthew Fuller added that he is working on a SGML application for library documentation as part of another project.
While noting that he has made a number of changes to the scheduler, Matt Dillon described a new tool called ‘wmake’. wmake allows you to run make in a subdirectory simulating a buildworld environment, without actually having to build world.
(quoted from his post, using libkvm for an example:)
Joshua Coombs has an extensive writeup of his ideas for routing, located at http://www.outofspec.com/routing/