bsdtalk202 - DragonFlyBSD 2.8 with Matthew Dillon

Also Available via Phone: +1 (210) 957-5481

This is an interview from day 2 of MeetBSD California.  He interviews Matthew Dillon about the recent Dragonfly BSD 2.8 release.

New Features include:

• Linux LVM (most of it)
• NetBSD DM ( Disk Manager / Storage Manager) which includes striping and crypto
• GUI release image working for this release
• A lot of stability work.
• A lot of Multi-Processor improvements
• Simplified the kernel framework to use kqueue everywhere
• Hammer Filesystem is doing well.
• swap cache improvements.

A GSoc project for dedupe for hammer is going into the code this week.
Software crypto was only using a single CPU, but in this release they did a bunch of work on the opencrypto in the kernel to make it Multi-Processor aware.

Crypto options for the installer are going into the codebase shortly.  The entire Hammer partition will be able to be encrypted.  Some VM adjustments after the release fix a heavily loaded machine working in low memory.  They just went it to the code.

Clustered Single Instance System
This is still a goal but they have backed off the SSI system for now. What they are doing is the clustered, multi-master filesystem and cache coherency.  This is a much more doable goal.  The focus recently has been on Multi-Processor work.  They are almost done removing the Giant Lock.  Nearly the entire backend is MP safe.  All the network stack, the entire IO path is MP safe through Hammer.  The front end is using token locks.  About half of them are actually still getting locks.  

The removal of the locks from the back end has improved performance a lot.  The system is no longer single threaded.

NFS

They fixed a lot of issues with NFS.  The RPCs are now asynchronous.  The OS now dedicates 2 kernel threads.  One for reading and one for writing.  Performance has been very good.

Hammer

They now have a catastrophic recovery tool.  It can't recover in place, but it can restore a hammer drive to another disk.  It requires almost no knowledge of the disk topography, but it handles it all automatically.  This utility was written because a user needed to recover a disk that had read errors and there were no backups.   A couple of the errors were right in the middle of the Hammer filesystem. An fsck tool was just impossible for Hammer, but a recovery tool was very simple.  It was written in just a couple of days.  The user was able to recover 99% of his files.

Swapcache

A lot of people don't understand how this works and how important it is.  It works with a single regular harddrive and a 20G SSD drive.  Even a small SSD creates major improvements to the original disk.