If you’re using pf to control how your bandwidth is used, you may want to look at the recent fairq updates from Matthew Dillon.  It should perform better now in situations where one traffic group is saturating its available bandwidth.  Here’s a handy link that explains this sort of problem, yoinked from IRC.

Sascha Wildner has added in binutils version 2.21, replacing 2.20.  Note that both 2.20 and 2.21 are ‘optional’, so you have to set BINUTILSVER to use them.  Otherwise the system defaults to the also-installed binutils 2.17.

This is the BSD-licensed version of libiberty, which was removed because it didn’t ever actually make it to being a replacement.

Matthew Dillon’s improved bridging to the point where you can now modify the MAC address of the bridge and most everything, including ARP, will come from it correctly.   It’s even possible to bond 2 or more interfaces together, with the side effect of dragonflybsd.org having a lot more bandwidth.

Update: the config for his bonded interfaces has been posted as an example.

Update 2: More notes here.

Matthew Dillon has continued his bridge work, with another commit adding various features.  Go, read.

Matthew Dillon has added transparent bridging, mostly to overcome issues with the AT&T DSL modem he’s using.  With this non-default feature, IP packets retain the original MAC address when retransmitted through a new interface.

The tmpfs(5) filesystem now runs without multiprocessor locks.  Yay!  Another hurdle down.

ps now has a new option: -R.  This lists processes in order by parent/child status, and indents to make it visually clear.  It looks like this.  I wish someone had done this 15 years ago.

Matthew Dillon’s made fork() and exit() (mostly) multiprocessor-safe.  It’s only two calls, but both are very frequently used.  One more step closer to removing the Giant Lock…

Matthew Dillon’s made some scheduler changes, which blogbench tests are showing give the default scheduler better performance under heavy load.  It’s a pretty technical writeup, so I’ll just point you at it rather than attempt to summarize.

Matthew Dillon’s made some changes that will speed up the booting process for people with a ridiculous amount of memory, like 64G.  This is x86_64 only, but that should not be a surprise if you think about it.

Sepherosa Ziehau has been continuing his work on improved interrupt support.  I have no one central commit to point to, but rather a zillion bits of work on this and other stuff.  I feel it’s always good to recognize when someone’s volunteering a lot of work – Thanks, sephe!

Global tokens are now running without the giant lock in DragonFly.  Neat!  There’s still plenty more to remove, but this is a big step.

Matthew Dillon’s updated fsstress for x86_64, so if you have a 64-bit machine that needs a workout – here you go.

While we’re talking about file system improvments, Venkatesh Srinivas has been removing the Giant Lock from tmpfs.

Ilya Dryomov has added live deduplication, or as he titles it, “efficient cp”.  It’s experimental and turned on with a sysctl, so approach with caution.

Xerox Network Services is gone from DragonFly.   Does anyone, anywhere, use this protocol?  Ironically, I don’t recall this even being visible on the Xerox hardware products I have at work.

Matthew Dillon fixed a rare and difficult-to-find bug on x86-64 Dragonfly.  This means much more of the system can be run ‘MPSAFE’, or without the Giant Lock.  Watch for this soon if you’re running 2.9.

Venkatesh Srinivas has created what he calls “Super Light Weight Reference Counting”, which he describes in a recent post, plus followup. He’s already converted sfbuf to use it.

Peter Avalos has updated libarchive to version 2.8.4.  The commit message has details on what’s changed (for us).  This is good, since the libarchive site release notes seems to not be up to date.

Update: Peter helpfully pointed at contrib/libarchive/NEWS.

Peter Avalos has updated zlib to version 1.25, and appears to have done some work with tnftp, though this is the only message I saw.

Ironically, I get a “this site is using an unsupported form of compression” error when browsing to the zlib web site.

As part of the ongoing work to support a lot of CPUs, Matthew Dillon has made some changes that have the side effect of benefiting virtual kernels.  How much?  I don’t have a benchmark, yet.

Peter Avalos has updated OpenSSL to 1.0.0c, to fix a recent security problem.  The problem doesn’t sound too catastrophic to my untrained ear, at least.

Matthew Dillon has moved CPU support to 63 processors and 512G of RAM.  This may cause issues, he warns.  It’s also just barely working, so don’t expect to go into production with half a terabyte of RAM in the next few days.

APIC_IO is back as a kernel config option, though it just toggles the sysctl loader tunable default.  This is so a kernel config file with that option still set won’t cause an error.

Peter Avalos has committed his speedups for OpenSSL encryption (using assembly), along with a lot of numbers to show performance changes.  It’s definitely sped up, but the quantity of values is so large that you’d have to visualize it differently to get a summary I could show here.

Peter Avalos is bringing in OpenSSL 1.0.0b.  I’m not sure what the difference between 1.0.0b and 1.0.1 would be.  Also, Alex Hornung has updated vn(4) – there’s more updates than the one I linked.

For a long time, APIC_IO was an option that you usually had to enable for multiprocessor capability in your kernel config.  Michael Neumann fixed up and changed this so it’s no longer a kernel config option for i386 or x86_64, but rather a loader tunable: hw.apic_io_enable.

DragonFly has shipped with a uniprocessor kernel by default forever.  Shipping with a SMP kernel may not work as well for all possible combinations.  With some recent changes by Matthew Dillon, both types of kernel are present and can be picked from at boot time – with the LiveCD!

Venkatesh Srinivas has a large writeup describing just how the memory allocator in DragonFly changed from 2.6 to the upcoming 2.8 version.   In all that text, you may notice the cheering statistic that it gave a 20% improvement in sysbench results.

Rumko came up with some changes for vkernel(7) that, among other things, made it possible to run them diskless; almost like booting a thin workstation.

OpenSSL on DragonFly can now dynamically load engines (cryptographic support modules) at runtime, thanks to Peter Avalos.

Here’s something: Pratyush Kshirsagar came along, saw the proportional RSS project idea, and did it.  It’s nice to have a completed project just sort of fall out of the sky.

It looks like my prior article was incorrect: it was a different issue than MADT causing problems with booting x86_64.  Scrambled memory report appears to be at the heart of the issue; in any case, it’s fixed now.

Did you know Fred is the name for the DragonFly mascot?  Well, you do now.  He’s also back in the bootloader, thanks to Joe Talbott.

Two recent changes in the way virtual kernels are constructed should make a speed difference.  The startup time is reduced (and more memory can be given to the vkernel), and the overall running speed should be quicker, too.

DragonFly’s version of pf (corresponding with OpenBSD’s 4.2 version) is now multiprocessor safe, to match the network stack.  pf itself isn’t using multiple processors; it’s just able to work without causing problems in an otherwise MPSAFE environment, thanks again to Jan Lentfer.  Note that there’s one minor caveat.

Jan Lentfer has now updated pf in DragonFly to version 4.2, on top of his earlier work to get to 4.1.  This upgrade apparently doubles speed from 4.1, plus he’s brought in some other, later fixes.  Thanks for doing a superhuman amount of work, Jan!

A familiar procedure in any open source project: irritation causes improvement.  In this case, the Forth-based boot loader irritated Matthew Dillon into writing a new replacement C-based one.  (See the commit too, and it may slightly affect the upgrade process for 2.7 users.)

All these recent locking changes seem to be adding up to a much more responsive system, incidentally.

Good news: the upcoming 2.8 release (that’s next month!) of DragonFly will be missing the Big Giant Lock from a significant part of its structure, and will be removed completely somewhere in 2.9.  Recent commits bear this out.

Thanks to the efforts of Venkatesh Srinivas, tmpfs file systems on DragonFly can now withstand fsstress testing.  Thanks, Venkatesh!

(One of the benefits of posting about people’s work is that the names are fun to type.)

Matthew Dillon sent an email to the kernel list detailing the performance improvements that he and Alex Hornung have recently made to dm_crypt and opencrypto. The disk encryption work does still come with a warning, however.

A few more fixes were committed to master improving the standards conformance of the new poll implementation and fixing a couple of bugs.