Dragonfly BSD 3.2 Released 85
An anonymous reader writes "Dragonfly BSD recently announced the release of version 3.2 of their operating system. Improvements include: USB4BSD, a second-generation USB stack; merging of a GSoC project to provide CPU topology awareness to the scheduler, giving a nice boost for hyperthreading Intel CPUs; and last but not least, a new largely rewritten scheduler. Some background is in order for the last one. PostgreSQL 9.3 will move
from SysV shared memory to mmap for its shared memory needs. It turned
out that the switch much hurts its performance on the BSDs. Matthew Dillon was fast to respond with a search for bottlenecks and got the performance up to par with Linux."
Yes, but (Score:1)
Re: (Score:1)
THE ssl port bug. How many ssl port bugs have there been. You could easily find it.
Re: (Score:1)
It's not a bug, it's a feature.
Re: (Score:2)
No results found for "the ssl port bug".
https://www.google.com/search?q=the+ssl+port+bug&oq=the+ssl+port+bug [google.com]
Re: (Score:1)
I think the main point is the near linear scalability curve for both Scientific Linux 6.2 and DragonFlyBSD which shows how far DragonFly has gone in the last 2 years, the team of developers is tiny. Sure Linux performance might have improved 10% since then, but it might also only be 2% or -2% ON THIS TEST, while other BSDs are sucking the dirt.
Re: (Score:1)
Since the attached PDFs to his post contain the benchmarks he ran I'd say yes.
Re: (Score:1)
The plots are based off of 3-run averages, all on the same hardware.
A good portion of the tuning was done against this same hardware as a reference platform, doing the same test,
with consistent results (plus improvements) across the development cycle.
Too many dimwits (Score:1, Insightful)
16 posts and every fucking one of them is an anonymous fucking coward dipshit.
Re: (Score:2)
Re:Too many dimwits (Score:5, Insightful)
Because this site is a giant linux circlejerk. I remember the good old days when FreeBSD ran linux binaries faster than linux itself.
Re: (Score:1, Funny)
Re: (Score:2)
"Reality" only exists in our imagination.
Re: (Score:1)
Unless "reality" happens to coincide with reality.
my experience with dragonfly 3.0 (Score:5, Informative)
1) hammer is beyond awesome, it's inspired. you owe it to yourself to live with dragonfly for a week or two for this reason alone.
2) the whole thing is pretty freaking snappy compared to freebsd (which itself is no slowpoke)
3) hardware support is patchy, you might need to hang onto that old nic/sas/raid card
4) linux emulation is not supported when running a 64-bit dragonfly system
5) otherwise, it's just like any other sane BSD.
my job forces me to run binary only linux crap so 4) rules out a wholesale move to dragonfly, but IMHO it is the BSD of choice for anyone with enough platform independence to seriously consider a BSD in the first place.
drivers & DHCP? (Score:2)
Re: (Score:2)
Don't buy a shitty nic, and RTFM before buying hardware?
You know - kind of like how Linux still is for some hardware, and definately used to be?
Re: (Score:1)
Thanks, this sums it up quite well. Id like to add that swapcache is pretty awesome, too.
Also I think that DFBSD is one of the very few open source OS projects worth watching closely. They
not only innovate, but still manage to keep a clean bsd OS. And while being small they are a very nice
and helpful community.
Way to go DFBSD!
Re: (Score:2)
But the Linux people got btrfs so HAMMER and ZFS isn't all that much to long for any more.
Re: (Score:2)
while many features of ZFS (and Hammer) are included in btrfs too, Hammer (specifically Hammer 2) has design goals that go way beyond those of ZFS (I believe it is going to be a fully fledged cluster FS). So there is something to wait for.
And I am still hoping that theiy will pursue their single-system-image design goal eventually.
Re: (Score:1)
What does single system image design mean?
I only really followed DragonflyBSD early and I know their focus has been towards multi core and clusters from the beginning. I guess for me personally the point is rather that I thought ZFS was cool but I use a desktop machine and by now I could use btrfs and get about the same effect so by now Solaris, FreeBSD or Linux doesn't matter much for me in that regard. Or well, DragonflyBSD in this case :)
It's always interesting to read about it though. (And I think it wo
Re: (Score:2)
SSI would be when multiple connected computer running an instance of, say, dragonfly bsd, can act like a single (multi-user and multi-tasking) computer. Tasks will migrate to any processor core in the cluster (and ideally factor in the cost of migration over network).
Similar to what openmosix did for linux ages ago - and very different from the infamous beowulf cluster.
By the way - Hammer does not only do simple snapshots but it does them automatically in intervals. So if you use it with samba or nfs you ge
Re: (Score:1)
But ZFS snapshots every modification and got de-duplication to?
Though from people in say #pcbsd or whatever it's called it seem like the memory requirements for using de-dup is massive?
btrfs doesn't snapshot the whole time but just when you tell it to?
The idea for me was to switch this usb flash drive to btrfs since zypper doesn't have the undo feature of yum but it can do undos using snapshots if you use btrfs.
I guess the yum alternative may be safer for your data though.
Guess I'm going away from the subje
Re: (Score:2)
Hammer can do deduplication with minimal memory requirements. For example only 512MB ram would still give a responsive and fast system. Hammer deduplication doesn't take a hard hit on performance like ZFS does, as ZFS dedup data in realtime while Hammer does it with a CRON job.
Re: (Score:1)
On the other hand just think about how much better dd if=/dev/zero of=file bs=1048576 count=10240 would perform on ZFS?!
(Or maybe not, what do I know?)
Re: (Score:1)
Hammer linear write performance is pretty good, since it is somewhat akin to a log-structured filesystem.
Re: (Score:1)
Similar to what openmosix did for linux ages ago
No, similar to what OpenSSI did for Linux years ago [wikipedia.org].
openmosix was much less SSI [wikipedia.org].
Re: (Score:1)
The problem with anonymous cowards is that there really are no reason to answer because your answer may never be noticed anyway. You're just wasting your time.
And no, I don't have extensive knowledge of either. I know they both do snapshots for instance and I know btrfs perform well. Whatever former Sun or Oracle can find a specific setup where ZFS beats btrfs is beyond me but I've never really trusted Suns benchmarks anyway. They feel just as useful as Apples to me.
Re: (Score:2)
If you were to look at benchmarks from a few years ago, FreeBSD blew out DragonFly on PostgreSQL testing. They've made real ground here, but it was a lot of work and it was only evaluated here against one application. PostgreSQL runs very well on the same core as it's a per CPU per connection setup. I believe the results but I do question if all applications would improve this much. The results are also very specific to the number of concurrent connections because the scheduler win goes away if there ar
Ahh, the irony... (Score:2)
Re: (Score:1)
Do enlighten me - where can I read up on this (shmem on dragonflybsd being slower than SysVshm)?
Re: (Score:2)
sigh (Score:1)
Re: (Score:2)