[TUHS] Dynamics between BSD and Linux

George Michaelson ggm at algebras.org
Fri Feb 2 16:43:13 AEST 2018

the other odd thing here is that Kirk McKusick used to pop up at lots
of places teaching. He is enormously engaging and a good presenter and
people love to hear about Filesystem design and models of virtual
filesystem.. but at the end of the day if they don't run the BSD
kernel, or code, then its .. kind of off-beam. I bet he could still
fill a room of people, but I doubt the room would equate to the volume
of linux nodes run inside DC.

Bhyve is a really nice virtualization model. It has zero traction.
Docker is it. Docker on Bhyve was mooted and died. CePH is barely on
BSD, 3-5 years down the track. So thats almost the entire world of
Docker -> Kubernetes you can't run, and the data/science world of
giant distributed filestores you can't run. Sucks to be a BSD nerd at
this point, because nobody wants you.

I'm running iSCSI backed from NetAPP for ZFS via FreeBSD. The ZFS is
rock solid. the iSCSI sucks, and can't do sensible dual channel
bonding, and a bunch of things in my life might be easier if I moved
it to Linux.  I run another 10-15 nodes as FreeBSD and I have to run
one as Debian to get BBR to get data from intractably slow ssh/rsync
feeds, on BSD or.. use this one Linux node and get a 10x speedup. Its
very hard to ignore this as a real-world signal of failure. (rsync
over ssh is hell, and there are much better models but this one is
bankable with other people)

right now, as a BSD bigot in a technology company, I'm in a group of
one. Its lonely, its cold, and its going to be a long slow winter (to
quote bill murray from groundhog day, kinda)

On Fri, Feb 2, 2018 at 4:36 PM, George Michaelson <ggm at algebras.org> wrote:
> Van Jacobsen did most of his ?ESNET? funded work on TCP/IP on BSD. He
> was associated with Berkeley. When BSD fragmented and there were 3,
> getting traction for rapid adoption of IP stack changes became hard.
> He worked on his suite of Mbone tools for a while, then gave up.
> He popped up again around the time codel was beginning, with random
> early drop (RED) changes, and they made BSD. But, things began to be
> released in linux kernels faster. BBR was a killer moment: its out in
> all modern Linux. Its barely adopted in FreeBSD, and the issues around
> BBR, Codel, models are being chained into the floor by interminable
> discussions (I am possibly wrong here, but this is the sense I get)
> Three or four smart guys at Swinburne were working on a really nice
> agile stack for IP changes in BSD. THey lost funds and traction and
> the team is now at Netflix.
> FQ-Codel aimed to home routers. Home routers for some reason are linux
> down the line. I also don't understand this: NetBSD would have been a
> fine model for tiny memory footprint SoC but somehow, it just didn't
> work out that way. Likewise the PI. I know NetBSD works, but somehow,
> its not on the main release cycle of the PI people. Its sideline.
> NAS people still like BSD. I think a lot of Open Source NAS projects
> wanted better underlying Disk IO models than Linux had, and ZFS, but
> now ZFS is in Linux.. I think the writing might be on the wall there.
> Juniper was (I believe) very BSD friendly. I don't know now. Was.
> On Fri, Feb 2, 2018 at 4:29 PM, Warner Losh <imp at bsdimp.com> wrote:
>> On Thu, Feb 1, 2018 at 11:11 PM, Bakul Shah <bakul at bitblocks.com> wrote:
>>> On Thu, 01 Feb 2018 19:44:48 -0800 Larry McVoy <lm at mcvoy.com> wrote:
>>> Larry McVoy writes:
>>> >
>>> > It's a bummer because BSD brings minimalism to the table.  You can run
>>> > a BSD machine in 128MB and it works.  Hell, it used to work great in
>>> > 4MB.
>>> I think this is the crux of the issue. As a group gets bigger,
>>> minimalism is hard to maintain. To have a fighting chance you
>>> have inculcate new people in the same minimalism culture and
>>> that takes time. This puts a higher bar to entry.
>> Though even in the early days of Linux, it could run in a slightly smaller
>> footprint. It just grew more quickly than BSD, though retained a better way
>> to subset that let it retain much of the lower end that BSD had grown too
>> large for for many years.
>>> > The BSD stuff isn't being taken seriously because the BSD people aren't
>>> > interested in taking new people seriously.  Which is a shame because the
>>> > work that Netflix and other BSD people have done is really cool.
>>> If you think what BSD folks have done is cool, just join in.
>>> Why not ignore the personalities and the popularity contest.
>> Honestly, most of the heavy BSD contributors do just that. There's drama
>> here and there, but it's mostly away from larger contributors...  And so it
>> goes...
>>> Regardless of how we got here, the reality is that BSD at this
>>> point has a tiny footprint in the market.  Even Linux has a
>>> small footprint in the desktop + laptop market, compared to
>>> Windows and Mac. BSD isn't even counted separately any more
>>> there. In the server market Linux is basically it. In the
>>> cloud market it is mostly Linux (almost all of it, if you
>>> don't count Azure). In the Mobile+desktop+laptop market, other
>>> than Android, Linux is under 1%. BSD numbers are just in the
>>> noise.
>> Yet, according to Sandvine, Netflix serves 35% of peak internet traffic, all
>> from FreeBSD. Go figure :)
>>> The reality is that BSD just doesn't matter to most folks. The
>>> same with minimalism. So it goes. [And neither fact matters to
>>> me for my non-pay work.]
>> Warner

More information about the TUHS mailing list