[TUHS] UVM VM system

[TUHS] UVM VM system

[Larry McVoy]

So I just read this

https://www.usenix.org/legacy/event/usenix99/full_papers/cranor/cranor.pdf

and it looks encouraging.  Apparently NetBSD is using it.  Does anyone
know if they are happy with it?

Has FreeBSD considered this?

Has anyone benchmarked FreeBSD against NetBSD to see which is faster
for VM stuff?

Re: [TUHS] UVM VM system

[Warner Losh]

[-- Attachment #1: Type: text/plain, Size: 657 bytes --]

On Sat, Sep 1, 2018 at 12:51 PM Larry McVoy <lm@mcvoy.com> wrote:

> So I just read this
>
> https://www.usenix.org/legacy/event/usenix99/full_papers/cranor/cranor.pdf
>
> and it looks encouraging.  Apparently NetBSD is using it.  Does anyone
> know if they are happy with it?
>

They are relatively happy with it...


> Has FreeBSD considered this?
>

Yes, but it would be a huge porting effort.


> Has anyone benchmarked FreeBSD against NetBSD to see which is faster
> for VM stuff?
>

Generally, the benchmarks favor FreeBSD. Again, it's a cleaner design, but
the time spent optimizing FreeBSD's and eliminating the bottle necks has
paid off...

Warner

[-- Attachment #2: Type: text/html, Size: 1400 bytes --]

Re: [TUHS] UVM VM system

[Clem Cole]

[-- Attachment #1: Type: text/plain, Size: 1832 bytes --]

On Sat, Sep 1, 2018 at 2:51 PM Larry McVoy <lm@mcvoy.com> wrote:

> So I just read this
>
> https://www.usenix.org/legacy/event/usenix99/full_papers/cranor/cranor.pdf

I remember that presentsation, it was exciting and very cool ;-)



>
>
> and it looks encouraging.  Apparently NetBSD is using it.

Hmm - integrated/used it at one time ... but ... I'm not sure of that is
still true of it it even really made it out.  That was all happening when I
was still hacking on Alphas (which was a long time ago).   We'd need the
active NetBSD folks to chime in on curent state.



>   Does anyone
>
> know if they are happy with it?
>
At the time (and I'm doing this by memory) the buffer cache stuff needed a
rewrite which I thought FreeBSD did/was doing at the time.
In those days, FreeBSD was within epislon on of Tru64 on Alpha performance
and NetBSD had a ways to go.  At the time, I gave a couple of Alphas to
somebody in the UK (I've forgotten whom); who was going to redo it.



>
> Has FreeBSD considered this?
>
Last I knew, no.  I was under the impression, the work FreeBSD did
rewriting the Mach stuff paid off for them at the time.  I have FreeBSD,
OpenBSD and Linux (and Mac OSx) all running on my systems here.   But the
problem is that the HW is all over the map in termns of release date, so
I'm not sure which is faster at this point.   The *BSD systems are the
easiest to admin and clean/simplest (which is why they only systems I have
exposed is an OpenBSD box).  But they have uses ;-)



>
> Has anyone benchmarked FreeBSD against NetBSD to see which is faster
> for VM stuff?
>

My data was from those days, and FreeBSD was winning, but thats a >>long<<
time ago.  Lots of bits have been types into to the kernel of both systems,
so you tell me,

Clem

ᐧ

[-- Attachment #2: Type: text/html, Size: 4143 bytes --]

Re: [TUHS] UVM VM system

[ron minnich]

[-- Attachment #1: Type: text/plain, Size: 575 bytes --]

I was his advisor on that thesis so I got to watch it roll out as it
happened.

uvm replaced the machvm in netbsd.

For a time, Chuck set it up to run in parallel with the existing VM. You
could start a process and pick which vm it used. For a while, it defaulted
to the existing one. Then, at some point, it defaulted to uvm. Then, at
some point, the old one was removed.

more here:

http://www.netbsd.org/docs/kernel/uvm.html

via search terms
uvm replaces machvm netbsd

chuck was a long time contributor to netbsd IIRC, but last time we talked,
he was using Linux.

ron

[-- Attachment #2: Type: text/html, Size: 874 bytes --]

Re: [TUHS] UVM VM system

[ron minnich]

[-- Attachment #1: Type: text/plain, Size: 969 bytes --]

I said that wrong, I was not his advisor, Guru was;I was the external
committee member or external advisor or some such thing.  Chuck and I
talked an awful lot because I was doing a lot in the sunos and AIX VMs at
the time, and Chuck and I had known each other from udel days.



On Sat, Sep 1, 2018 at 12:41 PM ron minnich <rminnich@gmail.com> wrote:

> I was his advisor on that thesis so I got to watch it roll out as it
> happened.
>
> uvm replaced the machvm in netbsd.
>
> For a time, Chuck set it up to run in parallel with the existing VM. You
> could start a process and pick which vm it used. For a while, it defaulted
> to the existing one. Then, at some point, it defaulted to uvm. Then, at
> some point, the old one was removed.
>
> more here:
>
> http://www.netbsd.org/docs/kernel/uvm.html
>
> via search terms
> uvm replaces machvm netbsd
>
> chuck was a long time contributor to netbsd IIRC, but last time we talked,
> he was using Linux.
>
> ron
>
>
>

[-- Attachment #2: Type: text/html, Size: 1531 bytes --]

Re: [TUHS] UVM VM system

[Brad Spencer]

Clem Cole <clemc@ccc.com> writes:

[snip]

>> and it looks encouraging.  Apparently NetBSD is using it.
>
> Hmm - integrated/used it at one time ... but ... I'm not sure of that is
> still true of it it even really made it out.  That was all happening when I
> was still hacking on Alphas (which was a long time ago).   We'd need the
> active NetBSD folks to chime in on curent state.
>

[snip]




Yes, UVM has been used with NetBSD for a very long time.  The browseable
CVS repo at www.netbsd.org suggests that work started in 1997 - 1998 or
so.  Looks like in 1999 the Mach VM was completely removed from NetBSD
except for some include files, but clean up appears to have continued
into 2000.



-- 
Brad Spencer - brad@anduin.eldar.org - KC8VKS - http://anduin.eldar.org

Re: [TUHS] UVM VM system

[Warner Losh]

[-- Attachment #1: Type: text/plain, Size: 846 bytes --]

On Sat, Sep 1, 2018, 1:42 PM ron minnich <rminnich@gmail.com> wrote:

> I was his advisor on that thesis so I got to watch it roll out as it
> happened.
>
> uvm replaced the machvm in netbsd.
>
> For a time, Chuck set it up to run in parallel with the existing VM. You
> could start a process and pick which vm it used. For a while, it defaulted
> to the existing one. Then, at some point, it defaulted to uvm. Then, at
> some point, the old one was removed.
>
> more here:
>
> http://www.netbsd.org/docs/kernel/uvm.html
>
> via search terms
> uvm replaces machvm netbsd
>
> chuck was a long time contributor to netbsd IIRC, but last time we talked,
> he was using Linux.
>

These days I know he's hacking on FreeBSD's storage stack with me at work
:). I think he's still a netbsd contributor. I see his name in the commit
log often..

Warner

>

[-- Attachment #2: Type: text/html, Size: 1608 bytes --]

Re: [TUHS] UVM VM system

[ron minnich]

[-- Attachment #1: Type: text/plain, Size: 968 bytes --]

good to know!

On Sat, Sep 1, 2018 at 1:26 PM Warner Losh <imp@bsdimp.com> wrote:

>
>
> On Sat, Sep 1, 2018, 1:42 PM ron minnich <rminnich@gmail.com> wrote:
>
>> I was his advisor on that thesis so I got to watch it roll out as it
>> happened.
>>
>> uvm replaced the machvm in netbsd.
>>
>> For a time, Chuck set it up to run in parallel with the existing VM. You
>> could start a process and pick which vm it used. For a while, it defaulted
>> to the existing one. Then, at some point, it defaulted to uvm. Then, at
>> some point, the old one was removed.
>>
>> more here:
>>
>> http://www.netbsd.org/docs/kernel/uvm.html
>>
>> via search terms
>> uvm replaces machvm netbsd
>>
>> chuck was a long time contributor to netbsd IIRC, but last time we
>> talked, he was using Linux.
>>
>
> These days I know he's hacking on FreeBSD's storage stack with me at work
> :). I think he's still a netbsd contributor. I see his name in the commit
> log often..
>
> Warner
>
>>

[-- Attachment #2: Type: text/html, Size: 1985 bytes --]

Re: [TUHS] UVM VM system

[Kevin Bowling]

Seems like Chuck Cranor is at CMU http://chuck.cranor.org/.  Chuck
Silvers is with you?

On Sat, Sep 1, 2018 at 1:25 PM, Warner Losh <imp@bsdimp.com> wrote:
>
>
> On Sat, Sep 1, 2018, 1:42 PM ron minnich <rminnich@gmail.com> wrote:
>>
>> I was his advisor on that thesis so I got to watch it roll out as it
>> happened.
>>
>> uvm replaced the machvm in netbsd.
>>
>> For a time, Chuck set it up to run in parallel with the existing VM. You
>> could start a process and pick which vm it used. For a while, it defaulted
>> to the existing one. Then, at some point, it defaulted to uvm. Then, at some
>> point, the old one was removed.
>>
>> more here:
>>
>> http://www.netbsd.org/docs/kernel/uvm.html
>>
>> via search terms
>> uvm replaces machvm netbsd
>>
>> chuck was a long time contributor to netbsd IIRC, but last time we talked,
>> he was using Linux.
>
>
> These days I know he's hacking on FreeBSD's storage stack with me at work
> :). I think he's still a netbsd contributor. I see his name in the commit
> log often..
>
> Warner

Re: [TUHS] UVM VM system

[Andy Kosela]

[-- Attachment #1: Type: text/plain, Size: 795 bytes --]

On Saturday, September 1, 2018, Clem Cole <clemc@ccc.com> wrote:
>
>
>
>>
>> Has FreeBSD considered this?
>>
> Last I knew, no.  I was under the impression, the work FreeBSD did
> rewriting the Mach stuff paid off for them at the time.  I have FreeBSD,
> OpenBSD and Linux (and Mac OSx) all running on my systems here.   But the
> problem is that the HW is all over the map in termns of release date, so
> I'm not sure which is faster at this point.   The *BSD systems are the
> easiest to admin and clean/simplest (which is why they only systems I have
> exposed is an OpenBSD box).
>

OpenBSD is also using uvm[1].  But these days it certainly differs from
NetBSD implementation as it was hacked on by different people during the
last several years.

[1] https://man.openbsd.org/uvm.9

--Andy

[-- Attachment #2: Type: text/html, Size: 1339 bytes --]

Re: [TUHS] UVM VM system

[Steve Mynott]

On Sun, Sep 02, 2018 at 08:29:42AM +0200, Andy Kosela typed:

> OpenBSD is also using uvm[1].  But these days it certainly differs from
> NetBSD implementation as it was hacked on by different people during the
> last several years.
> 
> [1] https://man.openbsd.org/uvm.9

Both forks now include a unified buffer cache.

There is an interesting series of blog posts at

<http://blog.pr4tt.com/2016/02/02/BSD-virtual-memory/>

The OpenBSD UVM has particularly diverged from the original with the
addition of a "dead entry queue" and the blog author complains about its
lack of documentation.  He also mentions an experimental RadixVM as
being current "state of the art" although its not available on any
mainstream systems.

-- 
Steve Mynott <steve.mynott@gmail.com>
cv25519/ECF8B611205B447E091246AF959E3D6197190DD5

Re: [TUHS] UVM VM system

[Warner Losh]

[-- Attachment #1: Type: text/plain, Size: 1293 bytes --]

On Sat, Sep 1, 2018, 11:19 PM Kevin Bowling <kevin.bowling@kev009.com>
wrote:

> Seems like Chuck Cranor is at CMU http://chuck.cranor.org/.  Chuck
> Silvers is with you?
>

Why I'm embarrassed to admit you are right. Chuck Silvers also did some VM
work, but not uvm.

Warner

On Sat, Sep 1, 2018 at 1:25 PM, Warner Losh <imp@bsdimp.com> wrote:
> >
> >
> > On Sat, Sep 1, 2018, 1:42 PM ron minnich <rminnich@gmail.com> wrote:
> >>
> >> I was his advisor on that thesis so I got to watch it roll out as it
> >> happened.
> >>
> >> uvm replaced the machvm in netbsd.
> >>
> >> For a time, Chuck set it up to run in parallel with the existing VM. You
> >> could start a process and pick which vm it used. For a while, it
> defaulted
> >> to the existing one. Then, at some point, it defaulted to uvm. Then, at
> some
> >> point, the old one was removed.
> >>
> >> more here:
> >>
> >> http://www.netbsd.org/docs/kernel/uvm.html
> >>
> >> via search terms
> >> uvm replaces machvm netbsd
> >>
> >> chuck was a long time contributor to netbsd IIRC, but last time we
> talked,
> >> he was using Linux.
> >
> >
> > These days I know he's hacking on FreeBSD's storage stack with me at work
> > :). I think he's still a netbsd contributor. I see his name in the commit
> > log often..
> >
> > Warner
>

[-- Attachment #2: Type: text/html, Size: 2438 bytes --]

Re: [TUHS] UVM VM system

[Will Senn]

[-- Attachment #1: Type: text/plain, Size: 1915 bytes --]

According to the Netbsd page:

Chuck Cranor designed and implemented UVM, Matthew Green handled integration issues and wrote the swap subsystem, Chuck Silvers wrote the anonymous memory object pager (which added support for shared memory), and various other developers have converted the appropriate ports across. Andrew Brown modified UVM to be able to do top down memory management.

It appears both Chuck’s contributed!

Will

Sent from my iPhone

> On Sep 2, 2018, at 9:55 AM, Warner Losh <imp@bsdimp.com> wrote:
> 
> 
> 
>> On Sat, Sep 1, 2018, 11:19 PM Kevin Bowling <kevin.bowling@kev009.com> wrote:
>> Seems like Chuck Cranor is at CMU http://chuck.cranor.org/.  Chuck
>> Silvers is with you?
> 
> 
> Why I'm embarrassed to admit you are right. Chuck Silvers also did some VM work, but not uvm.
> 
> Warner
> 
>> On Sat, Sep 1, 2018 at 1:25 PM, Warner Losh <imp@bsdimp.com> wrote:
>> >
>> >
>> > On Sat, Sep 1, 2018, 1:42 PM ron minnich <rminnich@gmail.com> wrote:
>> >>
>> >> I was his advisor on that thesis so I got to watch it roll out as it
>> >> happened.
>> >>
>> >> uvm replaced the machvm in netbsd.
>> >>
>> >> For a time, Chuck set it up to run in parallel with the existing VM. You
>> >> could start a process and pick which vm it used. For a while, it defaulted
>> >> to the existing one. Then, at some point, it defaulted to uvm. Then, at some
>> >> point, the old one was removed.
>> >>
>> >> more here:
>> >>
>> >> http://www.netbsd.org/docs/kernel/uvm.html
>> >>
>> >> via search terms
>> >> uvm replaces machvm netbsd
>> >>
>> >> chuck was a long time contributor to netbsd IIRC, but last time we talked,
>> >> he was using Linux.
>> >
>> >
>> > These days I know he's hacking on FreeBSD's storage stack with me at work
>> > :). I think he's still a netbsd contributor. I see his name in the commit
>> > log often..
>> >
>> > Warner

[-- Attachment #2: Type: text/html, Size: 3633 bytes --]