Status of >800 mounts

Status of >800 mounts

[Michael Blandford]

Hello,

What is the current status of having more than 800 nfs mounts ( on 2.4 
and 2.6 )?

I had heard there might be a patch to use one port per server instead of 
per mount. 

Thanks,

Michael


Disclaimer: The content of this message is my personal opinion only and 
although I am an employee of Intel, the statements I make here in no way 
represent Intel's position on the issue, nor am I authorized to speak on 
behalf of Intel on this matter.

Re: Status of >800 mounts

[Stew Smith]

On Thu, May 06, 2004 at 10:48:44AM -0700, Michael Blandford wrote:
> Hello,
> 
> What is the current status of having more than 800 nfs mounts ( on 2.4 
> and 2.6 )?
> 
> I had heard there might be a patch to use one port per server instead of 
> per mount. 
> 

i am working on a patch currently which multiplexes nfs connections
per sever so you end up with a 800 server limit, similar to the
way nfsv4 works now.. i will send it out for review very soon after
i ferret out a few bugs.

another idea we had was to add SO_REUSEADDR to rpc sockets so
you have more available on a busy system.

-- 
Stew Smith

Re: Status of >800 mounts

[Stew Smith]

On Thu, May 06, 2004 at 12:39:33PM -0700, Chris Croswhite wrote:
> Why only up to 800 (although this is a large number)?  There are
> instances where I could have 1200-1600 clients connecting to the same
> server (4 way opteron) if I could that many...
> 

this limit has been hardcoded in the sunrpc layer since the
birth of time. it starts at 800 and counts down to 0 until
it find an unused port. requring a privileged port gets
you the default level of security NFSv3 provides...

anybody know why rpc doesn't use the full port 0-1023 range?

--
Stew Smith

Re: Status of >800 mounts

[Ian Kent]

On Thu, 6 May 2004, Stew Smith wrote:

> On Thu, May 06, 2004 at 10:48:44AM -0700, Michael Blandford wrote:
> > Hello,
> > 
> > What is the current status of having more than 800 nfs mounts ( on 2.4 
> > and 2.6 )?
> > 
> > I had heard there might be a patch to use one port per server instead of 
> > per mount. 
> > 
> 
> i am working on a patch currently which multiplexes nfs connections
> per sever so you end up with a 800 server limit, similar to the
> way nfsv4 works now.. i will send it out for review very soon after
> i ferret out a few bugs.
> 

I tried porting a patch of Tronds' for 2.5, forward for 2.6 and backward 
to 2.4. My understanding of the NFS/rpc code was (is) not sufficient 
to be confident with this. When I asked for someone to check and comment 
on what I had done I got only silence. At least I think I posted that to 
the list?

After that I tried to merge it into a RedHat kernel, which proved to be 
too much for me to do at the time.

I look forward to seeing your patch.

Ian

Re: Status of >800 mounts

[Tim Hockin]

On Fri, May 07, 2004 at 11:22:40AM +0800, Ian Kent wrote:
> > i am working on a patch currently which multiplexes nfs connections
> > per sever so you end up with a 800 server limit, similar to the
> > way nfsv4 works now.. i will send it out for review very soon after
> > i ferret out a few bugs.

With an early version of Stew's patch, I hit over 2000 simultaneous NFS
mounts.  Once the last bugs are out, we can call this one closed, I think.

:)

-- 
Tim Hockin
Sun Microsystems, Linux Software Engineering
thockin@sun.com
All opinions are my own, not Sun's

Re: Status of >800 mounts

[Ian Kent]

On Thu, 6 May 2004, Tim Hockin wrote:

> On Fri, May 07, 2004 at 11:22:40AM +0800, Ian Kent wrote:
> > > i am working on a patch currently which multiplexes nfs connections
> > > per sever so you end up with a 800 server limit, similar to the
> > > way nfsv4 works now.. i will send it out for review very soon after
> > > i ferret out a few bugs.
> 
> With an early version of Stew's patch, I hit over 2000 simultaneous NFS
> mounts.  Once the last bugs are out, we can call this one closed, I think.
> 

Will it cover 2.4 as well or is it 2.6 only?

Ian

Re: Status of >800 mounts

[Tim Hockin]

On Fri, May 07, 2004 at 12:24:35PM +0800, Ian Kent wrote:
> > With an early version of Stew's patch, I hit over 2000 simultaneous NFS
> > mounts.  Once the last bugs are out, we can call this one closed, I think.
> 
> Will it cover 2.4 as well or is it 2.6 only?

It's developed on 2.6.  The backport is left as an excercize for the user :)

Re: Status of >800 mounts

[Ian Kent]

On Thu, 6 May 2004, Tim Hockin wrote:

> On Fri, May 07, 2004 at 12:24:35PM +0800, Ian Kent wrote:
> > > With an early version of Stew's patch, I hit over 2000 simultaneous NFS
> > > mounts.  Once the last bugs are out, we can call this one closed, I think.
> > 
> > Will it cover 2.4 as well or is it 2.6 only?
> 
> It's developed on 2.6.  The backport is left as an excercize for the user :)

Let me at it! Please!

Re: Status of >800 mounts

[Jeff Moyer]

==> Regarding Re: [autofs] Status of >800 mounts; Stew Smith <stew.smith@sun.com> adds:

stew.smith> On Thu, May 06, 2004 at 10:48:44AM -0700, Michael Blandford
stew.smith> wrote:
>> Hello,
>> 
>> What is the current status of having more than 800 nfs mounts ( on 2.4
>> and 2.6 )?
>> 
>> I had heard there might be a patch to use one port per server instead of
>> per mount.
>> 

stew.smith> i am working on a patch currently which multiplexes nfs
stew.smith> connections per sever so you end up with a 800 server limit,
stew.smith> similar to the way nfsv4 works now.. i will send it out for
stew.smith> review very soon after i ferret out a few bugs.

stew.smith> another idea we had was to add SO_REUSEADDR to rpc sockets so
stew.smith> you have more available on a busy system.

Are you going to post the patches (or have you already) to the linux-nfs
list?  If enough people find it useful, would be nice to get it upstream.

-Jeff

Re: Status of >800 mounts

[Michael Blandford]

Stew Smith wrote:

>i am working on a patch currently which multiplexes nfs connections
>per sever so you end up with a 800 server limit, similar to the
>way nfsv4 works now.. i will send it out for review very soon after
>i ferret out a few bugs.
>
>another idea we had was to add SO_REUSEADDR to rpc sockets so
>you have more available on a busy system.
>
>  
>
any updates on this patch? 

Michael

Disclaimer: The content of this message is my personal opinion only and 
although I am an employee of Intel, the statements I make here in no way 
represent Intel's position on the issue, nor am I authorized to speak on 
behalf of Intel on this matter.

Re: Status of >800 mounts

[Stew Smith]

On Mon, Jun 14, 2004 at 03:56:06PM -0700, Michael Blandford wrote:
> Stew Smith wrote:
> 
> >i am working on a patch currently which multiplexes nfs connections
> >per sever so you end up with a 800 server limit, similar to the
> >way nfsv4 works now.. i will send it out for review very soon after
> >i ferret out a few bugs.
> >
> >another idea we had was to add SO_REUSEADDR to rpc sockets so
> >you have more available on a busy system.
> >
> > 
> >
> any updates on this patch? 

we have code ready for review and testing but are awaiting
legal clearance to release it by Sun.

i will send it out the momment i am able to do so.

thanks,

--Stew