[9fans] Thank you very much

[9fans] Thank you very much

[Adriano Verardo]

First of all, thank you very much to all who, with their suggestions and 
comments, kindly helped me in understanding 9P2000.

I think that metadata can be a good solution for me. I agree that's 
gross but can be acceptable in some particular situations. All the more, 
it doesn't necessarily exclude the incidental use of normal clients too.

Let me issue another question.

Beeing constrained to mount small file server on FreeBSD-p9p the high 
number (up to 8630) of cached fids can be a problem.

Is it configurable ?

Thanks

Adriano

Re: [9fans] Thank you very much

[Russ Cox]

> Let me issue another question.
> 
> Beeing constrained to mount small file server on FreeBSD-p9p the high 
> number (up to 8630) of cached fids can be a problem.
> 
> Is it configurable ?

You're actually mounting the file system as a
FUSE file system, via the 9pfuse protocol converter.
The problem here is that the FreeBSD FUSE kernel
module is not issuing FUSE_FORGET (like Tclunk)
messages as aggressively as it could.

On Linux, if FUSE walks the same path twice and gets
different nodeids (like fids), then it tosses away one of
them -- no point caching them both -- but FreeBSD
appears to keeps them both in the cache. 

You'd have to ask the fuse4bsd guys about this behavior.

Russ

Re: [9fans] Thank you very much

[Jack Johnson]

On 4/12/07, Russ Cox <rsc@swtch.com> wrote:
> You'd have to ask the fuse4bsd guys about this behavior.

Any idea what the Mac FUSE behavior is?

-Jack

Re: [9fans] Thank you very much

[Russ Cox]

> Any idea what the Mac FUSE behavior is?

MacFUSE does not reissue lookups for names
already in the vnode cache.  So if you do

	ls -l 
	ls -l

Only the first will look up every name; the 
second will repeat the getattrs, but not the lookups.

Russ