Linux tmpfs (Was: [OpenAFS] Solaris 10u6: ZFS cache?)

[Christof Hanke]

Yes, thanks. That's what I meant.

Christof
________________________________________
From: Jeffrey Altman [jaltman@secure-endpoints.com]
Sent: Wednesday, November 26, 2008 11:43 PM
To: Christof Hanke
Cc: openafs-info@openafs.org
Subject: Re: Linux tmpfs (Was: [OpenAFS] Solaris 10u6: ZFS cache?)

Christof Hanke wrote:
> While talking about cache-performance (not that I know how they really wo=
rk)
> Would it be beneficial to have a separate Meta-data cache from a File-dat=
a cache?
> I'm thinking of situations where the the performance is poor (long distan=
ce, whatever)
> It would me frustrate a lot, if I had to wait, say 10 min, for an "ls" an=
d after downloading a file bigger than my cache another "ls" would
> take the same time.
> So, would it make sense to cache the stat() information in a separate cac=
he ?
> Just thinking loud here.
>
> Christof

in the AFS cache manager, volume info, vnode meta data, and
file/directory data are stored in separate classes of objects.

I think you are requesting that an arbitrary percentage of the file/dir
data cache be reserved for directory data vs file data.   The 1.5 tree
has an implementation of split cache management for segmenting
read/write vs read/only data.   Extending that to support other
classifications should not be hard.

Send requests or patches that implement features to openafs-bugs@openafs.or=
g

Jeffrey Altman