RE: Caching (*not* cache consistency) proposal (LONG)

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

From: Noveck, Dave (dave.noveck@netapp.com)
Date: 07/16/99-02:51:16 PM Z 

Message-ID: <7F608EC0BDE6D111B53A00805FA7F7DA05D10B55@tahoe.corp.netapp.com>
From: "Noveck, Dave" <dave.noveck@netapp.com>
Subject: RE: Caching (*not* cache consistency) proposal (LONG)
Date: Fri, 16 Jul 1999 12:51:16 -0700

 
> A very fine effort to address an important aspect of the 
> NFSv4 spec.  In
> my opinion, the claim of NFSv4 to relevance in WAN environments is
> vacuous without a solid story on caching.  Of course, I'm not 
> unbiased.

None of us is. :-)

> 
> On Wed, 7 Jul 1999 07:44:39 -0700  "Noveck, Dave" 
> <dave.noveck@netapp.com> wrote:
> > [[The revalidation procedure mentioned above may give a lot
> >   of false indications that invalidation is needed when a
> >   client is writing data as a result of doing unlocks on
> >   other area.  The problem is similar to that addressed by
> >   weak cache consistency info in nfs v3.  The problem with
> >   resurrecting that mechanism is that when large areas are
> >   written multiple write are done asynchronously.  It is not
> >   clear how to reliably infer whether any writes have been 
> >   done by other clients when multiple write requests are in
> >   flight.  Anyone have ideas?]]
> 
> Wouldn't the atomic before / after change attribute trick you describe
> later for directories work here?  If, after sorting out-of-order
> responses, the before and after change attributes dove-tail, then no
> intervening client writes have occurred and the cached data does not
> need invalidation.
 
Possibly.  The problem is that the change attribute is not defined as 
monotonic so that sorting per se not available.

I guess you can infer an ordering from the pre and post values that
you actually see.  So given a set of requests that were in flight
at the same time, if there exists an ordering (call it <before>)
of those requests such that for all a and b such that a <before> b,
post(a) == pre(b) then I think you are OK.  Now I just how to figure 
out how to implement that.

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.2 : 03/04/05-01:47:22 AM Z CST