[Gluster-devel] Cache Translator Questions

Martin Fick mogulguy at yahoo.com
Wed Dec 31 05:17:22 UTC 2008


--- On Tue, 12/30/08, Basavanagowda Kanur <gowda at zresearch.com> wrote:

> On Tue, Dec 30, 2008 at 11:11 PM, Martin Fick
> <mogulguy at yahoo.com> wrote:
> >
> > 1B) If memory only, are there any plans for a file
> based cache?
>
> Yes, we have plans for file based cache in our future
> release.

It sounds like this would simply require adding local disk cache read/writting to the translator without requiring any changes to the way coherency is maintained?  Seems like it might be well worth it.  If it is as simple as I am thinking, with disk caching, clients could maintain huge local caches and likely dramatically improve performance.  

These days clients may even have enough storage to cache an entire filesystem making the cluster servers simply authoritative repositories.  If this were the case, perhaps networks could be optimized for latency instead of throughput without suffering much of a FS throughput performance penalty since most large reads would come from the local cache.  Just some random thoughts...

Thanks for your replies,

-Martin



      





More information about the Gluster-devel mailing list