On Thu, Sep 10, 2009 at 7:22 PM, Jeffrey Hutzelman<[email protected]> wrote: > --On Thursday, September 10, 2009 07:00:34 PM -0400 "Matt W. Benjamin" > <[email protected]> wrote: > >> 5. directory entries cannot be looked up, as server doesn't know >> applicable normalization rules > > I'm not sure I'd go quite that far (and I'm pretty sure I was the one who > made this argument). I expect a lookup operation will _usually_ not be as > useful as doing the lookup on the client, for this reason and also due to > performance concerns. But occasionally it might, so I wouldn't completely > rule out the option of a lookup operation. >
The performance concerns in this space are something we shouldn't take lightly. Satyanarayanan, Kazar, and others have pointed out in numerous papers that experimentation with early vice prototypes revealed server-assisted path lookup to result in unacceptably high round trip counts, and a large drain on server resources. If we do add a lookup RPC, we should implore implementors to use it sparingly (if at all), or implement an adaptive algorithm whereby first N (for small N relative to dir size) lookups to a given directory fid use RXAFS_Lookup, and after that we start doing bulk fetches. While XCB would partially mitigate these effects due to the observed high degree of temporal locality of reference for dnlc entries, keep in mind that XCB can't solve for the spatial locality of reference problem that ls -l, graphical file browsers, etc. bring to the fore (hence why I think we need to push for threshold-based fallback to directory fetch, should we standardize a lookup interface). -Tom _______________________________________________ AFS3-standardization mailing list [email protected] http://michigan-openafs-lists.central.org/mailman/listinfo/afs3-standardization
