On Thu, Aug 31, 2006 at 08:57:30PM +0200, Thomas Moschny wrote: > This is absolutely unnecessary, because the validity of a cert (not > to be confused with the trust in that cert) is a constant. It could > be verified once and for all and stored in the database together > with the cert.
Well, the original intention of revalidation at every use was to catch 'unpredictable' errors, including such things as disk or memory corruption. Those are still worthy goals. That comes with a cost, and the cost is excessive, and we need to do something about it, but 'absolutely unnecessary' is too strong. For those working on performance, good test data is always useful. I recently finished a cvs_import of the NetBSD project's base src. {f,ht}ttp://ftp.netbsd.org:/ftp/pub/NetBSD/misc/dan/orig-NetBSD-src.mtn This is 1.7Gb and contains ~180k revisions. I also have a smaller pkgsrc import (~110k revs) which I'll upload beside that. I haven't yet imported some of the other modules. Both show up current scalabilty and performance issues quite dramatically. I'm going to try redoing those imports with branch reconstruction to test that feature soon. -- Dan.
pgpjhltgBKv03.pgp
Description: PGP signature
_______________________________________________ Monotone-devel mailing list Monotone-devel@nongnu.org http://lists.nongnu.org/mailman/listinfo/monotone-devel