I've integrated upto the readdir coalesing patch (07) and I've been doing some testing with each of the patches.
With patch 7, with readdir coalesing, I don't seem to achieve as much of a performance gain seems to marginal. How exactly did you test this since you claimed a much higher boost in performance? Regards, Ram. On Wed, 2010-06-23 at 17:50 -0700, Life is hard, and then you die wrote: > On Wed, Jun 23, 2010 at 02:23:57PM +0200, Ramshankar wrote: > > On Wed, 2010-06-23 at 04:51 -0700, Life is hard, and then you die wrote: > > > Ok, makes sense. I took a closer look at the Solaris coding standards > > > guide and realized I got a couple things wrong (mainly > > > line-continuations and switch statements); so I reworked the patches > > > to follows those better. > > > > > > Also, while looking at the latest linux shared-folders changes today I > > > realized that I had fallen into the same trap with the readdir fix > > > (patch 06) as the linux code (http://www.virtualbox.org/ticket/5251), > > > i.e. it wasn't properly resetting the directory listing for apps doing > > > a seekdir. I've fixed that now and tested it, both using the 'svnadmin > > > load' mentioned in the above ticket as well as with a small test-suite > > > I wrote for this. > > > > > > Attached therefore you'll find a new set of patches which include both > > > the style fixes and the readdir fix, but are otherwise identical to > > > the previous ones. > > > > > > > I'll take a look at the patches you provided and get back to you > > > > regarding them. Thanks a lot for the effort and time on this. > > > > > > Thanks. > > > > Great! Thanks for the quick fixes. I've begun reviewing & committing > > your patches. Since these patches are significant, I'll mentioned your > > name as "Ronald" (no mail address included) as a contributor in the > > source files involved, I hope this is agreeable. If you don't want to be > > mentioned for some reason let me know. > > Sounds good. Either way is fine with me. > > > I will mail you here once I complete reviewing, committing and testing > > these patches. > > Thanks - if you'd like me to rework/change anything, just let me know. > > A few things I should mention: > > - with the 08_stat_coalesce.patch the sfprov_get_{a,m,c}time() > and sfprov_get_mode() functions are not used anymore; I left them > in as I wasn't sure what the full reasoning behind the original > approach was, but they could be deleted now. > > - the sfprov_get_size() function could also easily be removed by > replacing the last remaining call to it with an sfnode_get_stat() > call. Again, I didn't do so because I wasn't completely sure if you > were thinking of other providers and whether you wanted vboxfs_prov > interface to be as fine-grained as possible or not. > > - lastly, I realize the stat-caching may be controversial (patch 09). > Maybe defaulting it to off (DEF_STAT_TTL_MS=0) may make people more > comfortable. > > > Cheers, > > Ronald > > > _______________________________________________ > vbox-dev mailing list > [email protected] > http://vbox.innotek.de/mailman/listinfo/vbox-dev _______________________________________________ vbox-dev mailing list [email protected] http://vbox.innotek.de/mailman/listinfo/vbox-dev
