Subject: Re: UBC status
To: Chuck Silvers <firstname.lastname@example.org>
From: Yuri Y. Chebotarev <email@example.com>
Date: 09/26/1999 13:09:03
On Fri, 24 Sep 1999, Chuck Silvers wrote:
> with the impending creation of the 1.5 release branch, I'd like to
> get people thinking about what else needs to be done before the UBC
> branch can be merged into -current. (the intent is to merge it soonish
> after the 1.5 branch is created, but it'll probably wait until sometime in
> November... my job has been consuming all my time lately, but that will be
> changing in the next month or so.)
> towards that end, here's some status:
> generally the UBC code works well enough to recompile a kernel,
> even with "make -j 100". I've also done some testing with copying
> large (100s of megs) trees with "tar | (cd && tar)", which works fine.
> I recall there were still some hanging problems with blocksize<pagesize
> on the alpha, but not on x86, so that's one thing to watch out for.
> here are some lists of things left to do:
> should be done before the initial commit to -current:
> - additional locking needed between VOP_GETPAGES() and VOPs that modify
> the file's size or blockmap. I've got a start on this.
> - implement error handling for uiomove() failures.
> - readahead, and writebehind flushing.
> - nfs write caching is currently write-thru, needs to become write-back again.
It could be implemented by adding one pointer to credentials to struct
vm_page and using it in nfs VOP_PUTPAGES().
just my $0.02,