Re: subversion 1.1.0 checkin crashes - repo needs recovery

Greg Hudson wrote:

> First I can recall hearing of this. What issues are these?
>
> (The only actual, concrete complaint I can remember seeing about FSFS
> storing many rev files in the same directory is that someone's NFS
> backup script became very slow.)

That's what I was referring to. NTFS performance degrades quickly with
multiple 10's of thousands of files in a single directory.

And I have personally noticed that this can happen even if the directory
in question isn't actually accessed directly. For example, we have a
Windows-based web server and we have a folder with lots of shopping cart
files (very small). _Any_ drive access on that volume is slow; I have
no idea why. My WAG is that since NTFS will store the file contents in
the directory table entry if the file contents are smaller than the
blocksize, having lots of small files make the entire directory table
much larger and causes a lot of swapping to handle it. Just typing
"DIR" in a directory several layers above the large folder will take
several minutes to return /any/ information.

I think the last time this came up, I opined that the FSFS design should
have used some form of hashing to store the repository files. Of
course, I didn't offer a patch to do it, so my complaints are largely
theoretical. No offense intended to the basic FSFS design; in my
experience, NTFS has not demonstrated itself to be nearly as robust as
most *nix filesystems.

John

-- 
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4501 Forbes Boulevard
Suite H
Lanham, MD  20706
301-459-3366 x.5010
fax 301-429-5748
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org