Nicolás Lichtmaier wrote:
>
>>>
>>> I don't know... I'm not sold on FSF yet. RHEL3 does not support
>>> "dir_index" ext2/3 feature. This means directory access is *linear*.
>>> And FSFS means having thousands of files...
>>
>>
>> Hmmmm, I've been using FSFS on WBEL3 on a 13.5GB repository on Linux
>> and not noticed any particular slowdowns, we've got 10687 revisions so
>> far.
>>
>> In fact, the Linux copy is about 4 - 5 times faster than than the same
>> thing on Window 2000 server.
>>
>> But of course this is just one data point, your mileage may vary.
>>
>> If we can figure out a way to easily shoe-horn in the BDB 4.2.52 to
>> RHEL3 without messing things up then I'm all ears.
>>
>> If people are willing to live with the problems then I could create
>> another rhel-3-bdb-4.2.52 build and take it back to like it was or
>> something.
>>
>> Let me know what you think the best options are.
>
>
> Perhaps my FSFS fears are not justified. I may well end up using it.
> Your repository is bigger than mine and you say you don't see any
> difference...
Note that while the old ext2 lookups were linear, they are pretty damn fast.
Besides, the ratio of disk seek time to data read time for a directory
fetch is going to be so high it's not funny. Also, they are held in the
kernel dentry cache, and with the skip deltas the set that gets used is
probably consistent and small enough for this to be quite effective.
So the for not having htree isn't going to be as bad as it might first look
until one reaches absolutely immense directory sizes where the dentry cache
can't hold the log2(n) revs that are being used, and the time to read in
all the directory entries is relevant compared to the seek time required to
get there in the first place.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Dec 17 19:49:10 2005