| 1 |
On Sat, Jan 24, 2004 at 04:37:35PM +0000, ?var Arnfj?r? Bjarmason wrote: |
| 2 |
> Attached is a recent test i did on subversion regarding exactly this, |
| 3 |
> the the size of the repository I came to the same conclusion as you |
| 4 |
> (the db is big) in this case 161MB for a raw import of portage versus |
| 5 |
> CVS's 70MB. |
| 6 |
However, 70MB is NOT the actual amount of data, as if we do use |
| 7 |
subversion, we WILL want our old CVS history. The CVS repository is |
| 8 |
currently 899Mb in size. The gentoo-x86 module that you see as the |
| 9 |
portage tree is 572Mb. |
| 10 |
|
| 11 |
Another major question I have, is how will it deal with lots of |
| 12 |
concurrent access? CVS excels in this regard, as it does locking on a |
| 13 |
per-file basis for writing (no locks on reading). |
| 14 |
AFAIK with BDB databases there is locking only for the entire database, |
| 15 |
so there may be an issue of a LOT more lock contention which would slow |
| 16 |
things down a lot. There are very seldom less than four simultanous cvs |
| 17 |
actions going on, and most of the time things are much busier than that |
| 18 |
(I've seen the server having 20 cvs operations going on at the same |
| 19 |
time). |
| 20 |
|
| 21 |
A normal CVS checkout of the gentoo-x86 from the tree (locally on the |
| 22 |
CVS server) takes 8 minutes. Your SVN co took 3.8 times as long as the |
| 23 |
CVS checkout so thats nearly 30 minutes to do a checkout :-(. |
| 24 |
|
| 25 |
I am really worried about lock contention however. We already keep all |
| 26 |
of the cvs locks on a tmpfs filesystem for speed and disk fragmentation |
| 27 |
reasons. |
| 28 |
|
| 29 |
-- |
| 30 |
Robin Hugh Johnson |
| 31 |
E-Mail : robbat2@××××××××××××××.net |
| 32 |
Home Page : http://www.orbis-terrarum.net/?l=people.robbat2 |
| 33 |
ICQ# : 30269588 or 41961639 |
| 34 |
GnuPG FP : 11AC BA4F 4778 E3F6 E4ED F38E B27B 944E 3488 4E85 |
Archives