Re: [Gnu-arch-users] Re: Why we might use subversion instead of arch.

[John Goerzen]

On Fri, Feb 27, 2004 at 12:42:16PM -0800, Tom Lord wrote:
> What tagging method are you using and what does a cold-cache
> `inventory --source --both --all --ids > /dev/null' take?

I am using explicit tags and that time is:

real    0m28.844s
user    0m1.020s
sys     0m1.620s

Run a second time, with things presumably in-cache:

real    0m1.609s
user    0m0.950s
sys     0m0.610s

> I'm hoping that the inventory will be reasonably expensive and that
> you're using explicit tags since there are changes in the queue for
> the 1.3preX series that will likely make a rather large difference.

You're batting .500 -- not bad :-)

You may also be interested in the following warm-cache statistics:

$ time tla what-changed > /dev/null

real    0m37.879s
user    0m7.150s
sys     0m5.410s

 ( no actual changes above )

$ time tla get linus--mainline--2.6--patch-6 wc-mainline-2.6 
* from revision library:
* address@hidden/linus--mainline--2.6--patch-6
* tree version set address@hidden/linus--mainline--2.6

real    1m58.330s
user    0m5.250s
sys     0m10.000s

$ time tla get linus--mainline--2.6--patch-4 wc-mainline-2.6
* from archive cached: address@hidden/linus--mainline--2.5--patch-76
* patching for revision: address@hidden/linus--pre--2.6--base-0
* patching for revision: address@hidden/linus--pre--2.6--patch-1
...
* patching for revision: address@hidden/linus--pre--2.6--patch-11
* patching for revision: address@hidden/linus--mainline--2.6--base-0
...
* patching for revision: address@hidden/linus--mainline--2.6--patch-4
* making pristine copy
* tree version set address@hidden/linus--mainline--2.6

real    5m17.601s
user    1m19.480s
sys     1m56.150s

$ time tla replay linus--mainline--2.6--patch-5
...
real    0m26.346s
user    0m4.610s
sys     0m11.030s

$ time tla update linus--mainline--2.6--patch-6 > /dev/null
 real    7m20.953s
 user    0m33.600s
 sys     1m3.930s

Conclusions:

1. The fact that patch-6 was in the revision library made *a* difference but
brought my performance nowhere near the level of what some are suggesting
it should be.

2. The difference between replay and update is larger than I thought.
   update is even slower than a fresh get.

3. I don't have exact statistics on commit.  I will try to get some when
   the next patch shows up.  One thought it that it seems to be a lot
   slower than what-changed, even if there were few changes -- but that's
   just from memory.

-- John