Re: svn commit/checkout performance issues (P5)

Hi,
    and here is the last piece of data for the client on a local (/tmp)
disk which basically tells the following:
the svn client is overloaded with disk io (should some of it be moved to
the server?) and that *fsfs* system is the best in performance and
repository size.

                 nosync sync fsfs
========================================================
(client: nfs)

import 8:42.89 16:07.87 2:14.71
size(k) 403,756 403,760 181,112

checkout 20:12.19 22:39.56 17:43.4
add --force 3:11.6 3:13.80 3:06.98
commit 57:59.91 1:00:44.99 45:00.68
size(k) 798,128 798,132 310,080

==================================================
(client: local)

import 8:40.40 16:48.25 3:03.10

checkout 2:03.00 2:15.52 1:41.79
add --force 1:10.14 0:51.61 0:50.82
commit 30:56.03 36:08.29 11:20.05

Brandon Ehle wrote:

>
>>
>> =========================================
>> checkout
>> ========
>> accurev update 5:02
>> svn checkout 17:36 (pretty long)
>> =========================================
>> repository size:
>>
>> svn 493,024K (~180,000 per kit version) (impressive)
>> accurev 1,379,852K (~400,000K per kit version)
>>
>
> What are the working copy (client) sizes after checkout? Also what
> filesystem are you running on the clients?
>
> Also, do you know about the --bdb-txn-nosync flag or did you use the
> FSFS repository format? During imports or commmits, setting the
> --bdb-txn-nosync flag on the repository can be as much as 30x or more
> speedup. On checkouts with this flag set, you should see about a 2x
> speedup.
>
>

-- 
>> Sergey Bogdanov
      INTEL massachusetts	
      HD2-2/246
      77 Reed Rd, Hudson, MA 01749
      Tel: (978) 553-2724