[cvsnt] High server load on update!

Terris Linenbach noreply at nowhere.nwh
Tue Oct 22 22:55:17 BST 2002


The same thing happens to me regularly.  The CPU gets pegged to 100% when
updating a folder that contains sub-folders.  Which specific file it decides
to hang on is random (i.e., the last file output in the WinCVS window).
There are no dangling locks prior to the update and only one client is using
the server.  Disk has gigabytes of free space, more than the raw CVS files
occupy.  I'm using hardware RAID (basic two-drive mirroring).

The cvs.exe process never completes and has to be killed using kill.exe.

I have tried using WinCVS and Tortoise as clients without compression.

I'm using 57j on the server.

"Tony Hoyle" <tmh at nodomain.org> wrote in message
news:pan.2002.10.22.19.39.33.278334 at nodomain.org...
> On Tue, 22 Oct 2002 13:28:44 +0200, Anders Truelsen wrote:
>
>
> > Updates on the other hand really pull the teeth out.
> > - A single update of a large workspace pushes the CPU to 100% for up to
> > 30 seconds.
> > Why is that? - I never observed that behavior on the old machine.
>
> Update is a two-step operation.  In the simplest case the names &
> datestamps of the files in each client directory are sent and the server
> then decides what (if anything) to send to a client.  If the client thinks
> something has changed on its side it sends the suspect file for
> diff/collision detection, which is stored temporarily on the server.
>
> This can use up a fair bit of temporary drive space in the worst case
> (such as all files appearing modified).
>
> It shouldn't affect CPU much though.  I'd check that your IDE interface
> has DMA enabled - if it's on PIO it'll be using up all the CPU accessing
> the disk.
>
> Tony
>





More information about the cvsnt mailing list