skiplist vs db format

[Igor Brezac]

On Fri, 23 Jan 2004 throwaway-it at cox.net wrote:

> We are using STK's Shared Virtual Array, V960's with PPRC (Peer-to-peer-remote-copy).
>
> Just for a test, we moved the /var/imap directory onto local disk (disk
> mirrored using Disksuite) and IO wait shot up to over 80%.

I use IQstore storage array with similar load and my asvc_t is much lower
than yours (5-20), %b is slightly lower and %w is always 0 which is what
you want.

>
> Is there a fix/solution for the SSL imap/pop daemons dieing other than to upgrade to latest Cyrus?
>

I am not aware of SSL issus with cyrus and solaris.  What entropy package
do you use?

-Igor

>   Bob
>
> >
> > Your drives are working overtime.  There is nothing waiting to be written
> > (good), but service time is high and I imagine things feels a bit
> > sluggish.  Is this software RAID?  Can you add more drives?
> >
> > I do not know if skiplist will reduce your disk load (my guess is yes
> > because skiplist does not maintain on-disk environment like berkeley
> > does), but I would change to skiplist regardless because of stability.
> > You will not have to restart cyrus as often if at all.
> >
> > -Igor
> >
> > On Fri, 23 Jan 2004 throwaway-it at cox.net wrote:
> >
> > > It 4.0.14 (November 18, 2001).
> > > Here's the output from 6 periods:
> > >
> > >     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
> > >     1.3   24.7    8.0  202.5  0.0  3.3    0.5  125.6   0  15 c3t17d2s0
> > >     2.1   57.8   10.8  414.6  0.0  7.0    0.0  116.4   0  45 c3t17d2s0
> > >     1.5   62.6    7.5  443.7  0.0 25.2    0.0  394.1   0  55 c3t17d2s0
> > >    63.6   63.6  502.9  443.1  0.0  7.7    0.0   60.8   0  64 c3t17d2s0
> > >     1.8   67.1    9.0  471.4  0.1 24.4    0.8  354.0   1  47 c3t17d2s0
> > >     1.9   69.7    8.9  485.0  0.0 16.5    0.0  229.9   0  53 c3t17d2s0
> > >
> > >
> > >
> > > >
> > > >
> > > > What version of Berkeley DB4?  If it is 4.1.25+ you need to upgrade cyrus
> > > > (2.1.16 or 2.2.3).  What does 'iostat -xPn 30' show for the RAID 1
> > > > partition?
> > > >
> > > > -Igor
> > > >
> > > >
> > > > On Fri, 23 Jan 2004 throwaway-it at cox.net wrote:
> > > >
> > > > > For those of you who have converted from the Berkeley db to skiplist, what kind of a performance improvement did you receive?
> > > > >
> > > > > We are using Berk. db4 and the IO wait for the disk containing the db is consistantly over 50%.  The partition is 7GB, which also contains the user, quota, proc, and db directory.  It resides on an STK V960 fibre channel, RAID 1 disks.  I am seeing "DBERROR db4: nnnn lockers" errors with the number over 10,000+ ... usually soon afterwards, the imap process needs to be restarted.
> > > > >
> > > > > Also, the 'imapd -s' and 'pop3d -s' process unexpectedly dies and SSL connections are no longer possible until imap process is restarted.  Is this a known problem?
> > > > >
> > > > > Our platform is :
> > > > > Cyrus 2.1.11
> > > > > Solaris 8, Sunfire 280R
> > > > > 4GB memory, 2 CPU
> > > > >
> > > > > Thanks,
> > > > >
> > > > >   Bob
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > >
> > > > --
> > > > Igor
> > > >
> > >
> > >
> > >
> >
> > --
> > Igor
> >
>
>

-- 
Igor