[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Scaling replicas

Alister Winfield wrote:
On Sat, 2002-11-16 at 03:28, Quanah Gibson-Mount wrote:


Has anyone seen any scaling issues with OpenLDAP -- i.e., if you have a single master, is there any sort of performance/replication slowdown noticed as you increase the number of replicas? Do people have a master with 10 replicas? 20? 30? 50 or more? Any issues noticed with a large number of replicas?


-- Quanah Gibson-Mount Senior Systems Administrator ITSS/TSS/Computing Systems Stanford University GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html

I suppose this should be reported but .... under load slurpd has in the
past lost updates for me. This has always happened when there are > ~10
slaves and higher than average update rate say >1 change a second.
Anyone else seen this ? Its very hard to replicate as its random and as
such I haven't ever found any hard proof as to why it happened. The most
important symptom was that the change never even got attempted according
to the logs on the affected slaves.

Alister Winfield - Easynet Systems Developer.

Yes, I've noticed this happening with OpenLDAP 2.0.* running under Solaris 2.7 . The problem is due to Solaris bug # 4472643 (fopen and fdopen fail in less than FOPEN_MAX or STREAM_MAX calls). I raised this with Sun some time ago - I got the reply that it might be fixed in Solaris 9. Our workaround is to take the load off the master and point clients at the replicas.

Dave Lewney
Principal Systems Programmer, Computing Service
University of Sussex, Brighton BN1 9QJ. Tel: 01273 678354 Fax: 01273 271956