Comment 17 for bug 1344940

Revision history for this message
Andrew Wilkins (axwalk) wrote :

> So, a 4.24 MB oplog over 1.18 hours. Given that, and that we can resize if necessary, I'm going to drop it to 1GB max. If someone (Kapil?) could get stats for a real production environment, it'd be great to get some more numbers.

John pointed me at the "peer-xplod" charm, wherein units will communicate with each other peer (so N^2) continuously, but updating relation settings. I've deployed 16 units of that charm, and it's less than doubled the oplog size. If I extrapolate to 24 hours of runtime, then it'll consume roughly 200MB of the oplog.

I'm happy to drop the size lower (512MB?), but I'd be more comfortable doing so if we had real numbers.