Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Mon, Jul 02, 2012 at 05:00:51PM -0500, J.R. Lillard wrote: > On Mon, Jun 25, 2012 at 2:10 PM, Lars Ellenberg > <lars.ellenberg at linbit.com>wrote: > > > On Fri, Jun 22, 2012 at 02:38:07PM -0500, J.R. Lillard wrote: > > > Witnessed another bandwidth spike that slowed my stacked layer down. > > > > > > 10: cs:Connected ro:Primary/Secondary ds:UpToDate/UpToDate A r----- > > > ns:192538032 nr:0 dw:599650316 dr:1701817080 al:4481613 bm:43214 lo:1 > > > pe:2050 ua:0 ap:2049 ep:1 wo:f oos:0 > > > > You want to increase your max-buffers. > > > > I set max-buffers and max-epoch-size to 8000 but I'm still having issues. > r1-U filled up almost 75% of it's 200M proxy buffer. During that time the > pe for r1-U kept climbing and all the disk writes were put on hold until it > could start catching up. I can tell this because my virtual machines all > freeze up and write latency reported from ESXi skyrockets as well. I'm > still wondering if I'm not understanding some fundamental part of DRBD or > Proxy. I am expecting my third node to never affect the performance of my > primary nodes unless the Proxy buffers filled up completely. Is that not > the case? "never" certainly not. "much less", more like. depending on a lot of things. > > And have a look at the other hints on drbd performance tuning. > > > > I did but nothing else seems to apply. We will likely be able to help tune that thing. There is no "for best performance just use these settings". If there was, we'd made "those settings" the default. And proper tuning need understanding, understanding may need some time and careful observation. If you are not satisfied with the proxy, or need help with performance tuning, why not contact your LINBIT support rep directly? Lars -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com