Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-08-01 22:38:44 +0200 \ Bernd Schubert: [ drbd proto A nfs export with sync seems to work fine now ] great. but you know that in case of a real failover, you may lose some of the most recent changes, even though your applications may think it had reached stable storage. rought estimate: 2*sndbuf size + latency*average throughput ... for home directories this may be good enough, though. > PS: Lars, do you want me to redo the latency tests, now with protocol A? maybe redo that with record size 4K, which will give the average latency for single blocks, so a good estimate for a round trip time. iozone -i0 -s 4M -r 4k -Q [-o -e -c] the same with large record size (-r 1M) does not say too much, just by comparing, one can see that linear streaming will do much better than wait for the commit of every single block... my hope for proto A would be that latency is near the unconnected values, which should be virtually identical to the values without drbd. Lars Ellenberg -- please use the "List-Reply" function of your email client.