Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2006-08-16 23:06:12 -0700 \ khaja mohideen: > Hi, > > Indeed Protocol A(write speed 10MB/Min) has given > double the write speed over Protocol C(write speed 6 > MB/min). But still the write is slow. Any way the WAN > Link is only 2 Mbps. I would try to increase the WAN > link speed. I am going to use this across IDC. So the > link will be burstable(Uplink 10 Mbps) i think. > > short write burst - how to increase/decrease this > value. local cache for most writes. socket send buffer size, as soon as it actually gets submitted to drbd. thats it. again: protocol A is cannot make small bandwith link apear as a gigabit link. protocol A is intended to make a high latency link (regardless of bandwidth) feel less sluggish by not requiring a full roundtrip for each write request. bandwidth and latency are not the same thing, and cause different problems, though the symptoms may look and feel somewhat similar sometimes. > Do I need to increase the RAM. Currently it has 2 GB > RAM with dual processor. even more ram will not make the replication link faster. it may cache more, and thus delay the "flush to stable storage" for writes, so it probably will lessen the slow down effect of the replication bandwith. but not eliminate it. and how much the impact will be is heavily dependant on your access pattern. > Is there any other parameter to fine tune, so that well, its more like coarse tune: up the replication link bandwidth. if you have 2 MBit link, you can only get sustained write rate of about 200KB per second plus protocol overhead and stuff for a dedicated link. thats 12 MByte/min theoretical maximum for single thread streaming only sequential access. you say you got 10MByte/min. You are on the limit of your replication link. -- : Lars Ellenberg Tel +43-1-8178292-0 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Schoenbrunner Str. 244, A-1120 Vienna/Europe http://www.linbit.com : __ please use the "List-Reply" function of your email client.