Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Tue, May 04, 2004 at 04:41:44PM +0300, dhz_bf at walla.co.il wrote: > > I have performed an experiment to get some idea about the overhead of > drbd when its partner (the secondary node) is down. > I am using version 0.6.12 of drbd. > > The drbd has been in CState of WFConnection. > > I have performed a certain operation on the disk and it tool about 5.5 > minutes when drbd was loaded and "owned" the disk on which the > operation was performed. The operation involved tar and gzip. > > I then took drbd down and mounted the partitions drbd "owns" directly. > I retried the operation and it took 4.5 minutes. > > There was an additional factor that the disk drbd "owns" contains the > file system of a number of remote machines. These machines were idle > at the time of the experiment. > > I looked at the code and it seemed to me like the overhead of drbd > when the secondary node is down (and CState < Connected) is pretty > small. Yet, the times I measured with this experiment show about 25% > overhead of drbd. > > Do these results look like the ones you got in your experiments? > Is this level of overhead reasonable? not at all. definetly a chaching issue in your "experimental setup". drbd "overhead" in unconnected state is typically within the uncertainty of the measurement. Lars Ellenberg