Yes, that's what we do with the direct connection.
The network cables were tested before with a Fluke DTX-1800 CAT6a testing
procedure to eliminate bad cabling as a reason for bad performance.
We have tested also without bonding on a single Intel NIC
Same result.
As mentionend in the first posting, replication through a 3com NIC
delivers poor 13MB/s...
Estimated replication speed of a single task would be about 100MB/s constantly.
We also do have a production setup of DSS-V6 with HP-hardware
with NC375i and NC360T NICs and smart array P410i,
overall replication speed is as estimated with 3 tasks running simultaneously
over a 802.3ad bond at about 180MB/s.
So this seems to be an issue how drdb handles the Intel or 3com NICs,
since the ISCSI performance without any replication is very OK on all NICs,
if we use the machine as standalone.
BTW, we know, that a bond will not increase performance for a single replication task,
but it will increase available bandwith for more than one replication task running,
as we can see in our production environment.
Reply With Quote