[DRBD-user] link aggregate ( bonding module )
igor at 3gnt.net
Thu Apr 23 15:28:17 CEST 2009
Lacp links don't aggregate bandwidth to the same destination mac
address. Use iperf to measure your link aggregation, and you will notice
you will not have 2Gbps link.
If you want that, you need to use bonding in mode 0, and do back 2 back
connection's in the both machines. If you put one switch in the midle,
you will get back to the same problem, because switch don't send traffic
to the same mac address in two different ports (at least all i have tested).
PS: I have played a lot with this kind of setups, I have in production
some servers with 4x 1Gbps intel nic's, and it sucks. You should really
buy 10Gbps NIC's, but first test your I/O very well too see if your
Daniel Bourque wrote:
> My disk writes are benching 2x faster than the GigE dedicated link
> connecting the DRBD nodes.
> [root at host1 ]# ./Bonnie -s 1000 -d /var/tmp/
> Bonnie 1.4: File '/var/tmp//Bonnie.9328', size: 1048576000, volumes: 1
> Writing with putc()... done: 50862 kB/s 99.7 %CPU
> Rewriting... done: 732229 kB/s 100.0 %CPU
> Writing intelligently... done: 248780 kB/s 54.9 %CPU
> Reading with getc()... done: 43119 kB/s 99.9 %CPU
> Reading intelligently... done: 2002381 kB/s 99.9 %CPU
> Thus, the DRBD link creates a bottleneck for disk writes. I would like
> to use 802.3ad Dynamic link aggregation but looks like DRBD works off 2
> unidirectional tcp/ip sessions, making it unable to max out the 2 gigE link.
> Does anyone have any suggestion ( other than buying a pair of 10gigE
> NICs ) ?
Igor Neves <igor.neves at 3gnt.net>
3GNTW - Tecnologias de Informação, Lda
SIP: igor at 3gnt.net JID: igor at 3gnt.net
ICQ: 249075444 MSN: igor at 3gnt.net
TLM: 00351914503611 PSTN: 00351252377120
More information about the drbd-user