Rick Barnes wrote:
> Since my switch *does* support 802.3ad, including layers 2,3 and 4
> hashing, should I use mode=4? Or would one of the other modes be
> "better" for providing fail-over and link-aggregation, specifically
> balance-tlb or balance-alb.
Use 802.3ad, also if performance is really an issue I'd suggest
setting up two networks on the NFS server one with standard frame
sizes and the other with jumbo frames, and directly connect systems
that need higher throughput to a dedicated VLAN(s) running jumbo
frames.
Also note that link aggregation will not increase throughput between
hosts, i.e. if you have 1 host talking to 1 server you will not
get higher throughput with link aggregation, what link aggregation
will do is allow many hosts to communicate to a single host with
higher aggregate throughput.
If you want faster, simpler single-stream throughput go with
10GbE, good quality line rate 10GbE switches are very cost
effective these days, probably 90%+ cheaper than 10GbE was
5-6 years ago.
Though your likely going to need a lot of disks to be able to
come close to saturating even a 1Gbps connection depending on
your workload.
nate
_______________________________________________
CentOS mailing list
CentOS@xxxxxxxxxx
http://lists.centos.org/mailman/listinfo/centos
