Sun, Aug 09, 2015 at 11:26:05PM CEST, jcasale(a)activenetwerx.com wrote:
Hey guys,
Following the docs at Redhat [1] I am not seeing a change in throughput for a
pair of Intel nics in a LACP trunk with an HP switch explicitly set as LACP as well.
When a client starts a long nfs upload it saturates the team interface, however
with or without the tx_balancer directive, when a second client starts a similar
upload the bandwith drops equally between the clients. The runner config I
am using is straight out of the docs.
Any ideas?
I'm not sure I understand what is the issues? You expect that the second
client will use the second teamed interface so both clients would use
one teamed iface? That depends on hash function. But setting that up
so you achieve that would be very tricky/random.
The thing is that hashing and balancing works nicely for a bigger
number of flows with hash function set up according to the traffic.
Hope this helps.
Jiri