> From: opensolaris-discuss-boun...@opensolaris.org [mailto:opensolaris-
> discuss-boun...@opensolaris.org] On Behalf Of valrh...@gmail.com
> 
> I actually have two ethernet ports on the server, so in principle I
> should be able to use automatic link-aggregation in OSOL to do this,
> right? If I understand correctly, the two adapters get teamed, and only
> require a single IP address, right?

You're 95% right.  Here are the caveats you'll want to know about:

With LACP, each data stream can only go as fast as a single network
interface.  So if you have aggregated 2x 1G adapters, you won't see 2G
speeds.  But you might be able to get 2 separate clients to each talk 1G
with the server.

Your switch must support it too.  Configuring it (at least in sol10, but
maybe it's different in osol) is a bitch.  Because you have to unplumb your
interface in order to add it to an aggregation group with another interface,
and solaris behaves *very* poorly when there is no plumbed network
interface.  

When one port is fully utilized, and another client comes along and tries to
establish another connection, and that client already had a data stream on
the same port which is already fully utilized, so the system has a new
desire to move one of the datastreams to the unutilized network interface
...  This is where I have experienced problems.  And it happens all the time
(like literally all the time) because that's the whole point of even
bothering to implement it.  LACP doesn't know that one of the datastreams
should be re-assigned to the other interface, until the traffic appears on
the overloaded interface.  When LACP decides to reassign the traffic to the
other interface, an error packet is issued, and the previously received
packet must be re-transmitted.  You see this occurring by watching the
interface error counter increasing.

Lots of traffic will not care about the error counter.  Lots of people use
LACP and don't even notice a problem.  But I do file transfers, and ssh
tunnels all over the place, and these error occurrences cause dropped and
stalled connections.  

My advice to you is:  Watch the error counter on your network interfaces,
and if it's anything other than zero, abandon LACP.


_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org

Reply via email to