On 2015-02-22 07:01 AM, Josh Grosse wrote:
On Sun, Feb 22, 2015 at 12:47:49PM +0000, Stuart Henderson wrote:
Are you clear on exactly what loadbalance does?
A hash of
the protocol header is used to maintain packet ordering.
The hash includes the Ethernet source and destination
address, and, if available, the VLAN tag, and the IP source
and destination address.
In this mode, traffic between two individual IP/MAC addresses on
the same vlan will only use a single port, you need multiple addresses
to see the traffic balanced between multiple ports.
I think that's clear. The problem experienced was half the TCP sessions
failing to connect. It's unclear why, as my testing was limited to
flood pings and tcpbench. Both of which appeared flawless. I'm running
roundrobin in production, now, and will schedule more test time for
loadbalance ... with multiple generators.
Logically, if roundrobin works, loadbalance should, too.
According to trunk(4), the only difference between them is the selection
algorithm that controls which outbound interface to use.
My first thought was the switch is doing some sort of MAC pinning as as
security feature, but that would have also defeated roundrobin.
Technically, it doesn't matter in this circumstance (barring strange
implementation issues on the switch) as both will produce similar
results on a DGS-1100 because all ports on the switch are limited to
1Gbps no matter what... so the fact that loadbalance is limited to 1Gbps
per stream and roundrobin 2Gbps per stream becomes irrelevant.
But it's still strange that one works and one doesn't.
--
-Adam Thompson
[email protected]
+1 (204) 291-7950 - cell
+1 (204) 489-6515 - fax
