2011-01-26 01:42 AM
There is a FAS2040A running with Data ONTAP 8.0.1RC3.
We config e0a, e0b, e0c, e0d with static multimode interface thourghput into a single ifgrp.
Each e0X is GbE interface.
What the throughput of ifgrp should be? 4Gb or 2Gb?
Cuz we heard that Data ONTAP use 2 interfaces for inbound and the other 2 interfaces for outbound,
(we did a test the result shows the throughput more like 2Gb.)
If this is ture? have any document to talk about this?
Or the ifgrp should have 4Gb throughput?
Thanks in advance.
2011-02-02 01:22 PM
I think it depends on how your switches are configured. Basically what you are creating here is a network trunk and all interfaces will work as one interface giving you higher throughput. Therefore your assumption of a 4GB connection is correct if your switches meet the IEEE 802.3ad standard for LACP.
Here is what our documentation says about static multimode interfaces:
"The static multimode interface group implementation in Data ONTAP is in compliance with IEEE 802.3ad (static). Any switch that supports aggregates, but does not have control packet exchange for configuring an aggregate, can be used with static multimode interface groups."
"A static multimode interface group requires a switch that supports link aggregation over multiple switch ports. The switch is configured so that all ports to which links of an interface group are connected are part of a single logical port. Some switches might not support link aggregation of ports configured for jumbo frames. For more information, see your switch vendor's documentation."
If "Interfaces e0, e1, e2, and e3 are part of the MultiTrunk1 multimode interface group. All four interfaces in the MultiTrunk1 multimode interface group are active."
"You can use round-robin for load balancing multimode interface groups. You should use the round-robin option for load balancing a single connection's traffic across multiple links to increase single connection throughput. However, this method might cause out-of-order packet delivery. If the remote TCP endpoints do not handle TCP reassembly correctly or lack enough memory to store out-of-order packets, they might be forced to drop packets. Therefore, this can lead to unnecessary retransmissions from the storage controller"
Here is a few links to the documentation I reference: