Does someone knows if there is a way to limit the bandwidth on a network interface in AIX?
I have done that in linux using iptables and tc
NOTICE: developerWorks Community will be offline May 29-30, 2015 while we upgrade to the latest version of IBM Connections. For more information, read our upgrade FAQ.
This topic has been locked.
4 replies Latest Post - 2013-01-17T17:09:26Z by jklotz
Pinned topic Limit network bandwith
Answered question This question has been answered.
Unanswered question This question has not been answered yet.
Updated on 2013-01-17T17:09:26Z at 2013-01-17T17:09:26Z by jklotz
nagger 100000MRSJ342 Posts
CristianLagos 27000414H18 Posts
tony.evans 0600007X8X412 PostsACCEPTED ANSWER
Re: Limit network bandwith2011-09-30T12:32:01Z in response to CristianLagosUncertainty fuels fear.
I suggest you baseline and measure the environment, and show if the LPAR is using all the bandwidth of the enclosure first. If it is, we should assume it needs to, or it's doing stuff it shouldn't be, in which case you either need more bandwidth or you need to fix the problem.
I assume you have a blade with a bunch of micro-partitions? What's the network config? How many physical adapters, what speeds, how many virtual adapters, how is it laid out (SEA's, HEA's, etc?)
jklotz 2700021VG426 PostsACCEPTED ANSWER
Re: Limit network bandwith2013-01-17T17:09:26Z in response to tony.evansI don't have an answer to the question (I'd like to), but being able to limit network bandwith is a very valid point.
First, a similar tool exist for disk bandwith (workload manager).
Second, you never know how a system can go wrong (bug or bad user action), so it's realistic to encounter some scenarios where a partition could outbound so much traffic network that other actual critical partitions (think of a database partition used by SAP for example) could suffer from network bandith degradation.
A way to control a bit the network trafic would be NOT to optimize the IP stack of a partition (which means that if you need to give it a network boost, you have to interrupt the network stack for several minutes).
For example, I have a case (specific to my environment, so may or may not be reproduced in yours to the same scale) where a partition can have a 90/100 MB/s data rate with standard IP parameters (mtu=1500 and mtu_bypass left to off), and can reach 450 MB/s when communicating with another partition located on a second physical server, with a mtu of 9000 and the mtu_bypass set to on).