I hate to reply to my own post, but someone is bound to point me to:
http://www.dell.com/downloads/global/power/ps2q05-20040286-Holmes-OE.pdf
And I read in that document:
"as traffic increases beyond
a single port’s bandwidth, the traffic does not spill over to the
next aggregated port. This type of load balancing is not possible
for link-aggregation groups based purely on Layer 2 information
because of the static nature of the calculation, and the requirement
to deliver Layer 3 packets in order. Thus, traffic for the same pair
of source and destination addresses always travels on the same
link, even if the link is oversubscribed."
Does this mean that there is no way of combining multiple links to get more bandwidth? If server A has 2 gigabit uplinks and server B has 2 gigabit uplinks, how is it possible for them to actually achieve throughput greater than 1 gigabit speeds?
LAG does provide you the ability to increase the overall BW between two switches. If you have a radom set of sources and a random set of destinations and you have traffic randomly distributed between these sources and destinations, some of the traffic will be hashed to a member link of the LAG and some will be on another member link. So over all the traffic between these two switches are randomly distribtued among all the member links in the LAG. So the more random a traffic pattern the better your distribution. What you will not get is if you have all traffic from a single source to a single destination then that traffic stream will not exceed 1Gig on a given link because all traffic from same source to same destination must go over the same physical link.
So really you are getting statistical load balance. So over a large sample or random packets from random source/destination we do get good distribution and will have increase BW between the two switches.
Now if you didn't use a LAG you would only be able to connect one physical wire between two switches otherwise you would get a bridging loop which would get disabled by spanning-tree. So LAG is the only way you could get any BW increase over the BW of the fastest link on the switch.
spacemky
2 Posts
0
August 10th, 2006 15:00
http://www.dell.com/downloads/global/power/ps2q05-20040286-Holmes-OE.pdf
And I read in that document:
"as traffic increases beyond
a single port’s bandwidth, the traffic does not spill over to the
next aggregated port. This type of load balancing is not possible
for link-aggregation groups based purely on Layer 2 information
because of the static nature of the calculation, and the requirement
to deliver Layer 3 packets in order. Thus, traffic for the same pair
of source and destination addresses always travels on the same
link, even if the link is oversubscribed."
Does this mean that there is no way of combining multiple links to get more bandwidth? If server A has 2 gigabit uplinks and server B has 2 gigabit uplinks, how is it possible for them to actually achieve throughput greater than 1 gigabit speeds?
Thanks!
DELL-Cuong N.
1K Posts
0
August 17th, 2006 18:00
LAG does provide you the ability to increase the overall BW between two switches. If you have a radom set of sources and a random set of destinations and you have traffic randomly distributed between these sources and destinations, some of the traffic will be hashed to a member link of the LAG and some will be on another member link. So over all the traffic between these two switches are randomly distribtued among all the member links in the LAG. So the more random a traffic pattern the better your distribution. What you will not get is if you have all traffic from a single source to a single destination then that traffic stream will not exceed 1Gig on a given link because all traffic from same source to same destination must go over the same physical link.
So really you are getting statistical load balance. So over a large sample or random packets from random source/destination we do get good distribution and will have increase BW between the two switches.
Now if you didn't use a LAG you would only be able to connect one physical wire between two switches otherwise you would get a bridging loop which would get disabled by spanning-tree. So LAG is the only way you could get any BW increase over the BW of the fastest link on the switch.
Cuong.