Link Aggregation

[richard.bumanlag]

Hi,

I am planning to purchase a 48 port 10/100 switch and 2 Gigabit Uplink ports. Just want to ask what is the computation for Link Aggregation? For me to be able to know if I will consume the Aggregated Link of the two Uplink Ports. Just curious on the computation for me to know if I am using the full capacity of the aggregat link or even just 1 GB uplink is enough.  Or should I find a 48 port 10/100 with 4 1GB uplink port and aggregate this uplink ports. I need to justify the purchase with computations.

Thanks

[Jo Kern]

Hi,

we do follow the standards

http://en.wikipedia.org/wiki/IEEE_802.1AX

to avoid misordering of packets the balancing is done based on sessions or flows  (identified by MAC/IP addresses).

Statistically your sessions will then be spread across all members in the trunk group, so the bandwidth is fully used (almost)

Cheers

Jo

[richard.bumanlag]

Hi,

Actually im aware that ling aggregation will have load balancing and failover. But the thing is I need to justify why do I need to buy a 48 port 10/100 with 2 uplink ports or do I need the 4 uplink ports?

That is why a computation with link aggregation for 48 ports will be a big help for me.

Thanks

[Edwin Summers]

Richard,

Just so I understand the question correctly, you're looking for information on uplink capacity vs load to justify the purchase, correct?

Can you share some contextual information?  In other words, what function will the switch be performing, where is it located in the topology, what is passing through it, etc.

[Jo Kern]

That depends who you have to justify it to.

20 ports 100 mbps = 2 gpbs

40 ports 100 mbps = 4 gbps

If you really looking for traffic engineering than like Edwin says there is other info required than the algorythm on how the LAG spreads traffic flows across the links.

If you read the article you see that the balancing of the 48 ports is based on values (MAC/IP address hashes, so statistically you will not reach 50-50 load balancing , so when one trunk link reaches 100% the other has still capacity avilable but traffic on the congested link will be discarded already.

The article says 70-30 distribution between 2 links.

It really depends on the the number of devices you connect to the 48ports.

Statistically you gain best load balancing the more devices you have.

Depending on your traffic not all devices will send constantly. And you need to configure QoS as well to make it work for your applications because if you want to be more precise you need numbers on actual throughput and that requires knowledge of the traffic type (TCP, RTP etc.) the typical throuput you actually  get on a 100mbps based on a device connected, etc.