Load balancing - WAAS

[Srin_G]

Hi all,

We got 2 x 674's in data center and use hash method for load balancing. Due to our IP address scheme, cisco's hash method puts most of the connections go to one WAE only. I know we can increase weight (currently 0) to get nearly 50-50 or 60-40 load balance, but i have no idea how to calculate the weight value. Currently it is 90-10 sharing! Any suggestions or doco is much appreciated.

Regards

Srini

[Amir Asfandyarov]

Hello Srini,

If you're stick to hash (and cannot use mask for some reason), then you're correct, you can use weights.

Couple of suggestions are there - https://supportforums.cisco.com/docs/DOC-21593#WCCP_best_practices_for_WAAS_deployment

Make sure that weight factors for individual devices are greater thane 100 - that will ensure complete "bucket" coverage in case one of the devices is down (that is, the remaining device will get 100 % load then).

When the sum of all weight factors is greater than 100, the specific percentage of buckets assigned to a specific WAAS device is the weight assigned to that WAAS device divided by the total weight and rounded up. Rounding up guarantees that each WAAS device will be assigned at least one bucket.

p.s. Still mask assignment gives you a bit more flexibility of load-balancing between devices in WCCP farm - see http://www.cisco.com/en/US/prod/collateral/contnetw/ps5680/ps6870/white_paper_c11-608042.html for recommended methods, depending on your HW (WCCP router/switch).

HTH,

Amir

[Srin_G]

Thanks Amir.

In that case, If i assign a weight value of 100 to WAE1 and 100 to WAE2 would i be able to get a 50-50 load balancing?

Cheers

Srini

[Amir Asfandyarov]

Hello Srini,

You will get 50/50 distribution of *buckets*, but not 50/50 distribution of *traffic*. Again, if you want to stick to hashing, you will need to find weights which will distribute *connections* (not buckets) more fairly. With hashing it is impossible to state that with particular weight you will get X/Y distribution of flows, I would say in this case it is a bit experimental.

Again, with mask you can be more predictive.

Regards, Amir

[Srin_G]

Thanks Amir.

Is there a mask calculator or formula for applying mask value? I will test it in the lab and will let you know. Just flat out here mate so couldn't reply early.

cheers

[Amir Asfandyarov · Accepted solution]

Hello Srini,

Take a look at this thread -

https://supportforums.cisco.com/thread/2153310, where my colleague Daniel gives some tips&tricks on mask calculation, you will also see there a simple xls file which will help you to calculate a better mask.

Also take a look at this one -

https://supportforums.cisco.com/thread/132907

Cheers,

Amir

[Srin_G]

Thanks mate. Appreciate your assistance.

regards