Mask Calculation's and Performance

Ryan Posey · ‎09-10-2011

We will have two 7341 WAE's in our Data Center. We have several hundred servers that branch sites pull (and sometimes push "SharePoint") centralized resources from. To even out the load, we are moving Service Group 62 to the WAN interface and 61 to the LAN interface on the Data Center Router. I would also like to create a "Mask" that effectivly doubles my DRE capacity by alternating WAE's based on the destinations last octet. Using a mask calculation that alternates 7341 boxes based on the last octet of the IP address seems to be the most effective method. Are their any performance hits or faults with this method of load distribution?

Using Cisco's "WCCP-Mask-calc-v103", I came up with 0x01 as the value to alternate based on the host address.

Is that the proper mask to alternate hosts based on their last octet?

Are their any inherited issues with using this mechanism?

Has anthing changed with 4.4.1 / 4.4.3 that would render this inoperable or unsupported?

Will this eat up resources on the router?

7341#1

10.0.0.0

10.0.0.2

10.0.0.4

10.0.0.6

7341#2

10.0.0.1

10.0.0.3

10.0.0.5

10.0.0.7

ktunugun · ‎09-13-2011

Hi Ryan,

This mechanism is fine and hold true for 4.4.1/4.4.3 as well.

However it is purely based on the requirement whether you want to load balance based on the last octet or the last but one octet. If you have multiple sites of Class C subnets you may probably want that all the users from one site should always go to the same WAE in datacentre so that it provides higher cache hit ratios quickly. If you have less number of sites and you may want to do the load balancing on a per user basis in which case both the WAEs DRE will be populated with the same kind of cache.

Hence it depends on the number of sites / users etc based on which you can decide on a mask value for load balancing.

Please refer to the below URL for the recommended assignment method on different platforms.

http://www.cisco.com/en/US/prod/collateral/contnetw/ps5680/ps6870/white_paper_c11-608042.html

Hope this clarifies.

Regards

Kiran.