WAAS: WCCP Mask or Hash on Routers?

Unanswered Question
Feb 19th, 2010

I'm starting thinking about using mask assign on an ISR router running 12:4(24)T with GRE/GRE. Has anyone done this before and can you use mask assign with GRE/GRE? We need to use it with GRE/GRE because our egress method has to be WCCP return. My thought was mask assign will be much better at load balancing across multiple WAEs in a cluster than hash because you can specify a long mask assignment. Right now, see more load on WAE than the other and are sometimes getting TFO overload.

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
Zach Seils Wed, 02/24/2010 - 06:41

Although it is possible to configure mask assignment on the ISR in recent versions of IOS, hash assignment is still recommended for the best performance.  Mask assignment was designed specifically for hardware-based platforms such as the Catalyst and Nexus series switches.

Regards,

Zach

timgunderson Thu, 03/07/2013 - 11:08

Zach,

I see that your response was more than 3 years ago on this question.  Current documentation (link below) seems to recommend Mask assignment.  Is the current recommendation on the ISRs to do L2 with Mask assignment?   What about instances where GRE redirection is required, then what is the appropriate assignment method, Hash or Mask?

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

aaedward Thu, 03/07/2013 - 13:54

The page you linked contains recommendations (in bold) for each platform. On the ISR G2 specifically, you should be able to use any combination of GRE/L2 and MASK/HASH assignment. Some other platforms require specific disribution and redirection methods to maintain the hardware acceleration of WCCP traffic. However, the ISR G2 does not have this requirement.

WCCP GRE and HASH distribution on ISR G2 is typically recommended to make deployment easier. With GRE, content devices can be an L3 hop away (if needed), and it reduces the chance of customers accidentally creating a WCCP redirect loop.

L2 distribution and HASH redirection method should typically require the least CPU and memory load on the ISR. These should perform the best in most cases.

The MASK distribution method gives better controls on how load is divided between multiple content devices, typically at the cost of more CPU and memory utilization. If you have only one or two content devices in your cluster, typically HASH will meet the need for slightly less CPU. As Zach said, most times MASK is used on the Datacenter side to give the ability to 'tweak' how the load is distributed across multiple devices.

Thanks,

Aaron

Actions

This Discussion