WAAS Clustering is it possible?

Unanswered Question
Oct 12th, 2009

Is it possible to cluster WAAS systems? I have heard that is it but I am having trouble locating any documentation on how to do it. Any assistance would be appreciated.

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
dstolt Mon, 10/12/2009 - 11:13

You may want to clarify on what you mean by clustering WAAS systems... Cluster interception? Cluster management? What are you trying to accomplish?

Thanks,

Dan

zebula Mon, 10/12/2009 - 11:36

I presently have a head end 612 unit that is running in inline mode. The plan is to move it to WCCP mode. Also it is being taxed pretty good so we are looking to cluster it (if possible) with another 612, to provide some redundancy and to of load some of the load. I see where you can put them inline and when the first one reaches it's connection max the second one starts to on some connections.

dstolt Mon, 10/12/2009 - 16:40

You are correct, your plan of WCCP redirection would be preferred to load balance and provide redundancy, especially at the head end. You can have up to 32 WAEs in a WCCP cluster, so you will be able to scale with more boxes as needed.

I would start here...

http://www.cisco.com/en/US/docs/app_ntwk_services/waas/waas/v413/configuration/guide/traffic.html#wp1041490

There is a lot of information there, however if you have questions, please post them and we'll see if we can assist you with them. Depending on your routers or L3 switches that you have installed, WCCP will have slightly different configuations for traffic redirection and return.

Thanks,

Dan

zebula Wed, 10/21/2009 - 11:53

Dan,

Thanks for the link. I checked it out and it talks about it at I high level but I can not find any real details.

I did some more digging and reading and I see where you can have a bunch of WAEs in a cluster and associated with a WCCP router. It looks like there are 2 WAAS Assignment methods, HASH and MASK.

The HASH looks to sense all the WAEs out there and splits up the address buckets evenly among the WAEs.

The MASK method looks to do something similar with the address buckets. This method gives you more control over the number of buckets.

The questions I have now are how does the WCCP router using the HASH method to the WAEs make a determination of what to send to what WAE?

I have the same question for the MASK method. I know it takes the MASK into account and performs an AND function but what do the results mean.

Also with the MASK method, do you know of any real examples that I can look at?

For just a basic setup (2 WAEs) I do not see how all the traffic would not end up going to the first WAE.

Any help would be appreciated.

dstolt Sat, 10/24/2009 - 06:05

Steve,

It depends on the Cisco gear as to whether you should use hash or mask. Hardware based redirection does not support hash in hardware asics (All Catalyst based switches, 7600s, ASR1K, etc.) so Mask is highly recommended to keep your CPU down. Hash is primarily for software based routers (think ISR and 72xx). Some of the newer 12.4T code can also support Mask on the software routers, however most customers still use hash based load balancing for the software routers.

Hash load balancing uses 256 IP buckets and cannot be influenced by configurations. It is less adaptable, but works well for most smaller sites with multiple WAEs. Mask based load balancing uses a 6 bit default mask of 0x1741 which should be altered depending on your situation. I usually use something like 0x3 for edges and 0x300 or 0xF00 in DC core situations depending on how many WAEs I need to load balance. Basically the more bits in your mask give you more buckets to spread over your cluster, but will use more TCAM space on your switch/router. A Mask of 0x3 will load balance on the last 2 bits in your address while 0x300 will use the last 2 bits in the 3rd octet, etc.

The other thing you should be concerned with is using either L2 or GRE (default) redirection. Rule of thumb is hardware based redirection should use L2, and software based platforms should use GRE redirection. Exceptions to this are the 6500 SUP 32 or 720 and ASR1000 which can do either in hardware.

I hope this helps, it's a lot of information but there is a lot to cover in these situations. The biggest thing to remember is to intercept traffic going both directions (wccp 61 and wccp 62) and avoid re-intercepting traffic egressing the WAEs as that will cause you headaches with loops. I often use negotiated return or generic-gre return in situation where you can't segregate the egress traffic, but again, this is platform dependent.

Dan

Actions

This Discussion