Assignment of WAVEs to Buckets WCCPv2

[j.a.m.e.s]

All,

Can anyone explain how WAVEs (caches in WCCP lexicon) are assigned to buckets please?

The context to my question is probably best illustrated by an example:

* You have two CEs (CE-A and CE-B), each with a provider circuit

* Asymmetric routing across the WAN providers

* Four WAVE units registered using WCCPv2

* An assignment mask of 0x0f00 on the source ip (16 bits)

* Therefore, there are four caches and 16 buckets (4 buckets per cache)

How does WCCPv2 ensure buckets 1-4 get assigned to WAVE-A on both CEs?

According to the post below, the assignment is based on WCCP registration time, but I am not confident in that, else rebooting WAVEs or CEs would cause loss of optimisation (unless they were all rebooted together in sequence):

https://supportforums.cisco.com/thread/2225305

This seems like a fundamental part of the WCCP operation, but I can't find any documentation for it.

Many thanks for anyone who can assist.

Regards

James.

[finn.poulsen · Accepted solution]

Hi James,

Please take a look here :

http://www.cisco.com/en/US/docs/ios/12_2/configfun/configuration/guide/fcf018_ps1835_TSD_Products_Configuration_Guide_Chapter.html#wp1000909

especially bullet point 3, where one of the WAVEs is elected/selected as the lead, and that WAVE controls the bucket assignments.

Know it's an (old) IOS 12.2 document, but that hasen't changed!

What you want to acheive is that you always redirect to the same WAVE from both routers, based on some common criteria.

So if you reboot one of the routers (CEs) the other one has to handle all the traffic, but should still redirect to the same WAVE.

If you reboot one of the WAVEs, this WAVE will deregister and the lead WAVE will inform everyone about the new assignment buckets.If it's the lead WAVE that is rebooted another lead WAVE is selected.

Off course TCP sessions being redirected to the rebooting WAVE will be disrupted and another TCP session will be set up from the client and eventually being redirected to one of the remaining WAVEs.

Hope this answers you question.

Best regards

Finn Poulsen

[j.a.m.e.s]

Hi Finn,

That's a great answer, thank you. I also found a useful reference in the draft RFC (section 4.8.2 "Mask/Value Sets") and a cloudshark trace showing the Redirect-Assign message containing a buckets-to-cache map.

Regards

James.