cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1754
Views
0
Helpful
4
Replies

Most Efficient WCCP Mask for 3 x WAE

anthony.brewley
Level 1
Level 1

Hi

I currently have a DC WAAS farm/cluster of 2 x WAE-7341's. 

I use the following for load balancing:

wccp tcp-promiscuous service-pair 61 62 mask src-ip-mask 0x300 dst-ip-mask 0x0

Which is working well and the traffic seems to be well balanced across both WAE's. 

My remote end sites all use /24 subnets.  I believe this mask above splits the data into 4 buckets based on the third octet, with two buckets being assigned to WAE#1 and two buckets being assigned to WAE#2.

Now I wish to add in a third DC WAE to the cluster. 

If I wish to continue to load balance based on the third octet but now with 3 WAE's, what is the best/most efficient mask to use?

regards

Anthony

2 Accepted Solutions

Accepted Solutions

Daniel Arrondo Ostiz
Cisco Employee
Cisco Employee

Hi Anthony,

I'm afraid the most efficient mask calculation is a very tricky thing, and it greatly depends on your environment. Let me just give you a couple of pointers on what you should consider.

The first thing to take into account is the bucket quantity and how they will split among the WAEs. With just 4 buckets, it would be 2 for one WAE and 1 for each of the other two. If you increase the mask to 3 bits (8 buckets) it would be "3,3,2", with 4 bits "6,5,5". Since you have an odd number of WAE and the bucket number is always even, there will always be some inbalance, but, the higher the mask, the more even the bucket distribution will be.

The second thing is the amount of traffic for each of the buckets. Even assuming that all the branches generate a similar amount of traffic (which is probably not true), you still need to take into account the number of branches you have and how many there will be per bucket. For example, if you for a 4-bit mask (16 buckets) with 30 branches, half of the buckets would match just one branch while the other half match 2.

In general, it's easier to define a mask with good distribution if the last octet is used instead of the third one, but of course, it has the drawback of not having a one-to-one relation between the branch and the WAE at the core location.

Regards

Daniel

View solution in original post

Hi Anthony,

The math behid the calculation is quite simple. The mask just defines the bits in the IP address that will be looked at. Each of the possible bit combinations is what it's called a bucket.

The distribution of the buckets between the different WAEs is done as part of the WCCP protocol, and unfortunately, it's not predictable. The only thing that can be know is that all the WAEs will get a even share of the total number of buckets.

You can use the Excel sheet I'm attaching to help you design best the mask for your setup. With it, you can easily calculate the IP to bucket association for different masks.

Regards

Daniel

View solution in original post

4 Replies 4

Daniel Arrondo Ostiz
Cisco Employee
Cisco Employee

Hi Anthony,

I'm afraid the most efficient mask calculation is a very tricky thing, and it greatly depends on your environment. Let me just give you a couple of pointers on what you should consider.

The first thing to take into account is the bucket quantity and how they will split among the WAEs. With just 4 buckets, it would be 2 for one WAE and 1 for each of the other two. If you increase the mask to 3 bits (8 buckets) it would be "3,3,2", with 4 bits "6,5,5". Since you have an odd number of WAE and the bucket number is always even, there will always be some inbalance, but, the higher the mask, the more even the bucket distribution will be.

The second thing is the amount of traffic for each of the buckets. Even assuming that all the branches generate a similar amount of traffic (which is probably not true), you still need to take into account the number of branches you have and how many there will be per bucket. For example, if you for a 4-bit mask (16 buckets) with 30 branches, half of the buckets would match just one branch while the other half match 2.

In general, it's easier to define a mask with good distribution if the last octet is used instead of the third one, but of course, it has the drawback of not having a one-to-one relation between the branch and the WAE at the core location.

Regards

Daniel

Daniel

Thanks sincerely for the reply, it is very much appreciated. 

I think I will have to accept that not all sites will generate as much traffic as each other but I would like to maintain the edge to core WAE relationship so that all traffic from a branch always goes to the same core WAE, simply because I believe that users within a location will be accessing the same or similar data.

So with that in mind, can you please explain to me the math behind how I would be able to get a '3,'3,2'  or a '6,5,5' split and also the mask I need to use.  My current mask (0x300) was 'given' to me but the math behind it wasn't explained in great detail and I'm struggling to work out how you arrive at the splits or how you know that 3 bits produces 8 buckets and 4 bits produces 16 buckets?

I'd like to be able to ultimately understand that if the third octet has value x, it will go to bucket #1, if y then bucket #2, etc.

From their I could look at my actual subnets and work out in advance if i will have an efficient enough balance.

thank you

Hi Anthony,

The math behid the calculation is quite simple. The mask just defines the bits in the IP address that will be looked at. Each of the possible bit combinations is what it's called a bucket.

The distribution of the buckets between the different WAEs is done as part of the WCCP protocol, and unfortunately, it's not predictable. The only thing that can be know is that all the WAEs will get a even share of the total number of buckets.

You can use the Excel sheet I'm attaching to help you design best the mask for your setup. With it, you can easily calculate the IP to bucket association for different masks.

Regards

Daniel

Daniel

I think the penny has just dropped for me, thank you very much. 

The xls doc is very very helpful.

regards

Anthony