08-24-2009 10:36 PM
A simple setup using cache-enabled virtual server:
clients --- internet ---- ServerIron --- real servers
Client traffic to the virtual servers is getting send to the caches first of all (reverse proxies) to get the load down at the reals. I do have 4 caches but the distribution is not very good. 2 of the caches are heavily loaded and 2 other are not that busy.
Is there any way to chance this?
08-25-2009 03:12 AM
The hash-mask is the following: hash-mask 255.255.255.255 255.255.255.255
We do have 4 virtual servers configured which are cache-enabled.
08-25-2009 05:22 AM
The traffic distribution in your case is based on 32-bits of the destination IP and 32-bits of the source IP. The syntax for the hash command is:
hash-mask <destination-ip-mask> <source-ip-mask>
You do not have a lot of destination IPs in your setup because the destination is always one of the virtual server and it looks like you do have 4 virtual servers only which are using the cache-enabled setting.
The destination IP is not really variable in your case but I guess the client are coming out of the internet and there are most probably tons of client IP address. All I can suggest is to play with the hash-mask.
My experience with the following setting is very good talking about some virtual servers only with a huge pool of client IPs:
hash-mask 0.0.0.0 255.255.255.255
This is going to look at the source IP only for the hash and traffic is getting distributed to the caches based on the source IP of every request. This works well in some of my setup - please try to change it and check the counters again.
The setting is getting activated as soon as you hit enter - please wait at least half an hour or maybe even longer because the stats are going to change slowly. Looking at the amount of current connections you should see a change very quickly. The amount of octects and packets is going to change slower because you do have base values now which are not perfectly load balanced.
08-25-2009 06:34 AM
It looks much better now - the amount of current connections is pretty equal across all caches. I am going to monitor it for a few more days but I guess this was the solution for my problem. Thanks oadam!