Load balancing question

["Sharma, Siddharth" <Si...@Staples.com>]

We have a 12 (tomcat 5.5.16) instance cluster on 4 physical machines running
redhat linux. There is an Apache installed on each machine (4 mod_jks). Each
mod_jk's worker.properties is configured exactly the same way i.e. each
mod-jk load-balances across all 12 tomcat instances. All instances have the
same weights in each mod_jk and sticky sessions are enabled.
>From Rainer Jung's email, I understand now that there are three
load-balancing algorithms and 'busyness' seems to be the default.
But with this algorithm, the workload distribution is uneven during a load
test.
The reason I can think of is that since there are 4 mod_jk load balancers
with no knowledge of each other's counts, the distribution is getting
screwed up. Is this a fair assessment? Is there a way to solve it?
We do need this redundancy in load balancers, so cannot get rid of any.


---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org

Re: Load balancing question

[Rainer Jung <ra...@kippdata.de>]

Hi,

first of all I assume you are using version 1.2.18 of mod_jk? Otherwise
please update.

Sharma, Siddharth schrieb:
> We have a 12 (tomcat 5.5.16) instance cluster on 4 physical machines running
> redhat linux. There is an Apache installed on each machine (4 mod_jks). Each
> mod_jk's worker.properties is configured exactly the same way i.e. each
> mod-jk load-balances across all 12 tomcat instances. All instances have the
> same weights in each mod_jk and sticky sessions are enabled.
> From Rainer Jung's email, I understand now that there are three
> load-balancing algorithms and 'busyness' seems to be the default.

Default is "Requests". You can set the method via the "method" attribute
of your lb worker in workers.properties, e.g.:

worker.mylbworker.method=B

> But with this algorithm, the workload distribution is uneven during a load
> test.

Now I don't know, if you use the Request method (default) or really
Busyness.

> The reason I can think of is that since there are 4 mod_jk load balancers
> with no knowledge of each other's counts, the distribution is getting
> screwed up. Is this a fair assessment? Is there a way to solve it?
> We do need this redundancy in load balancers, so cannot get rid of any.

You should configure the mod_jk status worker for the 4 Apaches. Then
have a look at the status page of mod_jk and try to understand the load
numbers there.

Stickyness is a natural conflict to load-balancing. When only a couple
of sessions are active, stickyness can easily lead to uneven
distributions. Only with increasing traffic, these effects should
statistically go away.


Regards,

Rainer

---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org