Load Balancing Docbase not run equally.

Experts,

We installed a content server and the docbase is split into three.

As this is load balancing environment - theoretically, three docbases should be sharing the load equally; however, based on performance monitoring, we found that one of the docbase is always on high usage while another two docbases are running as minimum capacity.

My question here is how to balance the load equally for these three docbases? Is there any guidance available?

Thanks for your help.

Find more posts tagged with

Comments

bacham2

You got the terminology wrong. You probably mean that you have one docbase with 3 Content Server (CS) instances managing it.

If all three CS project with the same proximity to the connection broker, then the load should be distributed more or less evenly. If you have a low load, it is possible that requests go more frequently to one CS due to DFC session pooling.

LawrenceLow

Hi, yes, you are right! Thanks for your correction.

So, is there any way to check it out or how to fine-tune it?

Thanks for your help.

bacham2

There's nothing to fine tune. If you want to check that your CS project to the connection broker with the same proximities, you can use the dmqdocbroker -i command on the CS box.

PanfilovAB

we found that one of the docbase is always on high usage while another two docbases are running as minimum capacity.

It can depend on workflow settings, e.g. wf_agent_notify_newtask=T can cause that almost all workflows will execute only on one content server. In my scenario the number of active sessions differs slightly across content servers:

29-05-2013 6-47-36 PM.png

LawrenceLow

Hi,

Thanks for your explanation. But I am newbie to CS....can you provide step-by-step how to use the dmqdocbroker -i command on the CS box?

Again, thanks for your help.

Thanks.

Regards,

Lawrence

LawrenceLow

Hi PanfilovAB,

How to prove that it is causing by workflow setting? Is there any way to show it - so that we can justify to the management.

Thanks.

Regards,

Lawrence

PanfilovAB

I use some software, that monitors different metrics. For example next graph shows auto-activity distribution across to servers where wf_agent_notify_newtask=T

next one shows distribution across four servers where wf_agent_notify_newtask=F

5-06-2013 1-54-56 AM.png

Still not evenly, but much better.

You can check auto-activity distribution by executing following query (i shows how many activities every server run during last 14 days):

SELECT co.object_name, COUNT (wi.r_object_id)

FROM dmi_workitem wi, dm_sysobject (all) co

WHERE r_runtime_state = 2

AND r_auto_method_id is not nullid

AND co.r_object_id = wi.a_wq_name

AND wi.r_creation_date > DATEADD(day,-14,date(TODAY))

GROUP BY co.object_name

Actually, that query has some systematic error, because it does not count finished workflows, but I think it's ok.

Further, if workflow settings is a root cause of your "problem", do not hurry up to set wf_agent_notify_newtask=F, because it has some side effects, and in most cases it's better to separate servers by roles: one executes workflows only, another serve http-users.

LawrenceLow

Hi PanfilovAB,

Thanks for the syntax, seem like workflow setting is the root cause.

By the way, may I know what is the "side-effect" might be happened? Instead of separate servers by roles - is there any other way to resolve it?

Thanks.