num_poolagents doesn't add up

[selma]

Hi Everyone:

I was reading the IBM article "The DB2 UDB Memory Model" and it talks
about for aix, one memory segment 256mb is dedicated to agent private
memory. That seems very low if you have many concurrent users. Also as
each agent completes its task the agent private memory is not free up
but remains in the num_poolagent pool for quick restart. Fine and good
until I add up how much that is. We are using memmaxfree (8mb) and
memdisclaim as recommended, and keep the num_poolagent at a modest 60.
But having read that only one 256 mb segment is dedicated for all
agents, 60 times 8 mb is 480mb which is bigger than the entire Agent
Private Memory segment not counting appleheap...etc. Would someone
please clarify this situation? I must be getting wrong information.

Incidently in our OLTP environment no more than 4 or 5 agents would be
active at any moment. Most tasks are short and done in 1-2 sec. Longer
ones are 8 to 10 sec but less used. Is num_poolagent too big? I just
simply count up average no of connected application.

Thank you in advance.
Selma

[Fan Ruo Xin]

"selma" <se*********@yahoo.com> wrote in message
news:11**********************@z14g2000cwz.googlegr oups.com...

Hi Everyone:

I was reading the IBM article "The DB2 UDB Memory Model" and it talks
about for aix, one memory segment 256mb is dedicated to agent private
memory. That seems very low if you have many concurrent users. Also as
each agent completes its task the agent private memory is not free up
but remains in the num_poolagent pool for quick restart. Fine and good
until I add up how much that is. We are using memmaxfree (8mb) and
memdisclaim as recommended, and keep the num_poolagent at a modest 60.
But having read that only one 256 mb segment is dedicated for all
agents, 60 times 8 mb is 480mb which is bigger than the entire Agent
Private Memory segment not counting appleheap...etc. Would someone
please clarify this situation? I must be getting wrong information.
===========
The more concurrent users does not mean the more private memory each agent
will use. On aix, each agent can use up to 256MB private memory. Once
allocated for one db2 agent, it can't be shared by the other db2 agents. The
shared memory is db2 buffer pool, not each agent private memory. This is why
db2 private sort and shared sort go to different place.
Incidently in our OLTP environment no more than 4 or 5 agents would be
active at any moment. Most tasks are short and done in 1-2 sec. Longer
ones are 8 to 10 sec but less used. Is num_poolagent too big? I just
simply count up average no of connected application.

Thank you in advance.
Selma

[Pierre Saint-Jacques]

Ruo Xin has answered part one.
As to num_poolagents at 60, as usual the answer is it depends.
1) If uni processor. Vlaue seems somewhat high as a connection has a
coordiantor agent from connect to until connect reset.
You also have to think about the idle connections. Your snapshot will
show you the high water of connections executing in the instance and
also the number of total remote/local connections. There's no need to
have many more pool agents than total number of connections.

2) With multi processors, each coordinator agents can dispatch subagents
so the number of agents initially available and also ket in the pool is
more important.

For an OLTP environment, you have few queries and therefore require
fewer add'l additional agents than the coordinators. With the numbers
you mention, I'd be comfortable with 20 agents in the pool unless your
snapshots tell you differently.
HTH, Pierre.

selma wrote:

Hi Everyone:

I was reading the IBM article "The DB2 UDB Memory Model" and it talks
about for aix, one memory segment 256mb is dedicated to agent private
memory. That seems very low if you have many concurrent users. Also as
each agent completes its task the agent private memory is not free up
but remains in the num_poolagent pool for quick restart. Fine and good
until I add up how much that is. We are using memmaxfree (8mb) and
memdisclaim as recommended, and keep the num_poolagent at a modest 60.
But having read that only one 256 mb segment is dedicated for all
agents, 60 times 8 mb is 480mb which is bigger than the entire Agent
Private Memory segment not counting appleheap...etc. Would someone
please clarify this situation? I must be getting wrong information.

Incidently in our OLTP environment no more than 4 or 5 agents would be
active at any moment. Most tasks are short and done in 1-2 sec. Longer
ones are 8 to 10 sec but less used. Is num_poolagent too big? I just
simply count up average no of connected application.

Thank you in advance.
Selma

--
Pierre Saint-Jacques - Reply to: sescons at attglobal dot net
IBM DB2 Cerified Solutions Expert - Administration
SES Consultants Inc.