MEMORY Utilization / Buffer Cache issue

[Jojo Castro]

Hi all.

We have 2 rx8640 running as DB cluster.
During the initial startup when two servers join in a partition, memory utilization is same with 9%. When DB is opened on both nodes, 40% utilization on both nodes. But when we started our POC testings, memory utilization on node1 is higher than 5% than node2.

On my initial investigation, below are the informations i have gathered:

*Both nodes are running 95GB each
*Node1 Buf Cache: 4.8gb
*Node2 Buf Cache: 1.9gb
*kernel parameters are equal on both nodes

ipcs in node1:

# ipcs -m
IPC status from /dev/kmem as of Thu Nov 15 11:40:17 2007
T ID KEY MODE OWNER GROUP
Shared Memory:
m 0 0x411c07db --rw-rw-rw- root root
m 1 0x4e0c0002 --rw-rw-rw- root root
m 2 0x412038e0 --rw-rw-rw- root root
m 3 0x00a5c581 --rw------- sfmdb users
m 819204 0x4d4e5251 --rw-r--r-- root root
m 32773 0x55315352 --rw-rw-rw- root root
m 32774 0x44525354 --rw-r--r-- root root
m 32775 0x53494152 --rw-r--r-- root root
m 32776 0x00005643 --rw-rw-rw- root root
m 9 0x00005654 --rw-rw-rw- root root
m 983050 0x06347849 --rw-rw-rw- root root
m 65547 0x0c6629c9 --rw-r----- root root
m 131084 0x49180468 --rw-r--r-- root root
m 3014669 0xf8029b68 --rw-r----- oracle dba


ipcs in node2:
# ipcs -m
IPC status from /dev/kmem as of Thu Nov 15 11:40:34 2007
T ID KEY MODE OWNER GROUP
Shared Memory:
m 0 0x411c07db --rw-rw-rw- root root
m 1 0x4e0c0002 --rw-rw-rw- root root
m 2 0x412089bb --rw-rw-rw- root root
m 3 0x00a5c581 --rw------- sfmdb users
m 720900 0x4d4e5251 --rw-r--r-- root root
m 32773 0x55315352 --rw-rw-rw- root root
m 32774 0x44525354 --rw-r--r-- root root
m 32775 0x53494152 --rw-r--r-- root root
m 65544 0x00005643 --rw-rw-rw- root root
m 9 0x00005654 --rw-rw-rw- root root
m 1900554 0x06347849 --rw-rw-rw- root root
m 11 0x0c6629c9 --rw-r----- root root
m 65548 0x49180468 --rw-r--r-- root root
m 1507341 0xa0b335ec --rw-r----- oracle dba


Is the result in ipcs has something to do with the memory utilization and buf cache difference between the two nodes?

Please be informed that I only have one more week before putting this machines into production.

Appreciate your advises and answers.

Thanks in advance.

[A. Clay Stephenson]

What part of DYNAMIC buffer cache do you not understand? I would be very surprised if two identically configured and identically loaded (from the standpoint of processes running) would display identical buffer cache utilization. The whole point of dynamic buffer cache is that it is dynamic. The differences you are seeing are normal and expected. You could set dbc_min_pct = dbc_max_pct and thus fix the buffer cache.

[Jojo Castro]

What part of DYNAMIC buffer cache do you not understand? I would be very surprised if two identically configured and identically loaded (from the standpoint of processes running) would display identical buffer cache utilization. The whole point of dynamic buffer cache is that it is dynamic. The differences you are seeing are normal and expected. You could set dbc_min_pct = dbc_max_pct and thus fix the buffer cache.

===>
Yes it is identically configured (including dbc_min_pct & dbc_max_pct). Also, identically loaded from the standpoint of processes. My questions are, thus dynamic buffer cache usage has something to do with the memory usage? Why is it that node1 memory utilization is 5% higher than that of node2?

[Jojo Castro]

What part of DYNAMIC buffer cache do you not understand? I would be very surprised if two identically configured and identically loaded (from the standpoint of processes running) would display identical buffer cache utilization. The whole point of dynamic buffer cache is that it is dynamic. The differences you are seeing are normal and expected. You could set dbc_min_pct = dbc_max_pct and thus fix the buffer cache.

===>
Yes it is identically configured (including dbc_min_pct & dbc_max_pct). Also, identically loaded from the standpoint of processes. My questions are, thus dynamic buffer cache usage has something to do with the memory usage? Why is it that node1 memory utilization is 5% higher than that of node2?

by the way, please find below setting of dbc_min_pct & dbc_max_pct on both nodes.

# kctune | grep dbc_max_pct
dbc_max_pct 5 5 Immed
# kctune | grep dbc_min_pct
dbc_min_pct 1 1 Immed

[Jojo Castro]

hello all, any answers please?

thanks

[whiteknight]

Jojo,

It is also depands how many Oracle process you are running on each individual nodes right ?

It is possible Node 1 has more running process than Node 2 ?

WK

[Patrick Wallek]

>>thus dynamic buffer cache usage has something to do with the memory usage?

Well, yes. The buffer cache is a portion of memory. So, as dbc usage increases, so does memory usage overall.

[whiteknight · Accepted solution]

Jojo,

1 more advise.

If you run the test for a while you might be able to use AWR (Statpack) to see what the Oracle SGA component sizes are....perhaps some are changed dynamically in the DB's auto-tuning mechanisms.

WK

[Jojo Castro]

Jojo,

1 more advise.

If you run the test for a while you might be able to use AWR (Statpack) to see what the Oracle SGA component sizes are....perhaps some are changed dynamically in the DB's auto-tuning mechanisms.

WK

====

thanks so much WK. What is AWR? How can i get those and how can i execute it?

[whiteknight]

Jojo,

AWR = Automated Workload Repository

You need to get Oracle folk to run.

WK

[Jojo Castro]

Ok thanks to all!!!

[Jojo Castro]

Have understand the behavior on the issue. I already cleared the issue with my group.