What determines the client connection memory size?

Question

We are trying to scale up the number of connections on our db - (dedicated not shared) .. but quickly consume the box.
Its 11gr1 - Linux .. 500G of memory ..
The "only" parameter we have set is:
*.memory_target=216522555392  (~200G)
Processes set to 6000 - but are are only around 1800 at this point.
We are seeing (via top) client connections with a reserved memory of 25-30g (usually the dbwr type processes) and client connections showing 5-10G in size.
With clients taking this much memory,  we start to see swapping on the box. In our dev/qa environment the clients are in the MB range .. of course they are not seeing real world traffic so I presume that memory requirements are growing as the app runs.
Can we set something to reduce the footprint of the client connections?
Thanks for any tips..
If we get one db bounce this year .. we want to be right .. cant guess here..
Daryl

Answer 1

DarylE. wrote:
We are trying to scale up the number of connections on our db - (dedicated not shared) .. but quickly consume the box.
Its 11gr1 - Linux .. 500G of memory ..
The "only" parameter we have set is:
*.memory_target=216522555392  (~200G)
Processes set to 6000 - but are are only around 1800 at this point.
We are seeing (via top) client connections with a reserved memory of 25-30g (usually the dbwr type processes) and client connections showing 5-10G in size.
With clients taking this much memory,  we start to see swapping on the box. In our dev/qa environment the clients are in the MB range .. of course they are not seeing real world traffic so I presume that memory requirements are growing as the app runs.
Can we set something to reduce the footprint of the client connections?
Thanks for any tips..
If we get one db bounce this year .. we want to be right .. cant guess here..
Daryl
>Can we set something to reduce the footprint of the client connections?
no
Since *NIX maps SGA into every client's process, displayed RAM size is distorted.
If you simply SUM every reported client size, it will greatly exceed total RAM (in most cases)
The fact that any swap is used is in itself not a negative indicator.
run vmstat like below
all is OK when (si + so) is less than (bi + bo)
[oracle@localhost dbs]$ vmstat 10 6
procs -----------memory---------- ---swap-- -----io---- --system-- -----cpu------
r  b   swpd   free   buff     cache     si   so    bi    bo   in   cs us sy id wa st
0  0     32  82684  20384 557096    0    0   120    32 1008 1057  3  7 88  1  0
0  0     32  82560  20408 557124    0    0     0    20 1031 1225  2  4 93  0  0
0  0     32  82560  20432 557116    0    0     0    16 1002 1183  2  6 92  0  0
0  0     32  79212  20456 557144    0    0     0    74 1007 1185  4 12 84  0  0
0  0     32  78592  20480 557148    0    0     0    21  999  998  2  5 92  0  0
0  0     32  78592  20504 557140    0    0     0    20 1002  929  2  7 91  0  0
[oracle@localhost dbs]$