Problem with swap space

Question

Hi All,
Using Oracle 10gR1, Solaris 9.
prtconf | grep "Memory size"
Memory size: 16384 Megabytesswap space is 16g
Pga_aggregate_target = 2G and after checking the v$pga_target_advice, it shows optimal.
Currently we have a issue of swap space getting used up nearly 100%. The system was out of memory and can't run any application
Alert logs error;
ORA-07445: exception encountered: core dump [_aio_free_stack_unlocked()+72] [SIGBUS] [Object specific hardware error] [0xFFFFFFFF7B602000] [] []
ORA-00600: internal error code, arguments: [ksnpost:ksnigb], [], [], [], [], [], [], []
ORA-04030: out of process memory when trying to allocate 588408 bytes (pga heap,kco buffer)On checking Metalink note, it said its an unhandled Oracle exception; have to increase the swap space.
AWR excerpt;
Cache Sizes (end)
~~~~~~~~~~~~~~~~~
               Buffer Cache:     4,848M      Std Block Size:         8K
           Shared Pool Size:     3,985M          Log Buffer:     1,024K
Load Profile
~~~~~~~~~~~~                            Per Second       Per Transaction
                  Redo size:             63,328.92              2,116.68
              Logical reads:              4,946.83                165.34
              Block changes:                257.73                  8.61
             Physical reads:              1,267.50                 42.36
            Physical writes:                 72.44                  2.42
                 User calls:                385.83                 12.90
                     Parses:                 84.22                  2.81
                Hard parses:                  0.04                  0.00
                      Sorts:                 10.53                  0.35
                     Logons:                  0.14                  0.00
                   Executes:                211.35                  7.06
               Transactions:                 29.92
  % Blocks changed per Read:    5.21    Recursive Call %:    13.56
Rollback per transaction %:    0.29       Rows per Sort:    94.72
Top 5 Timed Events
~~~~~~~~~~~~~~~~~~                                        % Total
Event                                 Waits    Time (s)   DB Time     Wait Class
row cache lock                       16,498      45,001     43.44    Concurrency
log file switch (checkpoint in       30,766      30,281     29.23  Configuration
db file sequential read           2,873,716      10,796     10.42       User I/O
buffer busy waits                     8,870       8,718      8.42    Concurrency
db file parallel write                  421       3,848      3.71     System I/O
          -------------------------------------------------------------On increasing the swap space, and restarting the instance, the system was normal.
Now why does oracle use up all the swap space and what can be the workaround to avoid this in future?

Answer 1

Verma wrote:
Using Oracle 10gR1, Solaris 9.
swap space is 16gOracle 10gR2 swap space recommendations are:
- 1 GB - 2 GB = 1.5 times the size of RAM
- 2 GB - 8 GB = equal to the size of RAM
- Greater than 8GB = 0.75 times the size of RAM
Currently we have a issue of swap space getting used up nearly 100%. The system was out of memory and can't run any applicationThat is unusual and should not be happening when you have what seems to be an excess of memory available. This could mean some kind of bug (memory leakage for example), or it could mean plain old fashion abuse of memory. Like PL/SQL code that attempts to bulk fetch a few million rows from the database with a single go.
On checking Metalink note, it said its an unhandled Oracle exception; have to increase the swap space.Wrong IMO (unless you were not following the recommended swap size factor). If you have have a giant swap area..moving the wall a few meters away (by increasing swap) will only enable you to run faster into it - I doubt that this is the intent of that Metalink note.
On increasing the swap space, and restarting the instance, the system was normal.
Now why does oracle use up all the swap space and what can be the workaround to avoid this in future?Oracle only use what is needed to perform the client requests that it receives. If the client request a mountain of data to be moved/processed, Oracle will try to oblige. If that mountain topples and crushes server resources, it is not Oracle that is to blame for trying to do exactly what was requested from it.