Hi Manuales:
The Unix buffer cache is a portion of physical memory that holds file buffers.
When an I/O is requested, a "logical" or a "phsical" transfer occurs. If the requested data resides in memory; that is, in the buffer cache, then the I/O was a "logical" one. If a disk access is required to retrieve the data, then a "physical" I/O occured.
Similarly, when a write is performed, data may not be written immediately to the physical medium (e.g. disk), but rather kept in the buffer cache. A read request may then be satisfied from this buffer.
Clearly, keeping some buffer cache generally improves performance since reading and writing memory is much, much faster than doing I/O to electromechanical devices.
The job of the 'syncer' daemon, which runs every 30-seconds, is to flush modified buffers from memory to disk. Too big a buffer cache can cause peaks in processor and I/O which otherwise can impede server performance.
With the latest HP-UX releases, the 'syncer' process has been improved.
The amount of memory you define for the buffer cache is controled by the kernel parameters 'cbc_max_pct', 'dbc_min_pct', 'bufpages' and 'nbuf'.
If you are running a database like Oracle with its own internal buffer cache (SGA) you generally want to keep a small Unix buffer cache. To do otherwise is to "double buffer" which gains nothing.
For more information on configuring the Unix buffer cache, see here:
http://docs.hp.com/en/TKP-90202/index.htmlRegards!
...JRF...
