Stress over system tables

Hi, I have one server with informix 12.10.FC10, it have more than 100 databases but we only use about [20-30] concurrent connections. Anyway I verify that when I try to read system tables like syslocks or others it takes long time. Is this a problem or my configuration error? I have searched about onstat results and how can I look for problems over results, but don't find much information, cab someone help me about this? Thanks for any help, SP

Sergio: There are two main classes of SMI tables in the sysmaster database: 1) those that expose in-memory data structues examples include syslocks, sysshmvals, sysprofile 2) those that expose on-disk structures examples include syschunks, sysdbspaces, sysdatabases, systabnames, sysptnhdr, sysextents 3) those that expose the memory images of disk structures examples include sysactptnhdr (a version of sysptnhdr that only includes tables that have actually been opened and have data in memory) The #2 SMI tables can become slow if the TABLESPACE TABLESPACE pages (object header pages exposed in sysptnhdr & others) in the dbspaces become badly fragmented or if you have many databases with many tables spread across many dbspaces even if the TABLESPACE TABLESPACES are not fragmented. The in-memory structures in #1 & #2 can become slow if the server's quiescent memory pages are being swapped out by the OS. One solution to this would be to be RESIDENT to -1 so that all shared memory is marked as non-swappable. This can also be because of the OS. Solaris, for example, while otherwise a good host for Informix, has problems with shared memory. For example it can take much longer for Solaris to allocate a new shared memory structure if the server requests one than Linux or AIX and sometimes onstats and queries against sysmaster on Solaris are just slow, especially in the older Solaris releases.

Hi Art, Thanks for your reply, I use CentOS Linux (RedHat clone) and I have done some experiences like configure PSORT_DBTEMP to use TMPFS. I also tryed to configure that kind of filesystem for temporary dbspaces, but I quit because the system restart takes about 40 minutes to backup and restore them... I also have my swappiness defined for 10 to only when I have 90% of memory occupied it uses swap. Even though I still have latency over my systems and don't know how to identify them and in case of lack of the code to show my colleagues the mistakes. SP