User manual

Miscellaneous Notes
65
see additional processors added to the system and will under-collect performance data, at best,
or may even crash.
1
On a 32-processor system MONITOR utility collects and displays load information only for first
31 processors. While 32nd processor is there and is performing system workload, MONITOR will
not see it and will not collect load information for it.
ANALYZE /SYSTEM
Similarly, if SDA utility is running on a live system while VSMP LOAD command is being executed,
restart SDA after completing VSMP LOAD to let it see additional CPUs added to the system.
Watchpoint utility and WPDRIVER
Use of WPDRIVER is unsupported in the current release of VAX MP when multiple processors
are active. It is safe to use WPDRIVER on VAX MP when only one processor is active, but not
when multiple processors are active.
WPDRIVER references WHAMI macro. This macro is not aware of VAX MP multiprocessing
capabilities and will incorrectly treat it as uniprocessor MicroVAX 3900, resulting in improper
operation and possible system crashes. It may be possible in the future to provide a patch for
WPDRIVER enabling its use in multiprocessor mode on VAX MP, should there be a need for such
use that justifies required effort.
Running cluster
When running VAX software simulator, whether uniprocessor or multiprocessor, as a member of
a cluster, it is prudent to adjust the values of SYSGEN parameters TIMVCFAIL and
RECNXINTERVAL across the cluster. TIMVCFAIL governs timeout interval for cluster virtual
circuits. If a cluster member does not receive acknowledgement from another member within
TIMVCFAIL interval, it considers the circuit broken and initiates reconnection attempt that lasts
up to maximum interval indicated by RECNXINTERVAL. Once RECNXINTERVAL expires, remaining
nodes abandon attempt to reestablish connection with the lost node and proceed to rebuild the
cluster without the lost node or nodes.
Unlike with silicon VAX, simulator-based VAX (or Alpha) node response can be delayed by
simulator’s thread preemption due to host system load, spin-up of disks previously spun down
according to host power saving policy, intervention of antivirus software and so on. To prevent
false circuit breakdown induced by these factors, it is prudent to increase the value of
TIMVCFAIL from default of 1600 (corresponding to 16 seconds) to 3000 (corresponding to 30
seconds). There is no point in further increasing TIMVCFAIL beyond the value 3000 as NI port
1
This issue can potentially be fixed later by converting resident part of VSMP to SYSMAN SYS_LOADABLE image,
however that would present its own inconveniences.