Conference turris::digital_unix

9956.0. "Monitoring tools fail with large configurations" by SPANIX::JULIANR (ALPHAbet = Our bet on ALPHA) Tue May 27 1997 07:36

	Hi all.

	We are running a heavy benchmark with as many as 278 rz devices.

	However, both Performance Manager and "iostat" don't show anything 
beyond rz270. Other monitoring tools don't help too much:

	- "monitor" v1.5 tool doesn't even start, because of 
problems in the "disk.c" module. 
	- "sydsmp" also crashes when trying to show CPUs performance in bar 
chart mode. 
	- "pvis" T1.1 doesn't show all block devices we have, and doesn't 
show the CPUs status in their real status.
	- "uerf" and DECevent 2.3 don't show most of the starting messages 
to check the system configuration.

 
	We also have 12 CPUs @440 Mhz, 8 GB of RAM and 34 KZPSA on a 8400 
with Digital UNIX 4.0b and most of the February patches. 

May this large configuration make such monitoring tools to fail ?

Which utilities should we use to have a clear and complete picture of large 
configurations like the ours ?

Regards,

Julian Rodriguez, UNIX Ambassador
Technical Sales Support 
Digital Spainz

	Monitor could be failing because the V4 version tries to
	collect the I/O queue depth using a method that simply
	doesn't work on all configurations.  Try removing the
	"-D=NEXUS_LIST" option from the compile line and recompiling.

	Also, I think the table(2) system call used to get the
	data is limited to supporting 256 disks.  If the various
	applications that use the interface encounter more each
	could fail in a unique way.  I know that Monitor has never
	been tested in such a large configuration.  Even if you could
	collect all the data, I doubt you could display it all.

    for the uerf case, you have to increse the message buffer size of the
    generic subsystem to a value that fit's the machines need.
    
    Pit