| Btw, besides fixing the basic boot-order problem on Virtual Hubs
(now you can boot either node first, doesn't matter), and some re-init
problems, the TIMA kit also does some other initialization checking
to avoid potential corruption problems:
1) Check Version match among SYS$MCDRIVER.EXE images. To detect if the
data structures have changed. Unlike other drivers, every MCdriver
with shared data structures has to agree what's where. This is why
you cannot have mixed MCdriver versions in the same community.
2) Resource availability. Checks to see that enough is left for the
rest of VMS after MCdriver is loaded. It's possible to boost the
MC sysgen params to consume all of the physical memory. In that
case, the VMS boot just hangs at some point waiting for resources
to free up (which they never do). Now MCdriver will go offline
if there aren't 32MB left after MCdriver loads. On a small system,
this could cause you to load MCA0 but not MCB0, unless you lower
MC sysgen parameters.
3) SYSGEN parameter match. All nodes in the community must agree on
the sizes of the shared data structures specified by MC_SERVICES_Pn.
The only differences allowed are P0 - Crash Other Nodes Too
and P7 - Display Informational Messages.
BTW, besides the INIT-time messages, there are some other messages
that might come out during heavy loads if you have P7 (which is
dynamic) = 1 or 2. You may get a timeslice message, which means
that MCdriver got so many interrupts at once that it could not
service all nodes in the community during its cpu allotment.
This used to result in the highest node being message starved, but
now the interrupts round-robin better. Still, if you see this,
it's an indication that some Sender node(s) is much faster than the
Receiver node.
With MC_SERVICES_P7=2, you might see PMstall messages. This means
that your MSCP_CREDITS are higher than the number of buffers per
channel (that MC_SERVICES_P9 specifies) can handle. It just means
that we failed to allocate a buffer and had to wait for one to be
released and retry, but if you get these a lot, you might either
increase P9 if you have the memory, or decrease MSCP_CREDITS.
On the other hand, if you never get these, you can maybe reduce
your MC_SERVICES_P9 message count per channel, and save some memory
space. P6 and P9 affect the bulk of the MC memory requirements.
|