Tuesday May 10th, 2022 NCLUG Meeting 6pm

Bob Proulx bob at proulx.com
Tue May 10 19:49:26 MDT 2022

j dewitt wrote:
> When: Tuesday May 10th, 2022, 6pm

As typical we started with general discussion.

VMS seems to have produced an actual supported release on 64-bit
amd64.  Wow.  Will wonders never cease.


Phil started out probing to see if anyone would like to swap a server
between houses on the FC city fiber and then they could use the
storage space space for criss-cross offsite backup across the fiber

Alex talked about the home guinea pigs.  One of them has a deformed
leg.  Sad.  Stephen talked about the joys of parenting a teenager.
Aaron related a similar parenting issue with his 12 year old.  Somehow
the story of the guinea pigs with the physical deformity sounds less
sad now.

Discussions are great because we also talked about electromigration
issues of VLSI metal connect.  Back to back with frozen exterior water

Discussion of tradeoffs of high core cpus and number of I/O channels.
The joys of large ram machines with more than 2 TB of RAM.

Pricing of Amazon AWS services.  Absolutely EVERYTHING costs a very
small micro-charge and that is multipled by a godzillian operations.
Trying to budget for this is very difficult.  What is the result of a
very small number multiplied by a very large number?  Who knows!

Apache problems on a small server.  Apache wasn't running.  MariaDB
was not running.  They were killed.  Various things were killed.  And
no one knew until it was noticed.  So that could use some monitoring.
But the problem was the Linux OOM Killer.  Due to external "Internet
background radiation" probes of the web server could cause Apache to
fork so many processes that the system would run completely out of
memory and then processes would get killed by the OOM.

When running Apache on the Internet it is necessary to benchmark how
much RAM it uses and to tune the limits to keep the max number of
processes below the available memory limits.  One can benchmark web
server using "ab" the apache-benchmark program.  This was fired up
immediately and and we watched it crash the web server by having it
spawn too many processes.  Definitely a problem.

Logcheck.  It's good for monitoring.  But it starts out as a firehose
of noise from things that you don't care about.  So *if* you want to
use logcheck then you have to immediately and agressively add filter
rules to allow system log entries that are not important.

Tonight were more disorganized than even out typical level.  But a
good time was had by all.

More information about the NCLUG mailing list