[Info-vax] Accuweather new contract

[Stephen Hoffman]

On 2015-03-30 03:25:23 +0000, Kerry Main said:

> 
> As I recall, Eric's previous analysis showed OpenVMS network numbersto 
> be approx. half those of Linux which would be exactly what would be he 
> case if the autoconfig issue were present when he did the testing.

One million UDP packets, versus 20,000.

Related:
  <https://groups.google.com/forum/#!msg/comp.os.vms/wu7afPpr0G0/bZtiEIwZZMoJ>
  <https://groups.google.com/d/msg/comp.os.vms/9wxcJ0bGgSw/KwATgBwa1H8J>


> Either that or one must believe that Linux network numbers are 
> twicethose of OpenVMS using similar network configs - something I find 
> a bitof a stretch.

Looks closer to 50x, with some general profiling information through 
the kernel stack.

Given the test traffic was UDP, packet retransmissions are not expected 
— which would mean that a duplex error would have to be explicitly 
missed, and an unaccounted loss of some large percentage of packets 
would also have to arise   But this is probably worth firing up a few 
tests, just to get some more current numbers than what EJ encountered 
some years ago.

Haven't tried building either of these on OpenVMS, but — via 
<http://www.csee.usf.edu/~christen/tools/toolpage.html> — see 
<http://www.csee.usf.edu/~christen/tools/blast.c> for a simple tool 
that blasts UDP.  Alternatively, here's a client-server 
<https://www.abc.se/~m6695/udp.html> version.  The latter one should be 
pretty easy to wire up packet loss detection.

For folks new to this socket stuff, see <http://beej.us/guide/bgnet/>.

On another Linux-related but otherwise unrelated topic, Linux 4.0 is 
rolling out the infrastructures for hot-patching the kernel.



-- 
Pure Personal Opinion | HoffmanLabs LLC