[comp.os.vms] DECnet and Ethernet performance

nagy%warner.hepnet@LBL.ARPA (06/14/87)

>mike hopper
>Hopper.XRCC-NS@XEROX.COM
     
>We have both a 750 and uVAX on our local Ethernet.  We are running
>DECNET on both and find that the speed of communications between the
>VAXes is very slow.  I read in the "DEC Professional" magazine  of April
>1987 that the file transfer rate between two DECNETted 780's was 40 KB
>but that between a QBUS based pdp-11/44 and a 780 was only 10KB.
>     
>Does anyone know if the uVAX to 750 transfer rate is also very low?  If
>so, how much is this degraded by having a number of users doing
>transfers simultaneously?
     
One thing that has a lot to do with DECnet transfers on the Ethernet
is what controller you have.  The DEUNA is just about the slowest
(except for the "new" BI controller) Ethernet controller Digital
makes.  The rated throughput for the DEUNA is about 1.6 Mbits/sec -
which will translate to about 200 KBytes/sec at best.  The MicroVAX
DEQNA controller is considerably faster; I remember bandwidths as
being quoted in the 2.4-2.6 MBits/sec range (about 300 KBytes/sec
flat-out).

In preparation for designing a new distributed control system here,
we undertook some measurements of DECnet performance.  The test
systems were:

	1. Two MicroVAX-II's with DEQNAs on the same Ethernet segment.

	2. Running MicroVMS V4.2

	3. Systems were idle except for our test programs (running at
	   normal interactive base priority of 4).

	4. Ethernet was essentially idle also.

The test setup involved a pair of processes, one on each system.  Process
F setup a logical link to process E.  Process F then proceeded to sit
in a loop and send messages to process E which would echo them back to
process F.  The loop was timed.  The results, after analysis, were:

	- Overhead to send a zero-length message task-to-task using
	  DECnet is 4 milliseconds on a uVAX-II

	- Bandwidth is approximately 270 Kbytes/sec.

= Frank J. Nagy
= Fermilab Research Division EED/Controls
= HEPNET: WARNER::NAGY (43198::NAGY) or FNAL::NAGY (43009::NAGY)
= BitNet: NAGY@FNAL
= USnail: Fermilab POB 500 MS/220 Batavia, Ih!adh