Index Journeys
RFC 334 UNKNOWN OPERATIONS · TESTING

Network Use on May 8

A warning that network performance tests might cause temporary outage

On rfc-editor.org ↗
PUBLISHED
May 1972
AUTHORS
A.M. McKenzie
STREAM
Legacy
TEST DATE
May 8, 1972
RESEARCHER
Vint Cerf, UCLA
NEW IMP SOFTWARE DATE
May 15, 1972
TEST FOCUS
Heavy traffic to single site
DOI
10.17487/RFC0334

RFC text

Original text · RFC Editor View on rfc-editor.org ↗ 
Network Working Group                     Alex McKenzie
RFC #334                                  BBN
NIC 9927                                  1 May 1972


                          Network Use on MAY 8



     Vint Cerf of UCLA Network Measurement Center has just indicated
to us a requirement to carry out certain performance experiments on
the network prior to installation of the new IMP software on May 15.
These experiments are designed to measure network performance under
heavy traffic load to a single site, and thus may cause the network to
"lock up" such that _no_ traffic can flow (a problem which the new IMP
software is de- signed to prevent). Vint has indicated to us that ARPA
regards these experiments as extremely important in spite of the
possibility of lockup; we have therefore agreed to assist with the
experiments and scheduled them for May 8.

     Although we do not know whether these experiments will cause
lockup, we suggest that sites do not schedule important network work
for Monday, May 8. We apologize for the lateness of this announcement.








       [ This RFC was put into machine readable form for entry ]
       [ into the online RFC archives by BBN Corp. under the   ]
       [ direction of Alex McKenzie.                   12/96   ]

AI ELI5

On this page
PROBLEMResearchers needed to test how the ARPANET would perform under extremely heavy traffic to a single site before new software was deployed that would prevent lockup.
MECHANISMVint Cerf's group at UCLA's Network Measurement Center scheduled performance experiments for May 8, 1972, that could potentially cause the entire network to lock up with no traffic flowing.
NETWORK MAINTENANCE WINDOWAll ARPANET sites were warned not to schedule important network operations on May 8 due to the risk of complete network outage during the experiments.

§1The May 8 Performance Tests

On May 1, 1972, Alex McKenzie of BBN announced that Vint Cerf's Network Measurement Center group at UCLA would conduct critical performance experiments on the ARPANET on Monday, May 8, 1972. These experiments were designed to measure how the network would perform under heavy traffic concentrated at a single destination site.

Vint Cerf's team at UCLA planned to test what happens when lots and lots of messages go to one computer at the same time.

§2The Risk

The experiments could potentially cause the network to lock up such that no traffic could flow in any direction. This problem—where the network's finite buffers fill up and prevent any progress—was exactly what the new IMPSYS 2600 software (scheduled for release on May 15) was designed to prevent. Despite the risk of complete outage, ARPA regarded these experiments as extremely important to understanding network behavior before the software deployment.

The test might break the network completely, but the government thought the test was important to do before putting in the new software.

§3Recommendation to Sites

BBN recommended that all ARPANET sites refrain from scheduling important network work on May 8. The announcement acknowledged the lateness of the notice and apologized for the short notice, as the experiments had just been scheduled. Sites were expected to understand that despite the disruption risk, the research was essential for understanding and improving ARPANET reliability.

Sites were told: don't do anything important on the network that day. BBN said sorry for telling them at the last minute.