This page contains information to update and supplement the paper Forecasting Disk Resources for a Usenet Server, which was presented at the 7th Usenix Large Installation System Administration (LISA VII) Conference. The graphs are updated every week or so; commentary will be updated as need and motivation dictate.
SLAC stopped carrying certain newsgroups, primary a subset of alt.binaries.*, on 4 November 1996. No correction has been made for this change, so the graphs presented here show a substantial drop in traffic after that date. Then, in late January 1997, SLAC shut down its news server entirely. Other data sources for continuation of this project were not readily available. |
The paper's figure 1 showed the number of articles accepted by SLAC's news server per week, with interpolation of data during the collapse of the old server and installation of the new one in September 1992. The graph of current articles per week includes interpolated data for several subsequent disruptions plus the raw data. |
Figure 2 showed exponential and quadratic curves fitted to the actual number of articles per week, with five holiday low points excluded from the curve fit. The current trend graph shows only the 67% exponential growth curve used later in the paper, the raw data with interpolation for periods of server disruption, and a new 100% growth curve intersecting the old curve at mid-1993 which seemed to be a better fit until early 1995. Since then, the growth of news has slowed dramatically, probably due in large part to the growth of the World Wide Web. As of April, 1996, the trend appears to have dropped back to match the historical 67% growth curve. Whether or not it will drop even further will be interesting to watch during the remainder of 1996 and beyond. |
Another graph attempts to compare each week with the
same week in the previous year.
Defining the same week is not as
easy as it might seem. This graph simply compares weeks which are
52 weeks apart. The result is rather noisy, which is why this graph
was not included in the paper, since holidays, with their attendant
large dips in Usenet traffic, are not always exactly 52 weeks apart.
Several types of averages (which should be described here; yet another
thing to do!) help only somewhat.
The surge in growth during the latter half of 1994 and continuing into 1995 is most likely due to large, formerly closed providers such as AOL and CompuServe adding access to Usenet, thereby dramatically increasing the Usenet readership over a fairly short period of time. The drop to nearly zero growth in early 1996 is more an artifact of this one-time boom than a real drop. Comparing this time period to two years before might be more meaningful. |
Copyright © 1995-1998,2001,2002, Karl L. Swartz. All rights reserved.
All trademarks mentioned herein belong to their respective owners.
Karl Swartz <karl@kls2.com>