Summary of Results
In terms of overall technical feasibility of online journals access,
we found two universities where it clearly worked very well (MUK and UG)
and two where there were severe problems (UNZA and UCAD). At UCAD, we
were unable to download large PDF files during regular hours, although
we did succeed (albeit very slowly) during off-hours. At UNZA, we were
unable to download any PDF files over 130 kilobytes (Kb), regardless of
time of day. In both places we had much better success with HTML articles.
At both UG and MUK, the process for both PDF and HTML articles was fairly
quick and trouble-free.
Judging "by the numbers" in raw bandwidth terms, these results
may be considered somewhat surprising. The University of Zambia has the
highest bandwidth connection of the four universities, at 128 kilobytes
per second (kbps); its performance, however, was the worst of the four.
Cheikh Anta Diop has a 64 kbps radio link. Both universities have fiber-optic
LANs in place. Their relatively poor performance may arise from several
factors, including the number of simultaneous local users relative to
the available bandwidth, the LAN architecture, and the quality of the
"upstream" Internet links. On the other hand, MUK, which has
only a 33.6 kbps modem dial-up connection to a local commercial service
provider, shared by ten computers, enjoyed quite speedy access. The results
at UG, where they have a fiber LAN connected to a dedicated 64 kbps radio
link as well as dial-up access, were very similar to those at MUK.
Some observations:
- Given the situation at MUK, where a 33.6 baud modem serves as the
university's main Internet connection and feeds ten computers capable
of efficiently downloading journals at the same time, there would seem
to be no technical reason that online journals are inherently unfeasible
in Africa. Network setup and management are key.
- There are several possible reasons the connection worked so poorly
at UNZA. One is that the LAN is a single physical ethernet segment connecting
some one thousand computers, with all those computers competing for
LAN resources, including bandwidth for Internet connectivity. In short,
that's a lot of traffic. By contrast, the LAN at UG is organized into
smaller ethernet segments connected to each other by an ethernet switch.
This setup manages the traffic more efficiently, as computers compete
only with other computers in the same segment, and inter-segment traffic
is minimized. Thus utilization of the ethernet is made more efficient,
making it possible to add more terminals to the network with less degradation
of service. Another possible reason is that the problems were a result
of the way that Windows handles Internet data transfers. Downloading
is a two-way process involving data coming in and receipt confirmations
going out. When there is a delay in the incoming data, Windows slows
down its response rate, but if the incoming rate increases, Windows
does not speed back up to match it. This creates a tendency toward ever-decreasing
speed that eventually kills the connection once it reaches a delay of
a certain length.
- Strategically placed mirror sites for journals may prove to be a key
aid to feasibility in certain locations. At UCAD, although efforts to
download large PDF files failed, it was nevertheless possible to download
very large files from a server in South Africa because the UCAD Internet
connection happens to share a common service provider with the particular
South African server in question. Interestingly, this common provider
is in Canada, not Africa, so it is more a question of optimal routing
than geographical location of servers. Network congestion is more likely
to occur when the information being accessed must travel across the
"peering points" that connect different service providers.
- At UCAD, which has a 64 kbps radio link like UG, but which performed
much worse than UG, the difference seemed to be attributable to the
higher local usage at UCAD. UCAD resells its bandwidth through additional
64 kbps leased lines to four other non-university institutions. These
four other lines are fed through the UCAD radio link. While this arrangement
helps UCAD pay for its Internet connection, it seems to have significant
negative effects on network performance.
- While there is some evidence that network performance is enhanced
during African morning hours (i.e., before large usage begins in the
US morning), there is no doubt that the biggest bottleneck is the amount
of local usage. In other words, the Internet connection is faster during
times of low local but high US use than it is during times of high local
but low US use.
- HTML-based journals are far easier to access than PDF-based ones.
Even on the slowest, most difficult connections, HTML articles remained
accessible while even mid-sized PDF files were impossible to download.
- University level caching proxy servers provide significant performance
improvements on Internet access generally, but are of limited usefulness
for online journals specifically. The reason is that online journal
articles are accessed by means of CGI requests, a method that allows
journal providers to keep accurate databases of access statistics. Proxy
servers do not normally cache CGI requests, because by their nature
CGI requests need to be "new" each time in order to be useful.
Still, the proxy server improved the online journals experience by providing
nearly instantaneous access to all the intermediate steps one must go
through before actually downloading an article, e.g., accessing the
journal’s front page, table of contents, and recurring graphics (logos,
etc.). In the case of the IDEAL catalog, there are some nine preliminary
steps before one gets to download an article, from the IDEAL homepage
and log-in page to selecting subject categories, sub-categories, journals,
years, issues, and finally articles. Without the proxy server, each
step typically took two minutes or more, adding up to nearly 20 minutes
(and sometimes more) of terminal time before beginning an article download.
With the proxy server, these steps were virtually instantaneous.
- Unix skills (a key to network management) existed at each university,
but were often limited to one or two people and sometimes at relatively
low levels of knowledge. These people were uniformly interested in improving
their skills and extending them to other people. Unix training would
be a good use of donor funds.
- Another good use of donor funds would be to offset the initial capital
costs of improving bandwidth. In Uganda a radio link would cost $250
per month on a flat rate basis, whereas they now pay from $500 (usually)
to $2000 (sometimes) per month in phone bills for their dial-up access.
Why don't they switch? The initial, one-time setup cost for the radio
link equipment is $18,000, and for the moment that is enough to be an
insurmountable barrier. At UCAD, they are interested in setting up a
VSAT system for direct satellite access, which again would greatly expand
the bandwidth while reducing the monthly cost. Again, the barrier is
the initial setup cost, a one-time expense that has prevented the move
toward better communications and greater sustainability.
- One of the elements of feasibility is the cost of the online journals
subscriptions. While these may be made available at discounted rates
to African universities, some payment will be necessary if online journals
access is to be made sustainable. This reality raises difficult questions
about the number of users that can be supported by the networks, i.e.,
what number of journal users is necessary to justify the cost of a subscription,
and can that number of total journal users be supported by the existing
network infrastructure? If not, what is the cost of upgrading the network,
and how does that affect the journals subscription calculus?
 |
 |
|
