This page attempts to answer some of the most Frequently Asked
Questions that the Webperf team has received.
-
Why are particular ISPs not
listed?
-
Webperf only tests companies that have asked to be tested.
Submissions can only be accepted from authorised
representatives of the company concerned. If an ISP is
absent from the testing, you should contact the ISP itself.
Please also note that sites under test (SUTs) are
automatically removed from the testing if their web site has
generated test errors for a prolonged period of time.
-
Why has my company disappeared from
the testing?
-
If a SUT generates test errors for a continuous period longer
than twelve hours, from multiple test hosts, it will be
temporarily removed from the testing and the company
concerned is notified by email of the removal. Common causes
for this are:
- the static object being tested is moved
- there is a prolonged network outage between test hosts
and the SUT
- the SUT's web site is unavailable.
-
What do the availability results
mean?
-
The availability results are an average assessment of how
successfully a typical web client was able to request and
download the front page of the SUT. There are many reasons
why this test could fail, but it does not automatically mean
that the server is down.
-
connect_error
This indicates that we could not connect to the web site
under test. This is equivalent to the Netscape error
message "connection was refused by the server",
or a connect() system call failing. This error
case is also given if no IP address was returned by the DNS
lookup.
This error can be returned if the web server is overloaded
and unable to respond to the request made.
-
http_error
This indicates that we managed to connect to the server but:
-
we received an invalid (i.e. non-RFC-compliant) HTTP response
-
we received a valid HTTP response, but that response was
an error (an HTTP status code other than one between 200
and 399)
-
the server broke the TCP connection before we had
received all the data
-
timed_out
This indicates that we were unable to resolve the DNS name,
connect to the web site and download the front page
(excluding images) within 60 seconds - at which point most
people accessing the web site will have pressed the stop
button and wandered off.
timed_out errors can also indicate network outages when the
errors are only seen from a couple of test hosts
(especially if these test hosts are on the same network).
If the timed_out errors are more widespread, it probably
indicates a problem closer to the site under test.
This test is based on the definitive white paper on the
subject of web site download speeds and their potentially
adverse effect on your business. This phenomenon, also
known as the eight second rule can be found
on Zona Research's web site.
-
failed_requests
This indicates that something strange happened to the
static file being tested during the ApacheBench run. The
number in bracket correlates to the number of failed
requested encountered.
This test highlights extreme web server misconfiguration,
as it is unlikely that errors of this nature could happen
in any other way.
-
Why are my results not as good as
the people at the top of the table?
-
Network performance along a connection is largely based on two
factors:
- the latency of the channel, and
- the maximum data throughput of the channel.
Poor performance can be the result of many factors:
- limited bandwidth
- poor peering
- overloaded bandwidth, or
- overloaded hardware/software serving the data.
Your overall performance is going to be limited by a
bottleneck caused by any one of these factors. The SUTs at
the top of the Webperf performance tables have successfully
found an efficient combination of the above factors.
-
Why do some test hosts give me
very slow performance results when other companies get much
faster results from the same test host?
-
Each ISP has different peering. If the bandwidth provider
for a particular SUT has peering arrangements with the
network on which there is a test host, it will get better
performance results from that test host, than a SUT whose ISP
is not peered.
Good peering is intrinsic to your web hosting business - if
you are poorly peered, then customer's sites will have
correspondingly poor connection times, and as the
eight-second rule shows, this can have a detrimental on their
(and your) business.
All the test hosts are now standarised on 100Mb connections,
so peering arrangements with other providers can be assessed
from per-test host performance data. To maximise your
business potential, ensure that your network is well-peered.
An indication of good peering is when a SUT receives
consistently high performance throughputs from multiple test
hosts.
-
How do I advertise on
Webperf?
-
Webperf does not currently sell advertising space. Banner
adverts are available for companies who provide a shell
account on a machine on their network connected to 100Mb link
for Webperf to test from.
We are currently looking for test hosts in the US. If you
would like further information please contact webperf@webperf.net.
-
How do I change the testing details
I've submitted?
-
Currently, the quickest way to have your testing details
changed is to fill out the
submission form. A system is currently in development to
allow subscribers to manage their own details. More news on
this will follow.
-
Why has my submission been
rejected?
-
There are a number of basic criteria that must be fulfilled
before a site can be added to the testing:
-
your company must be an ISP or Reseller, i.e. you must
provide mass web hosting services for customers, corporate
or otherwise - Webperf is unable to test personal web sites
-
your company must not already be being tested by Webperf
-
the URL of the static object specified points to a
different server to that of the main company web site, and
it is not clear whether that server hosts other customers -
this is to ensure that dedicated machines are not set up
purely for testing purposes
-
Why is the text on the web site white
on a white background with IE on Mac?
-
We believe this is due to brokenness in the rendering of
Cascading Style Sheets by Internet Explorer on MacOS.
Currently the only workaround is to turn off CSS in your
browser.
-
Why is my site not tested by all the test
hosts?
-
Before a performance test is carried out, a network proximity test is
first run to ensure that there is at least a WAN interface between
a test host and a SUT. SUTs that are too close to a test host in
network terms will not be tested.
-
Why are the results on Webperf
different to the results in Internet Magazine?
-
Webperf publishes its results in real-time, but due to time
constraints, Internet Magazine's results are usually gathered
from Webperf between 1-2 months prior to the magazine's
publication. For the most up-to-date results, please refer
to our web site.
Please note that explicit permission must be sought from
Webperf before any part or whole of the data we collect is
published.
