Streaming - A Practical Guide to Video and Audio Compression - page 359

This is a problem that Pixar encountered in its workflow. Apple introduced a new
QuickTime codec they had developed to deal with the issue. The new codec is called
Pixlet, which is a contraction of “Pixar wavelet.” The Pixlet compression is somewhat like
a motion JPEG file because it only codes I-frames. Pixlet is designed to compress efficiently
but without dropped frames and with no visible artifacts at all. This yields sufficient com-
pression to get the bit rate down to about 1 Mbps. At that scale, the video can be played
across the network with no dropped frames. This is sufficient to be able to post-sync the
sound, for example.
The Pixlet codec now ships as a standard component in QuickTime (version 6.4),
which is an integral part of the Mac OS X 10.3 (Panther) release.

20.11 Very-High-Performance Solutions

Some important work is being done by companies such as Kasenna and Akamai in the
area of distributing the streaming servers out to the edge of the Internet. Kasenna in par-
ticular has some very interesting ideas, and, as a member of the Internet Streaming Media
Alliance, (ISMA) is clearly committed to making this work with open standards. You can
choose proprietary standards and stream them through the Kasenna architecture as well.

20.12 Edge Servers

A major problem with serving content across the Internet is the long distance between the
client and the server. The distances are large in geographical terms but also in logical
terms. The signals have to travel a long way though the wires, and the logical connectiv-
ity describes the number of different sub-networks and routers that are involved in the
connection. Recall that these are called “hops.”
Kasenna and Akamai have worked hard to address this problem. Although they both
deploy edge servers, the way each company connects them to the central repository is
slightly different.
An edge server is a CPU that is placed near the end user's connection, close to the
edge of the Internet. This reduces the distances involved, both physically and logically. As
a consequence, the service enjoys reduced latency. Locally cached content is delivered
more rapidly and responsively. This works especially well for streamed content and for
very high-traffic loadings when serving static content. It divides the audience into smaller
communities and avoids everyone in the world hitting your server at once. This model is
illustrated in Figure 20-6.
Because there are many edge servers involved, the loading is spread across more
machines and distributed over a wider part of the Internet. Multiple connections become
effectively aggregated into a single request from the edge server back to the central repos-
itory, and in many ways it looks like a proxy server connected in the reverse direction.
Kasenna and Akamai have solved replication of streams and delivery of content, and
some very innovative work is being done to enable broadband services to make full use of