Web Page Delivery

Reference:

Whit Andrews, "Planning for PUSH," Internet World, May 1997, 48-52.

"Pull" vs. "Push"

Original Web model:

Pull: user idenfities URL (e.g., via search engine) and requests URL

Push:

• It's become a buzz word that means different things to different people
• Content-provider defines "channel" (like a TV channel)

Logic behind channels: People are used to the broadcast model, so reduce the Internet to a broadcast medium!

• Users subscribe to channels based on interest
• Easier to use for mass market than pull
• Some push firms are into content (e.g., PointCast), and others into technology (e.g., Marimba)

One Example:  Marimba's CastaNet

Overview

• Used primarily for Java applets
• Can be used for many media types (especially html)
• Castanet downloads and installs software,automatically updates it, and automatically removes obsolete versions.
•  Castanet is scalable through repeaters to distribute channels to thousands or millions of users from a single administrative sites.
•  Castanet proxies make the system work efficiently across corporate firewalls.

Channels (from Marimba's documentation):

Like applets, channels are:

• Multi-platform: The same code can run on Wintel, Solaris, and other computers.
• Downloaded: You get channels from the Internet.
• Secure: Channels are subject to the same constraints (with one relaxation) as applets.

• Stand-alone: Channels run in their own windows, not in a Web page.
• Local: Channels are stored on your hard disk where they are available when the network isn't.
• Persistent: Channels can read and write into one designated directory of your hard disk, preserving channel documents and preferences without threatening files that don't belong to them.

• Automatically updated: You don't discover new versions, the tuner brings them to you; a channel can even be updated while it's running.
• Bidirectional: Channels can return user-provided data to developers, such as counts of feature usage and errors.
• Customizable: developer software called transmitter plug-ins can modify channels as they are transmitted to users."

• Automatically distribute the latest version of software to a large organization.  (No more loading new versions of software onto your PC by hand.)
• Education channels could distribute a new lesson each week.
• Sports channels could distribute new sports scores every fifteen minutes.
• Software developers can distribute enhancements, bug fixes, and patches frequently.  (You could have a new release of your software every week!)

Subscribing to a channel

• "You subscribe to channels with the Castanet tuner.
• Subscribing downloads the channel, installs it on your hard disk, and launches it.
• Subscribing also initiates the automatic update schedule (hourly or weekly, for example) saved in the channel by its developer; you can also direct the tuner to update a channel whenever you wish.
• You can launch a subscribed channel from the tuner, from a Web Browser, or, from a platform-specific launch facility such as a shortcut or alias."

Castanet Tuner

• "The receiver component of the Castanet software distribution system.
• Tuner is a small Java application that does three main things:
• Subscribes to new channels by downloading and installing them on the users disk.
• Updates subscribed channels by pulling changes from transmitters; updates can be performed automatically or manually.
• Sarts and monitors execution of channels."
• Each channel carries an update schedule provided by its developer.
• When it's time to update a channel, or you direct the tuner to so so, the tuner connects to the channel's transmitter.
• The tuner and transmitter compare the files constituting the channel. If the transmitter has a newer version of a channel file (or a new file) it sends the file to the tuner.
• Channel updates are differential: only files that have changed since the channel was last updated are transmitted to the tuner.
• If a running channel's developer has indicated that the channel can be updated on the fly, the tuner installs the new files immediately; otherwise it waits until the channel stops.
• Channel updates are performed under the protection of a transaction so that channel files are always consistent with one another, even if power fails in the midst of an update.
• Tuners are themselves channels, and automatically update themselves once a week.

Castanet Transmitter

• Transmitter is analogous to Web server
• Transmitter can be stand-alone software or Web server extension
• "Unlike HTTP servers, transmitters send tuners only the files that are out of date on the tuner's machine, and they send all of the files in a single connection"
• When developer wants to send new version of files to users,
• developer points Castanet publisher to new files
• Publisher copies changed files to channel
• Transmitter serves updates to tuners while rest of files are being loaded
• Per-channel plug-ins available:
• Could select language (e.g., French vs. English) that tuner user prefers
• Could report how often tuner user clicked on advertisements

Push product offerings:

Partnerships

Netscape's Netcaster is based on Marimba. It will have channels for (see http://www.pagecrafter.com/lighthouse/push.html):

• Marimba
• ABCNews.com
• CBS Sportsline
• CNNfn
• Gartner Group
• Sesame Street
• Planet Out
• City Search.

• Wall Street Journal Interactive Edition
• Fortune
• Forbes
• first Call
• Newspage
• Dun and Bradstreet Online Channel
• Lexis Nexis and several other publications from Reed Elsevier.

Pull Acceleration: Caching Proxy Servers

(But caching has limitations.)

How proxies work:

• Client configures Web browser to use a proxy P either for caching or as firewall
• When client requests URL http://S/doc.html, request is sent to P (not S).
• If P has doc.html, and it's not expired, P returns doc.html to client.
• Else if doc.html is expired, send GET-IF-MODIFIED with last modification time to server S. S either returns new version of doc, or a return code indicating P's copy is still up to date.
• Else P doesn't have doc.html, and issues GET to server S.

Popular as proxy-only in US, essential as caches outside US

Caches can optimize one of three things:

2. Server hit rate perhaps retain popular documents

3. Time users wait to load document perhaps retain documents from other countries

Caching proxies be used in four locations

Inter-cache protocol ICP allows caches to share data items.

Some popular proxies:

• Harvest (outdated today, but introduced ICP)
• Squid (public domain decendant of Harvest, used by most ISPs worldwide)
• Cache3D (commercialized version of Harvest, owned by NetApp)
• Netscape Proxy Server (department level proxy; comes with SuiteSpot)
• Microsoft Proxy (department level proxy, and semi-firewall; comes with BackOffice)
• CERN Proxy

Return to CS6204 home page.

Last modified on 30 May 1997.

Send comments to abrams@vt.edu.
[This is http://ei.cs.vt.edu/~jwww/courseNotes/PushPullProxy/index.html.]