Oracle Corporation

Interactive Media Server Group

The Edge Server

A Proposal for Internet Media Servers

Introduction

By any measure, the Internet, more specifically the graphical World Wide Web (WWW) is a resounding success. However, that very success may be its undoing. The simple facts are that the amount of WWW subscribers, content providers, and requests by those subscribers for said content grows exponentially faster than the capability of the network to meet the demand and it's quite likely we will never get ahead of this curve. The business and trade press continually preach of its demise, citing lengthy delays and outages, warning of the impending gridlock when the infrastructure totally collapses. This gloomy prediction is based on a profound, new paradigm shift…when the text and graphics of today's WWW evolves into tomorrow's world of full motion video and audio. The real-time delivery requirements of video and audio demand a contention-free data channel.

You may ask "haven't technical innovation and competition always solved issues like this in the past? " Perhaps, but this particular problem is quite complex:

1. The very nature of the web creates a massive traffic congestion problem. Users don't know and don't care where the information is stored, what type of computer it's stored on, what application was used to create it, etc. he same flexibility and ease-of-use features that made the WWW such a success results in a serious contention issue since everyone competes with everyone else for available network resources. Since the Internet is IP-based, all packets must be evaluated by routers to determine the destination delivery paths.
2. Backbone, Sub-Net, and Router upgrades are not sufficient. case in point; MCI Corporation has quadrupled their backbone circuits in the past 12 months and they still do not have the capacity to meet the demands placed on their network. Most experts agree that Internet throughput cannot increase substantially enough to offset the increasing bandwidth requirements of the WWW itself.
3. Developing and deploying faster access methodologies are not sufficient . Besides faster analog modems, there are a plethora of new access technologies being market tested with the promise of significantly improving the connection between the end user's computer and the ISP POP. ISDN offers up to 128Kb/sec, various Digital Subscriber Line (xDSL) technologies and cable modems promise more than 6Mb/sec and may be eclipsing LAN speeds in the next 12 months. Why won't innovation carry us past the bottleneck known as the last mile? After all, we are increasing the performance of today's fastest modems by a factor of 20X or more! The answer is that 'bigger pipes' to the POP simply send bigger chunks of data onto the web. Moreover, history has proven that improved access takes many years to become pervasive. Just consider how long 28.8Kb/sec analog and ISDN 128Kb/sec have been available and they are not considered pervasive by any definition.
4. QoS solutions are not sufficient. Real-time protocols such as RTP, RSVP, etc. and specialized backbones (like M-BONE) go a long way towards improving transports for scheduled or premium events, but are unsuitable for the proliferation of multimedia content that is expected in the near future.
5. Streaming Video and Audio will exacerbate the problem. Today, multimedia data types represent a minuscule fragment of the total WWW content, but video and audio are quickly becoming more prevalent at many web sites. Hyperlinks point to multimedia files that are downloaded in their entirety to user's local disks. New streaming technology allows clients (equipped with compatible web browsers) to watch live video and audio as the file is downloading. While streaming promises to greatly improve the delivery problem, what will occur when tens of thousands or even millions of streams hit the already overtaxed infrastructure? And while improved compression techniques promise to squeeze multimedia files sizes smaller and smaller, video and audio will continue to require a 'big pipe' as a result of the real-time transport requirements.

The Edge Server Solution

Oracle Corporation believes the only solution to this problem is to place the multimedia data repository closer to the consumer of the information. The manifestation of this belief is the 'Edge Server'. The Edge Server caches and/or mirrors multimedia content otherwise stored on the WWW. The principle of the Edge Server could be applied to any web-based information but we should start with video & audio, solve the worst case problems first and then apply the same architectural model to less bandwidth-intensive data types.

In the Edge Server, multimedia data is replicated at the edge of the WWW, where the user's connection terminates at the POP. Hyperlinks on web pages become pointers to streaming media servers that are physically closest to the consumer.

Deployment at the POP

Oracle proposes that an Edge Server gets deployed at each POP. The POP is the physical interface since standard network connections are available such as Ethernet, ATM, and WAN. The POP is the also the logical termination of the user's access point. Packets flowing into or out of the POP are only limited by the access speed of the user's connection. Any data packets that flow behind or through the ISP backchannel (e.g. router) can encounter serious routing delays and contention latency, usually slowing the effective data rate.

By placing the media repository at the POP and behind the router, the user is insulated from the traffic conditions that exist on the Internet at any given time. Network outages and traffic gridlock are held transparent (once data starts streaming from the edge server), so the client device receives the stream as fast as the access connection allows.

Consider a trip to the airport. Your automobile may be capable of traveling much faster than the posted speed limit and it might take 30 minutes to get there. However, if traffic is heavy, the same trip could take 2 hours or more. If you could board a jet from your own neighborhood (without the noise, of course), your trip would be much shorter, more predictable and far less stressful.

Although ISPs do not typically operate large server farms, they offer co-location of web servers and many are installing proxy servers that cache most recently accessed URLs. By necessity they must expand their systems administration capability and add value or cease to exist in the face of competition (i.e. AT&T WorldNet, CompuServe, etc.).

Mirroring vs. Caching - Push or Pull

Content providers and web publishers can use GUI tools to stage, propagate and update multimedia data to replicated, distributed Edge Server repositories in a mirroring model. An alternative methodology might involve caching algorithms that update edge servers after first user touches a URL. If the requested data was not already cached, a dialog box could inform the user with the approximate time their media would be available, and might suggest they visit other sites in the interim. Either methodology has its strengths and weaknesses.

Push or mirroring tends to increase the likelihood that a URL is stored locally on the Edge Server and does not incur the initial write cost of a caching model, but requires more storage and must be constantly refreshed. Pull or caching models would incur a high cost for the first 'hit', any additional requests would be fulfilled locally and the Edge Server would be more likely to contain the latest update. Oracle believes a combination mirroring/caching hybrid may provide the most flexibility and the ability to manage tradeoffs between performance, storage space, popularity of content and frequent updates.

Once the data is in place, users make requests such as clicking on hyperlinks and software will broker these requests to the appropriate Edge Server based on the POP they are connected to. Only pointer and other control information moves across the core of the WWW, thus conserving precious Internet bandwidth. Differential Writeback technology could be utilized to just send just the updates instead of the entire file, further reducing network traffic. Logging agents will ensure web publishers and content providers retain the critical demographic data required to identify domains, audit trails, etc.

The Challenge

Oracle Corporation is proposing that W3C members collaborate together and define the most efficient architecture that will facilitate distributed edge servers on the periphery of the Internet and explicitly allow:

1. Content to be migrated from central or staging servers to edge servers
2. Content to be refreshed or updated transparently
3. Hyperlinks to dispatch multimedia data requests to edge servers
4. Logging feedback loop to web pages and central servers

Conclusion

Response time is slow and getting slower. When content providers and web publishers roll out streaming, real time multimedia data in large volumes, the infrastructure of the World Wide Web will collapse as hundreds of thousands of concurrent streams attempt to traverse an already clogged Internet. Network upgrades or faster access cannot be expected to remedy the expected gridlock. Oracle Corporation proposes the Edge Server; a streaming multimedia server that is replicated and deployed at the ISP POP on the periphery of the WWW. Web hyperlinks will point to the Edge Server that is physically closest to the consumer of the information. Edge Servers will enable bandwidth-intensive video and audio content to be played at the fastest possible speed that the user's access connection can support, insulated from the uncertain conditions that regularly occur on the web.