Strategic Computing and Communications Technology
CS 294-3, EE290X, BA296-11, and SIMS 290-2
Spring Semester 1998

Inktomi's Traffic Server

Analysis and Recommendations

Pedro Ferreira
Jonathan Martin
Jun Yang
Han-Shen Yuan
 

Table of Contents

 1. Introduction

 2. Customers

 3. Competition

 4. Complementors & Strategic Alliances

 5. Success or Failure?!

 6. Summary

 7. References

1.  Introduction

The rampant adoption of the Internet has produced the greatest infrastructure build-out challenge in history—that of providing sufficient network bandwidth to meet the overwhelming demands. Fortunately, a unique property of the Internet can be exploited to meet this challenge. This property is the high frequency of redundant content access. Caching refers to the local storage of frequently used electronic data to reduce access time. Network caching makes networks smarter by making it possible to move information once and then store it close to a group of users. The stored information is then used to service redundant requests from the group. Injecting this intelligent storage capacity into the network infrastructure allows the network to automatically optimize its traffic flow and reduce wasteful resource usage. It eliminates the need to transport data repeatedly over long haul links, and translates into more efficient systems, which provide higher performance at a lower cost.

Network caching technologies and implementations are in their infancy today. There are a number of obstacles to the successful use of  caches in the infrastructure of the Internet. These range from the ability to handle millions of users to the ability to economically provide reliable networking infrastructure systems.

Inktomi develops scalable network applications that improve Internet information access and delivery. These applications deliver high performance by leveraging Inktomi's unique parallel and cluster computing technologies. Inktomi applications, including the world's largest search engines and carrier-class network cache systems, are designed for Internet service providers, backbone carriers and other customers seeking to solve large problems.

2.  Customers

There are three broad market segments with very different performance and scaling requirements.   At the high-end of the market are the large ISPs and backbone providers. The middle tier of the market is characterized by smaller ISPs and large corporations. Finally, individual users make up the lower end of the market and they are primarily catered to by browsers. The market segments are described below in more detail.
 
Fig 1: Network Caching Market Players and Segments
 
Large ISPs and Backbone Providers
Well-designed carrier-class hardware tends to follow Moore's law -- every 18 to 24 months the capacity of hardware devices double, while their price remains stable.  However, Moore's law does not apply to bandwidth -- available WAN bandwidth remains constrained and long-haul infrastructure upgrades remain expensive.  Injecting intelligent caches into the network infrastructure allows the network to automatically optimize its traffic flow, and reduce wasteful resource usage thus alleviating the constrained bandwidth problem to some degree.  This market segment is primarily composed of large ISPs and backbone providers, like UUNET, MCI, PSI, AT&T, AOL, Compuserve, and Sprint, who have very high performance requirements and require that the caches scale to several terabytes of data.

Small ISPs and Corporations
Data networking is growing at a dizzying rate.  Over 80 percent of Fortune 500 companies have Web sites.  More than half of these companies have implemented intranets, and are putting graphically rich data onto the corporate WAN.  The resulting uncontrolled growth of Web access requirements is straining all attempts to meet the bandwidth demand.  Caching eliminates the need to transport data repeatedly over long-haul links, and translates into more efficient systems which provide higher performance at a lower cost.  Large corporations can reduce their web connectivity charges by installing caches at strategic points in their intranets.

Individual Users
This segment caters to the needs of individual users and hence has low performance and scalability requirements. With its cluster based  high performance solution Inktomi is more suited to compete in the two segments listed above, so we will not be analyzing this market segment.

3.  Competition

The major competitors in the high and mid-end market segments are analyzed.  Their product performance and features are summarized below:

3.1 High-end competitors

Cisco
Cisco’s Cache Director System is a carrier-class solution -- inherently fast, scalable, hierarchical, fault tolerant/fail safe and simple to deploy. The special-purpose hardware/software solution is optimized to serve two primary goals: to reduce WAN access bandwidth requirements and to speed Web access for users.

The Cisco Cache Engine requires no client configuration, making it transparent to the user. Up to 32 cache engines can be combined to form a cache farm, scaling performance and storage linearly as Cache Engines are added. To reduce WAN access costs, enterprise customers can install a Cache Engine (or a cache farm) near the router that provides the Internet connection at the top of the network hierarchy. In addition, customers can place a Cache Engine at remote offices that connect to the main office via WAN links. This arrangement reduces congestion at the central servers and reduces bandwidth usage on the intersite links.

3.2 Mid-end Competitors

Novell
BorderManager offers easily administered high-performance proxy cache support, complete integrated firewall protection, remote access, and virtual private networks.  BorderManager FastCache™ is marketed as a cost-effective and easy-to-use software package that significantly accelerates Web access performance (e.g., delivers Web-based content to users 10 to 200 times faster than traditional Web access methods), reduces network/intranet traffic, and offloads Web server workload by up to 90 percent. This new addition to the Novell BorderManager family is rated at over 4,000 active connections per second -- four to ten times more scalable than any competing cache product -- and can service 85,000 active connections using Intel architecture servers. BorderManager FastCache also enables businesses and Internet Service Providers (ISPs) to improve customer service by dramatically boosting the performance and responsiveness of their hosted Web sites.

Network Appliance
NetApp’s NetCache is a high performance and secure software product that allows network administrators to replicate (or cache) Web content. This reduces bandwidth costs by avoiding duplicate requests generated over the WAN, and significantly speeds up access for users. NetCache also provides greater control and selection of information permitted over the network. The product is designed for enterprise-wide corporate intranets and for Internet service providers (ISPs) who have multiple points-of-presence (POPs) and data centers.
 
 

Comparison of Performance and Features 
Company Features Performance Data Hardware Requirement
Cisco No client configuration, transparent to the user, hierarchical, time sensitivity, scalable, fault tolerant/fail safe and simple to deploy. 
$30,000		 32 cache engine farm supports up to 500,000 at single point presence, store 25 mil web pages; providing 768 GB of cache storage and 28,800 concurrent session Supported on Cisco 7200 and 7500 series routers with 10BaseT and 100BaseTx interfaces
Inktomi Scalability of users; scalability of content, fault tolerance, transparent to users A 16 node cluster of Sun Ultra 2s with a half TB cache supported over 3,488 hits/second saving 147.7Mbps of bandwidth Sun SuperSPARC or UltraSPARC systems with Solaris 2.5.1
Novell Scalable, hierarchical caching , firewall filtering, single point administration, virtual private networking, address translation, routing, and remote access services 4000 active connections/s with a cache size of a few GB, service unlimited # of web users Industry standard Intel processor-based hardware and compatible with existing Novell, UNIX, and NT networks
Netapp Scalable, easy to install, administered remotely using browser interface, transparent proxy capability, block filter or redirect URL 637,500 URLs/hour at under 1 s response time, NetCache provides support for unlimited URLs/hour while NetCache Lite supports a maximum of 15,000 URL/hour Supports Windows NT, SPARC solaris and Digital Unix
 

4.  Complementors & Strategic Alliances

Some of the complementing factors which will give a boost the industry are enumerated below. In choosing these strategic alliances, Inktomi will need to aware of the following: Currently, Inktomi has formed strategic alliances with Sun Microsystem, Intel,  UUNET and NTT in the network caching market.  These prominent corporations have added substantial credibility to Inktomi.

5.  Success or Failure ?!

On the technology side, Traffic Server is clearly the leader.  The success or failure of Traffic Server (TS) will be determined by the business strategy.  Looking through the recent history, there have been cases in which the technological leader did not emerge as the final winner as in the case of Beta vs. VHS.  The business strategies are analyzed for each of the market segment.  In each case, we look at the value proposition of the customers and compare that will the product attributes of Inktomi, and find out what kind of resources are need to "complete the package".  Based on the resources required, we then recommended strategies accordingly.

5.1 High-end Market - Backbone provider/ISP market
 

5.2 Mid-end Market - Enterprise & Small ISP Market
  In this market segment, the enterprises do not require as much scalability and cache size as the high-end market.  In addition, ease of administration, customer support and security are extremely important.  Traffic Server lacks some of these desirable features and Inktomi lacks some of the key parts of the value chain.   Marketing/Sales, technical support, and distribution are the main complementary assets within the value chain that Inktomi must evaluate.  Although the marketing and selling of Inktomi products may be farmed out to a strategic partner, such as Sun (i.e., through bundling of Traffic Server with Sun hardware), these functions are so vital to the diffusion of the products into the market, that we conclude that the marketing and selling functions are specialized complementary assets.  In other words, the marketing and selling approach should highlight Inktomi solutions (as opposed to a Sun solution with an Inktomi input).  To ensure this, the program should be executed by an internal marketing department and sales force.  Considering the potential for coordination and incentive problems in a partnering arrangement, developing internal sales and marketing skills will likely be less costly to the company in the long-run (although potentially more costly in the short-run).  As Inktomi considers competing in the enterprise space with a caching product, these skills will become even more important (i.e., because relationships with and direct feedback links from enterprise customers are very important).

6.  Summary

In the high-end market, Inktomi has effectively leveraged off their technological leadership, first mover advantage and utilized their strategic alliances to become the dominant player in this market segment.

In the mid-end market, Inktomi faces a number of challenges.

  1. Lack of marketing/Sales, technical support, and distribution force.  Inktomi needs to evaluate methods to acquire these complementary assets within the value chain.
  2. Inktomi is relatively unknown in the enterprise market.  Forming strategic alliances with prominent corporations should increase market awareness and Inktomi's credibility.
  3. Potential entrants. Right now, Inktomi is the cache leader.  But industry experts say it's only a matter of time before Cisco, 3Com, and computer giants such as Sun and IBM try to wrest this type of business away.  Already, these giants are testing new traffic-management tools on private corporate networks.  If the new approaches speed things up on smaller LANs, then they can probably be scaled up for the public Internet.
Overall, we believe that forming strategic alliances to complete the virtual value chain is the key to Inktomi's success.  However, Inktomi needs to proceed with caution in order to retain its core competency in technology and prevent today's partners becoming tomorrow's competitors.

7.  References

  1. Jack Richard, "Internet Architecture", Boardwatch  Magazine,  http://www.boardwatch.com/isp/archit.htm
  2. Emerging Technologies Research Group, http://etrg.findsvp.com/index.html
  3. Bestavros  et al., "Application-level document  caching in the Internet", Workshop on Services in Distributed and Networked Environments, Summer 1995. ftp://cs-ftp.bu.edu/techreports/95-002-web-client-caching.ps.Z
  4. James Gwertzman and  Margo   Seltzer, "The case for geographical push-caching", HotOS Conference, 1994. ftp://das-ftp.harvard.edu/techreports/tr-34-94.ps.gz
  5. http://www.mci.net
  6. http://www.uunet.net
  7. http://www.netapp.com
  8. http://www.cisco.com