"To win the battle for public opinion in America, you have to present yourself as a victim; that is how you win the heart of Americans. He who apologizes for using force loses.."
With supply chains stretching to China, customers as likely in Paris as Peoria, outsourcing operations in India and development in Russia, the reality of today's economy is that we must be global to succeed. But how do we succeed at being global? Here are three areas where most American companies need to get better, if they are going to succeed globally.
"Exploring the New World a thousand years ago, a Viking woman gave birth to what is likely the first European-American baby. The discovery of the house the family built upon their return to Iceland has scholars rethinking the Norse sagas."
The EU struck at the commercial backbone of Microsoft’s global software empire today (12/22/04) – forcing the company for the first time to strip its Windows platform of a popular feature to give competitors a better shot at success.
An EU court ruled Microsoft must immediately divulge some trade secrets to competitors and produce a version of its flagship Windows operating system without the program that plays music and video.
If you haven’t read Franz Kafka’s The Metamorphosis, I recommend it for some profoundly humorous holiday reading. The subject of the book is a human who one day finds himself transformed into an insect, and as a result feels the stress of losing his humanity. I don’t know of a similar story of a bug becoming human, but Mary Shelley’s Frankenstein is a more widely recognized discourse on the challenges of architecting intelligence and experience “up the stack” and the ultimate and more complex issue of acceptance in a complex environment.
With the enterprise migration to Web-enabled applications, new levels of complexity are challenging IT managers and traditional switching vendors who’ve traditionally relied upon capacity to solve their problems. Recent surveys of eWeek and Computerworld readers suggest a Web tier quagmire driven by the failure of brute force capacity to address the performance and security issues inherent with complex, noisy Web applications. While enterprises are currently overspending on capacity-related gear, they are becoming increasingly aware that things aren’t going well. The research results are available at www.redlinenetworks.com and also at http://redline.alwayson-network.com/
The question is: Can you transform a low layer switching architecture into an elegant, effective, intelligent, experienced and acceptable (to complex Web applications) high layer processing architecture? Can you give a bug a brain and a consciousness’ to interact at a much higher level? Can you add an experience chip? The answer in networking is perhaps yes, but that the process takes time. Time, because we all know that experience counts.
Take adaptive content processing and compression for example. Compression algorithms are like central nervous systems, they’re fairly common in complex life forms, from bugs and earthworms to humans. But being able to comprehend, manipulate, correlate, etc is a more cerebral function based on experience and knowledge… which is much more challenging as an add-on. Directing traffic based on a few rules or a packet sample from a stream of data requires much less experience than being able to comprehend, manipulate and/or act on the entire meaningful stream of data in a way that can interoperate with much more complex applications in networks never designed for that noise and complexity.
This year has been, indeed, a year of metamorphosis for low layer server load balancing vendors now laying claims to becoming high layer application front ends. As Mary Shelley predicts, sometimes it isn’t about merely having critical organs either, but by having the judgment, experience and appearance required to interact in more complex environments. Effective compression requires a policy engine developed for thousands of exceptions. It requires knowledge and experience.
Then why are the SLB vendors quickly positioning themselves as application front ends? Next week I’ll share with you what we’re learning about the AFE market.
Tell us what you think of this post using our On or Off rating system. Only your most recent vote will count.
Member Comments
Thomas (post 2):
Essentially we're talking about anomaly management versus any kind of high level, proactive analysis. That will come from IT managers... but these devices will give those managers the tools to better manage the exception explosion. It managers will still be behind the wheel, driving the delivery of the applications, but they're getting better tools to manage the complexity.
I'm talking more specifically about dedicated devices that are capable of managing both the complexity and capacity requirements for delivering enterprise Web applications, especially the new Web-enabled versions of PeopleSoft, SAP, Siebel, Oracle, even custom Web apps. The first generation of Web applications were designed for the broswer, so they were more elegant, straightforward, designed with page download times in mind, etc. The next generation of applications moving via browser were not designed for the Web, but rather client-server access. As a result, the challenges that IT managers face with their delivery are much different.
The research I sited of (eWeek and Computerworld readers managing Web applications) suggests that increased capacity is not solving the challenges of these new Web-enabled applications. From a vendor perspective we're seeing the load balancer space cycle to solutions with higher level capabilities... ultimately these devices need to have high "anomaly quotients"; they need to be able to manage the explosion of exceptions that take place when applications never designed for browsers (and networks never designed for the noise and complexity of these applications) collide.
By intelligence at this point in the product cycle I'm really referring to the ability to manage the "exception explosion" which is driving IT managers crazy. Many have tested devices that can compress based on algorithms like GZIP but cannot handle the exceptional cases. For compression, for example, there are thousands of exceptions that can render the power of compression useless. Devices that can manage to those exceptions can deliver on the promise.
To embrace a racing metaphor, the early load balancing devices were ultimately about horsepower, because their tracks were relatively straightforward. Fat tires, huge engines, limited steering and visibility. Delivering these new Web applications is less like a drag strip and more like a Grand Prix, with lots of twists and turns... horsepower is still important, but management, flexibility, visibility is critical. The lesson of the eWeek and Computerworld reader surveys is that taking a dragster to a Grand Prix is like entering an Olympic weightlifting champion into Olympic Greco-Roman wrestling competition. Hope that helps...
Far from it.....................bandwitdth isn't eveerything..........but applications never have and never will drive bandwidth deploymnets; that was my point. It has been and always will be the other way around. The apparent overbuild of long hual tier one fiber-optic netowrks was not an investment anomly as some would think; those networks are imporatnt capacity enabling backhaul resources for wireless as well as cable operators. Level(3) deal with Comcast highlights this tighteing relationhip. So your assertion that "capacity not complexity" was the element of you statement I was challenging.
And to be honest IT managers should not have to be overly concerned about broadband networks ability to handle low latency real time web application------or as you have put it "Can you transform a low layer switching architecture into an elegant, effective, intelligent, experienced and acceptable (to complex Web applications) high layer processing architecture?" It is one the reason we have MPLS.
Are you telling me that eventually Front-end Application devices will be able to learn ?
Are you building algorithms into your products or perhaps even self-learning algorithms ?
Is Prolog embeddable and suitable for web application in networking environments ?
You're sounding more and more like one of those "bandwidth will solve everything" kinda guys. Some guy told me the same thing in '99 and gave me a bunch of stock tips. The rest is history...
I think we can see more intelligence both inside and outside the firewall for the enterprise delivering Web-enabled PeopleSoft, SAP, Siebel and custom apps. Inside the firewall there are architecture-level issues based on the proliferation of point products (as discussed)... more capacity and intelligence outside the firewall will help but not solve the device proliferation problem completely. It appears from your excellent posts that the network outside the firewall is also to go through a necessary metamorphosis. Thank you.
Researchers from Lucent Technologies' Bell Labs are presenting two papers based on innovative research this week at MobiCom 2004 in Philadelphia, the premier international forum for mobile computing and wireless networking. First, they'll describe a method for dynamically improving how data packets are routed through a wireless network by modifying its topology in response to changing traffic patterns and user demand. Next, they'll describe how the performance of wireless local area networks (WLANs) can be greatly improved by seamlessly shifting users from heavily loaded to lightly loaded access points - thereby relieving network congestion and increasing the number of users that can access the network at any given time. Both of the approaches described at the conference hold the promise of improving the performance, reliability and availability of wireless communications. This work is yet another example of how Lucent continues to push the envelope and lead the evolution towards high-speed mobile data.
In a paper called, "Network Deformation: Traffic-Aware Algorithms for Dynamically Reducing End-to-End Delay in Multi-hop Wireless Networks," the authors describe a method for estimating in real time the mean end-to-end transit time of packets through the network based only upon the size of the queues at the network nodes and the network layout. This estimate allows the researchers to develop a class of "topology modification algorithms" to dynamically reconfigure nodes, thereby creating new links and also changing the capacities of existing links. According to the researchers, these algorithms alleviate network congestion and improve overall routing performance because the maximum load that the network can carry, before the end-to-end transit time starts to increase without bound, is significantly increased as a consequence of the changes to network connectivity.
"Adding new connections to alleviate congestion and to speed the flow of traffic in a data network is not unlike building new roads or tunnels to do the same in the physical world," said Sayandev Mukherjee, a researcher in Bell Labs' Wireless Research Laboratory. "Our algorithms indicate that building additional connections between nodes upstream - before traffic even reaches a bottleneck link - will reduce the end-to-end transit time resulting in greatly improved routing performance. This makes more sense than simply enhancing the capacity of the bottleneck links themselves."
Other members of the research team include Sharad Ramanathan, a researcher in Bell Labs' Physical Sciences Laboratory; Anindya Basu, formerly of Bell Labs, and now working at Morgan Stanley; and Brian Boshes, a student at the University of California, Berkeley, who is also an alumni of Lucent's Global Science Scholars Program.
In the other paper being presented this week, "Fairness and Load Balancing in Wireless LANs Using Association Control," a new method for balancing the traffic load in WLANs based on the IEEE 802.11 specification is described. WLANs enable a person with a wireless-enabled computer or personal digital assistant to connect to the Internet by moving within, for example, 15 meters of an access point, called a "hotspot." Recent studies on operational WLANs have shown that user load is often unevenly distributed among wireless access points resulting in unfair bandwidth allocation among users. To rectify this imbalance, the researchers have developed an algorithm that can intelligently and efficiently shift users from heavily loaded to lightly loaded access points - which guarantees near optimal bandwidth allocation for all users. According to lab simulations, these techniques outperform commonly used heuristic approaches, and they could be used as the foundation of a practical network management system.
"In the presence of hotspots, our algorithms provide fair service to all users accessing the network, while also maximizing the amount of bandwidth they receive," said Yigal Bejerano, a researcher in Bell Labs' Internet Management Lab. Bejarano continued, "Typically our algorithms also yield higher network utilization than the most commonly used 'strongest signal approach, while today's approaches tend to focus on overall throughput when allocating network resources. We believe that understanding the correlation between fairness and load-balancing are critical in order to maximize bandwidth for all users."
Bejerano's collaborators include Seung-Jae Han of the Wireless Research Lab, and Li Li of the Networking Research Lab - a testament to the multi-disciplinary nature of the collaborations happening at Bell Labs.
MobiCom is a highly selective conference - with less than an eight percent acceptance rate for papers to be presented - focusing on all issues in mobile computing and wireless and mobile networking at the link layer and above. The conference regularly attracts over 500 of the world's top researchers, practitioners, students, and executives, who are active in bringing about the future of mobile computing and networking.
Capacity can help but only to a point. In the Web tier in front of these Web applications that data is having to make multiple round trips between various devices (load balancers, servers, cache, authentication, compression, etc.); ultimately the issue isn't how many lanes are on the freeway and the speed limit but rather how many stops (and return trips) cars have to make at each overpass. Add to the basic spider Web of trips between devices the challenges that come from Web application noise and you have a recipe for frustration... IT managers are scrambling to find out where the problem is from a device standpoint when the problem is architectural. Both the eWeek and Computerworld reseaarch referenced points to budget overruns and performance/security challenges.
IMHO the bigger picture is that networks handling the more complex Web applications require much more intelligence than pure switching power and bandwidth. All of the Web tier functionality scattered about in different devices needs to be consolidated, hence the application front end (the consolidated device that reduces hardware requirements) opportunity.
Application front ends can replace the load balancers and other Web tier point devices as well as teh Web servers currently in demand. They represent, therefore, a larger market opportunity that load balancers. Even the load balancer market is churning toward complexity management, as the more primitive load balancers are now being replaced by more sophisticated layer 4-7 appliances that are starting to resemble primitive AFEs (strong on scalability, SSL termination, etc. and with some competence in compression, caching, etc.) while the private AFE players are on very healthy growth ramps with admirable margins.
Thanks for your comment; your input would be more than welcome as well on the Redline Networks Keiretsu.
Good point Craig. The earlier Web applications essentially were for consumer-facing Websites, and were fairly standard and straightforward. IT managers were correctly more interested in capacity... because the challenge was serving content to more browsers. With enterprises using browsers for the likes of PeopleSoft, Siebel and other custom applications enterprises are now serving up more kinds of content. So the issue is no longer capacity (the server load balancers learned to scale to new heights relatively easily... it was about switching packets to more servers) but complexity. The AFEs bring higher level judgement and experience to the Web tier, with much more power and sophistication to address the challenges that arise when these new applications (not initially designed for browser access) and networks (never designed for the level of noise and complexity from these new Web applications) collide.
The trick to metamorphosis is pattern recognition, which must begin at the most simplied level before growing into a more complex (virtual) organism. Those applications that have the most repetitive, with the fewest deviations/permutations would be the best fit for transfroming low layer switching architectures. Does anyone have specific examples where this would work best?
CBS MarketWatch
Michael Powell Read This: FCC Needs Broadband Trust Fund
E-NEWSLETTER SERIES