Computer Scientists and Industry Partners Discuss Advances in Networked Computing at CNS Research Review
By Tiffany Fox, (858) 246-0353, tfox@ucsd.edu
San Diego, CA, Jan. 20, 2008 — The world is on the brink of a revolution in computing that will make any type of data available to us, anytime, anywhere. But this transformation isn’t happening in a vacuum, and the computer scientists and engineers at the University of California, San Diego's Center for Networked Systems (CNS) are actively involved in making networked infrastructure faster, safer, more energy-efficient and increasingly reliant on Internet or "cloud" based computing.
|
The members of CNS – which include several researchers at the UC San Diego division of the California Institute of Telecommunications and Information Technology (Calit2) – held their half-yearly research review last week to outline progress-to-date on their collective projects across all aspects of networked systems. In addition, more than 40 industrial participants from CNS member companies presented some of the top challenges facing their respective organizations, and provided feedback on the ongoing work at the center.
The two-day review, which took place in UCSD's Computer Science and Engineering (CSE) building Jan. 14 and 15, spanned a variety of topics, from conserving system memory through a technique known as "identical page sharing," to detecting malicious Web sites by analyzing the various characteristics associated with their URLs. CNS graduate students played a significant role in the review, both as presenters and as participants in a poster session and reception that showcased their work.
“The CNS research presentations showcased both the breadth and depth of our work in networked systems,” remarked Amin Vahdat, director of CNS and a professor of CSE in UCSD’s Jacobs School of Engineering. “Our faculty, staff, and graduate students are working toward a vision of computing and storage delivered efficiently, reliably, securely, and with low power. All presentations were made by our graduate students, giving them invaluable presentation experience and showcasing their talents for our member companies.”
Also speaking at the review were representatives from the computing industry, including Google's Technical Manager, Bob Felderman; Motorola Senior Fellow and Chief Architect Hamid Ahmadi; Cisco Distinguished Engineer Flavio Bonomi; and the event's keynote speaker, Amazon.com's Chief Technology Officer and Vice President Werner Vogels, who presented a talk titled "Ahead in the Cloud: The Power of Infrastructure as a Service.”
Cloud computing was a recurring theme at the review, and a number of presentations addressed the challenges inherent to this emerging style of computing. The general concept of cloud computing incorporates "software as a service," whereby a resource like Google Docs is provided over the Internet to users who need not have any expertise or control over the technology. Presenter Kevin Webb, a CSE graduate student, discussed techniques for "Seeding Cloud-Based Services" by addressing the need for distributed resource control and processing unstructured data (the associated research was conducted by Vahdat, CSE researcher Ken Yocum and CSE Associate Professor Alex Snoeren). In a presentation titled "Bluesky: System Support for Transparent Could Computing," CSE grad student James Anderson discussed research conducted by Associate Professor Geoffrey Voelker and Assistant Professor Stefan Savage (both of CSE), which looked at ways to enhance networked systems built in a “transparent cloud” model. Transparent cloud computing is akin to using an out-sourced virtual data center, where organizations can decide how much computing power they need and what hardware/software systems they require.
|
The advent of cloud computing has placed a consequent emphasis on applications, and several of the CNS presentations focused on improving the ways applications are developed and shared among machines. A presentation by graduate student Cynthia Taylor, for example, investigated "Proximal Resource Architectures for Thinner Client Computing," which was essentially a primer on how to get applications like video to run faster and more smoothly on "thin-clients" like iPhones and other tiny machines. Taylor advised using a cross-platform system called Virtual Network Computing (VNC) to speed up the “send-receive” update loop between clients and servers. By using the VNC with a smart proxy for added performance, the researchers were able to achieve a rate of 13 data updates per second, as opposed to 2.5 updates per second with no smart proxy.
Explained Taylor: "In conclusion, we can improve VNC performance by having the smart proxy mediate the update rate despite high amounts of latency," or server-client lag time. "And we can do this in a way that doesn't change the existing source code. Our future work will focus on significantly increasing performance and adding functionality to the smart proxy, like face-detection technology for video applications."
The explosion of Internet-based "killer apps" and the simultaneous consolidation of datacenters have also posed a problem in terms of machine memory limits – a topic that was discussed in a presentation by CSE grad Michael Vrable on "Extending Virtual Cluster Management and Resource Utilization." The research, which was conducted by Savage, Sneoren, Varghese, Voelker, and Vahdat, sought to minimize the hurdles involved in sharing memory among virtual machines. Noting that "memory is a key bottleneck to running virtual machines," Vrable explained the team's proposed solution: A system called "Difference Engine" that couples "identical page sharing" with methods of patching and compressing memory pages that are similar, but not identical.
Noted Vrable: "With homogenous workloads, we were able to achieve 80 percent memory savings, with most of the savings coming from identical page sharing. With heterogeneous workflows, we were still able to reclaim substantial amounts of memory savings, and patching contributed significantly to those savings."
Keeping the Internet safe and free of scams and malicious Web sites is also a priority for CNS, and two presentations focused on ways to identify and analyze the many ways spammers and hackers try to target users. A lecture titled "Beyond Blacklists: Learning to Detect Malicious Web Sites from Suspicious URLs ," for example, provided a number of lexical features and IP-based characteristics for detecting malicious URLs, spam phishing and other exploits.
"There are various characteristics associated with these sites," said CSE grad/presenter Justin Ma. "The question is: How do you relate these properties of the URLs to the maliciousness of the Web sites?"
|
To find out, Voelker and Savage drew malicious URLs from those submitted to "phish tanks" by online users, and compared them with benign URLS from certain online directories that had been previously vetted for validity. Using a probabilistic linear model called "logistic regression" as a classifier, they reduced a set of 30,000 URL features down to 4,000 features for model analysis. They discovered that certain "red flags" indicate malicious intent, including 1) suspicious ownership of the site, 2) where the site is hosted geographically, 3) the registration date of the site, 4) what kind of connection the server is using, and 5) the presence of certain URL extensions. The extension".com," for example, tends to signify a malicious Web site when it is found in the middle of a URL (i.e. "bankofamerica.com" is probably fine, but "bankofamerica.com.cz.rnl" should raise some eyebrows). Ultimately, the researchers would like to create a URL reputation service that will allow users to query URLs via a database to determine their validity.
With more and more users not only downloading Web content but also creating it, security is a primary concern — but so is profit. Motorola's Hamid Ahmadi outlined his company's approach to expanding its market sphere of influence in a presentation called "Multimedia and Information without Limit: A Motorola Applied Research and Technology Center." Ahmadi said Motorola's main focus at the moment is to use data in a more analytical way to enable new services — or, primarily, "bringing multimedia and video capabilities to devices."
"The goal we have is to potentially create new business and new business opportunities," he said, noting that the core components of Motorola's current approach are broadband access, intelligent edge networking, next-generation video experiences, platforms and infrastructure. "But not all the research we do is internal — we have more and more joint projects with UCSD and other big universities. “
Ahmadi also pointed out the current challenges in the market, which include the gaps between those who create Internet content (YouTube, for example), those who aggregate or collect it (e.g. iTunes) and those, like Motorola, who distribute it through their devices.
"In the future, the promise will be 'any content on any device, any time you want it,’" he predicted. "For example, right now, television programming is linear and not on-demand. Next-generation TV will be based on your schedule, and will be mobile." He even predicted that TV will become social, with Motorola-designed set-top boxes making it possible for users to create videos of themselves in their living rooms and immediately broadcast the footage to others' television sets in real-time.
Cisco Systems' Flavio Bonomi made similarly mind-boggling predictions in his presentation, "Towards a More Application Aware Networking." By understanding the requirements of various applications and by allocating computing, storage and networking resources effectively across distributed systems, Bonomi expects that the Internet will evolve from its current arena of computers and hand-held devices to become the "Internet of things." It will support the interconnection of a huge number of smaller devices, including a proliferation of sensors, and also will connect "the elephant in the room," he says, "which is the car."
|
Savage noted that researchers and CNS and beyond are eager to develop such technologies, but are often stymied by a lack of appropriate platforms and test beds. Vahdat suggested that companies like Motorola and Cisco set up test beds at universities so that researchers could "live the experience" and serve as guinea pigs – and collaborators – as the new technologies are developed.
Such collaborations are on the radar for CNS, which works with its member companies on all active research projects. In the past, CNS researchers worked with AT&T to deploy more efficient routing protocols running in the core of AT&T’s global network. Currently, the unique perspectives of CNS member companies heavily influence the center’s work on future data center networks and on Spam analysis, and a number of joint publications and patents have come from ongoing collaborations between Sun and CNS interns.
Looking to the future, Vahdat predicts that computing is at another transition point in its history.
“With near-ubiquitous wireless network coverage and continued exponential increase in available computing power, network speed, and storage capacity, we are approaching a time when all of our data, from applications, documents, video, music, photos, movies, to medical records will be available to us instantly at any time,” he noted. “Achieving such a vision will require a re-architecting of the global network from the core all the way to its edge, considering security, privacy, wireless and cellular protocols as first-class requirements. The infrastructure will be delivered by emerging mega data centers, consisting of upwards of 100,000 compute nodes with petabits/sec of network bandwidth and exabytes of storage. As we saw in the research review, CNS researchers – in collaboration with their industry partners – are at the forefront of this activity.”
Media Contacts
Tiffany Fox, (858) 246-0353, tfox@ucsd.edu
Related Links
Center for Networked Systems