TACC Update Daedalus/BARWAN Winter Retreat, January 1998

1. TACC UpdateDaedalus/BARWAN Winter Retreat,January 1998 Armando Fox fox@cs.berkeley.edu

2. Outline • TACC Review & Status • New TACC Services and CS 294-6: Internal Technology Transfer • Thin Clients & Computing By Proxy • Publication/Ongoing Research Activities • Summary

3. Layered Model of Internet Services httpd, etc. • TACC Layer • Programming model based on composable building blocks • SNS Layer: “large virtual server” • Implements SNS requirements • Cluster computing for hardware F/T and incremental scaling TACC ScalableNetwork Svc • Exploit TACC model semantics for software F/T • SNS layer is reusable and isolated from TACC • Focus on application content • Applications get SNS behaviors for free

4. What’s TACC? • Transformation (“one-to-one”) • TranSend, Anonymizer • Aggregation (“many-to-one”) • Search engines, crawlers, “my headlines” • Caching • Both original and TACC-generated content • Mass Customization • Per user: for content generation • Per device: data delivery, content “packaging”

5. C C T TACC Examples • HotBot search engine • Query crawler’s DB • Cache recent searches • Customize UI/presentation • TranSend transformation proxy • On-the-fly lossy compression of inline images (GIF, JPG, etc.) • Cache original & transformed • User specifies aggressiveness, “refinement” UI, etc. A $ DB T html $

6. $ C W W W W W W T A TACC Conceptual Data Flow To Internet FE User request • Front end accepts RPC-like user requests • Original data fetched from cache or Internet • User’s customization profile retrieved • Aggregators/transformers operate on data according to customization profile

7. Interconnect Logical TACC Server Architecture • Component replication for scaling and FT • NOW with scalable interconnect • Incremental scaling: commodity building blocks C FE $ $ $ FE W W W A W W W T FE GUI LB/FT

8. TACC Contributions • TACC programming model • Transformation, Aggregation, Caching, Customization can all be composed • Encourages software structure that admits of simple F/T mechanisms • Captures a large class of services • Cluster-based framework to instantiate TACC • Allows TACC authors to ignore SNS requirements • Performance, scalability, and failure management adequate for real workloads • Stock hardware

9. TACC Server Status • “Production” cluster running at UCB since 4/97 • TranSend web accelerator • Virtually zero downtime and zero maintenance • Software distribution in progress • Binaries for Solaris and Linux • Source snapshot • “SDK” and additional modules coming soon • Used as testbed for CS 294-6 graduate seminar

10. CS 294-6: Internet Services • Survey: design & implementation of Internet services • Design and implement a service using TACC • Best ones deployed permanently, like TranSend • Real developers: a way to stress the software infrastructure (and documentation!) • Encourage other research projects to leverage TACC

11. New Services Deployed • Safe worker upload and execution • Cha-Cha, combining browsing and searching • Group Annotation for the Web • Top Gun Wingman, world’s only graphical Web browser for the Palm Pilot • Top Gun MediaBoard • RTSP Archive Server • Others… http://www.cs.berkeley.edu/~fox/cs294-6/

12. Safe Worker Upload/Execution • ISAAC group: David Wagner, Prof. Eric Brewer • Any user can author a new worker/module • New workers are uploaded dynamically into a running TACC server • TACC “route selection” forms-based interface used to select worker(s) for each MIME type • Workers run in a Janus sandbox

13. Cha-Cha: Contextualizing Search • BAILANDO project: Mike Chen, Jason Hong, Prof. Marti Hearst (UCB SIMS) • TACC implementation improves performance & scaling of Cha-Cha prototype Query(A) HotBot html Cha-Cha Combine(A) Layout(T)

14. Anno. DB Stock Webservers Group Annotation • Multivalent Documents/Digital Libraries project: Marcel Kornacker, Ray Gilstrap, Tom Phelps, Prof. Robert Wilensky • Annotate the Web, without owning it Web browser Anno.Svr. MVDEditor(Java) UI

15. RTSP Archive Server Front End • MASH project: Angela Schuett, Suchi Raman, Prof. Steve McCanne • Goal: Add scalability and availability using TACC

16. Thin Clients: Top Gun Wingman • World’s only graphical browser for PalmPilot handheld device • >5000 users worldwide • Richest feature set of any Pilot browser, many uniquely enabled by proxy • #9 out of 100 in “People’s Choice” awards

17. Groupware: Top Gun MediaBoard • mb comes to the Pilot • Combines MASH tools with TACC server • Enables future MASH collaboration • Inherits TACC robustness & scalability Superiority of proxy-based approach to enabling “thin” clients successfully demonstrated.

18. Publication Activities • Invited submission to IEEE Personal Comms. • Special issue: Adapting to network and client variability • Structure, data semantics, and implementation of interactive Internet applications • Collaboration with Dr. Murray Mazer, Open Group Research Institute • Past work on annotating the Web and using HTTP proxies to build Web applications • Top Gun Wingman

19. Summary • Validation of proxy architecture for thin clients • Wingman is the fastest browser and the only one with graphics; MediaBoard is unprecedented • (Continuing) validation of methodology & platform for building scalable Internet services • Successful internal & external tech transfer • CS 294-6 projects • Wingman, MediaBoard, MASH collaboration… • TranSend at UC Davis, SRI, soon Lucent • Next time: the eye-watering demo

20. Credits • Top Gun Wingman: Ian Goldberg, Anthony Polito, David Lee • Top Gun MediaBoard: Yatin Chawathe, Steve Fink • MASH/TACC integration: Yatin Chawathe • RTSP Archive Server:Angela Schuett, Suchi Raman • Cha-Cha: Mike Chen, Jason Hong, Prof. Marti Hearst • Annotate the Web: Marcel Kornacker, Ray Gilstrap, Tom Phelps, Prof. Robert Wilensky • Secure upload: David Wagner