Midterm Examination

CSci 555 / Neuman-Obraczka Fall 1997
Directions for completing exam

Answer the following two questions:

1. (50 points) The ever increasing computational needs of the National Weather Service became even more apparent after several recent weather phenomena such as the El Nino. You have been hired to re-design their computing architecture, which includes:
   • A wide-area, distributed weather information gathering system that collects satellite weather information.
   • A parallel weather forecasting system that runs weather simulation tools using gathered weather information. Machines used in this parallel computation system are connected through a high-speed Local Area Network.
   • A wide-area, distributed weather information dissemination system that provides weather forecast information for public access.
   • A wide-area, distributed weather information dissemination system tailored to be used by collaborating meteorological researchers around the world. Researchers use the data to feed their meteorological models and publish their results for other researchers to use and comment on.

   1. (20 points) Given the communication paradigms covered in class, discuss the tradeoffs of using each one of them to implement each of the above services? Justify your answer.
   2. (20 points) All the distributed systems above share data. However, the data sharing patterns vary from one system to another. Describe the data access pattern corresponding to each system. What kind of synchronization/concurrency control techniques would you use for each system to ensure data consistency? Justify your answer.
   3. (10 points) If you use replication to improve the performance of the systems above, what kind of replication approach would you choose for each system? Why?

2. (50 points) WorldCom, CompuServe, and America Online (AOL) recently announced their intent to re-structure the online services business of all parties so that AOL will acquire the online content business of CompuServe and get out of the network access part of this business, selling it to WorldCom, who will also pick up the network access business of CompuServe.

   With this change we will see significant consolidation of both the network access business in WorldCom and the online content services in AOL. The size of the customer base will make the scalability and reliability of the new system even more critical than it is today.

   You have been hired by AOL to help them design a new content and user services architecture that will better meet the scalability and reliability requirements of the restructured companies. For the purpose of this exam question, assume that today (though not necessarily in the system you will design) all AOL user services including content retrieval, login, and e-mail are provided by servers at AOL's headquarters in Virginia. Their network access service (which will be sold to WorldCom) is distributed with network access points in all major US cities, and CompuServe's network access services are similarly distributed across the US and Europe.

   1. (20 points) User names in AOL are drawn from a flat global name space. Why do you think the name space is structured in this way, and what implications does this have for scalability? How would you structure the name space for e-mail addresses and user names in the combined online service? Why? How would users find the e-mail address of other users with whom they wish to communicate, and how does this relate to naming. How might they name those users once they have found them? Above all, when answering this question, be practical, and don't over design. For the approaches to naming other than the approach you choose, explain why you didn't choose it.
   2. (15 points) Where would you place AOL servers, and why? How does your placement of servers affect manageability, reliability, and performance? What kinds of data will be managed by these servers, and for each kind of data, will it be be distributed, replicated, or both? Based on what characteristics will you allocate data to particular servers? Where else might data be stored in the system besides on these servers, and under what conditions? How does your placement of data affect manageability, reliability, and performance?
   3. (15 points) How will you deal with security in the system? When will authentication be required, and how will it be accomplished? How will the structure of the system you designed in parts (a) and (b) affect the security of the system and the points at which authentication will be necessary. What user operations will require authentication and authorization and what kinds of queries will require confidentiality and/or integrity of the communications channel? Why? Again, be practical here and balance the cost of such protections against the actual benefit.

• About this document ...