Pygmy.com is celebrating it's one month anniversary since launching it's online bookstore. During Spring Break, it is offering every tenth customer a 10% discount coupon for a book of their choice.
Also, since they expect their sale to be a wide success, they have added capacity to their store by replicating the front-end tier. Any customer can contact any of the k front-end replicas with their requests. The back-end database is not replicated and, like before, each front-end must contact the back-end database server to process requests.
This assignment has two parts. Each part uses a different set of techniques to solve the problem of issuing discount for every tenth customer.
You can be creative as to when elections are triggered. For example, the leader can resign at any time and another time server will need to be chosen.
Part I. Leader election and clock syncronization
Assume k front-end replicas and one backend database server. Also assume one client process makes a sequence of requests to each front-end server. These k+1 processes run a leader election algorithm to elect a time server. You can use any leader election algorithm discussed in class for this purpose. Once a leader has been elected, this node also acts as a time server. Implement the Berkeley clock syncronization algorithm to syncromnize their clocks. Each node then maintains a clock-offset variable that is the amount of time by which the clock must be adjusted, as per the Berkeley algorithm. We will not actually adjust the system clock by this value. Rather to timestamp any request, we simply read the current system time, add this offset to the time, and use this adjusted time to timestamp request.Every incoming request at each front-end replica is timestamped with the syncronized clock value; this time stamp is relayed to the database server for buy requests. The time stamps are used to determine an ordering of buy requests and every 10th buy request is automatically given a 10% discount (by charging a lower amount for the requested book.
Part II. Logical/ vector clocks
Next, we will solve the same problem using logical/vector clocks. In this version, we assume that clocks are not syncronized and yet we wish to determine the ordering of events (here events refer to requests made by clients).To do so, each client process and each front-end server maintain logical clocks. You can either use Lamports clocks or vector clocks for this purpose. Pick either the totally ordered multicast or causally ordered mutlicast algorithm to determine an ordering of all requests made by the clients to the Pygmy.com front-end servers. Note that in this case, requests must be multicast to all processes, in addition to being sent to the front-end server that will process the request. You are free to design the system with either algorithm.
Use the logical clock values to determine the odering of buy requests. Again, every 10th buy request gets a 10% discount on the book being purchased.
A pictorial representation of the system is shown in the figure below:
