Blog

Use a linear approximation of the function at to approximate the value of .

Potpourri-2c Potpourri (2c) [2 points] (Answer True/False with justification)  Tuning the overshoot parameter associated with firm timers to zero is equivalent to using soft timers.

Potpourri-1 Potpourri 1. [2 points] (Answer True/False with justification) Your friend is submitting a map reduce job. Her input data is split into 100 shards. Ignoring set up time, the actual time for the map function to execute on an input shard is T. To speed up the map phase, she asks for 200 nodes. With this allocation, the map phase on the entire input will complete in T/2 time units. 

Internet_Scale_Computing_3a Map Reduce Consider the following implementation of a MapReduce Application. It operates on a cluster of server nodes with the following execution model:  Each worker thread executes its assigned map tasks sequentially (one map task at a time)  Intermediate data from each map task is stored on the worker’s local disk  Data transfer occurs for reducers to collect the intermediate data from the mapper tasks   No network cost for accessing data on the same server node  Network transfer cost applies only between different server nodes  All inter-server-node data transfers can occur in parallel  The reduce phase starts only after all the intermediate data from all the map tasks have been transferred to the nodes.  Each worker thread executes its assigned reduce tasks sequentially (one reduce task at a time)  Specifications of the MapReduce Application to be run:  Input data: 150GB split into 50 shards of 3GB each.  Number of map tasks: 50 (one per shard).  Number of reduce tasks: 15 (the desired number of outputs from the Map-Reduce Application).  Each map task produces 300MB of intermediate data.  Each reduce task gets equal of amount of intermediate data from each of the map tasks to process for generating the final output.  Simplifying assumptions:  Ignore local disk I/O time  All network paths between server nodes have same bandwidth.   Parallel network transfers don’t affect each other (no bandwidth contention).  All data transfers occur ONLY after ALL the map tasks have completed execution  Perfect load balancing (work distributed evenly to all reduce tasks)  All server nodes have identical performance  Assume 1000MB=1GB (instead of 1024MB) for ease of calculations.  All nodes mentioned in the configuration below are workers and mappers/reducers can be scheduled on them. You can assume a separate node for master which is in addition to what is stated.  You should ignore time spent by master for doing the orchestration. You should ignore the time taken to shard and time taken to send shards to nodes running map tasks. You should ignore the communication time for anything except file transfer.  Calculate the execution time for the map phase and on the following configuration:   5 server nodes  Processing speed: 1 minute per GB (for either map or reduce task)  Assume that the data needed for processing is already present on the nodes.

Real_time_and_Multimedia_3 PTS Your friend is a geologist who studies earthquakes and tsunamis. They have placed 1000 seismograph sensors all around the world. Each sensor produces a 1 kilobyte datapoint every second. They have also placed 50 cameras in coastal areas around the world that record at 60 frames per second. Each video frame is 1 MB.  Your friend wants to record this data and analyze it for two purposes:  To send urgent alerts when multiple sensors in a region report strong shaking in the same brief time window. To correlate timestamped video recordings with timestamped seismograph data in order to study what types of shaking cause tsunamis. Is PTS a good basis to build a system for analyzing this data? Justify your answer. 

Internet_Scale_Computing_4 CDN In principle, the Coral system’s key-based routing serves the same fundamental purpose as traditional DHT routing: enabling a source node to locate and communicate with the appropriate destination node. Given this, what specific problem does Coral’s key-based routing address that traditional DHT routing does not? 

Internet_Scale_Computing_1a Giant Scale Services You are deploying a large-scale machine learning model for inference in a cloud data center. The model is 960 GB in size and can be broken down into 8 GB chunks that must be executed in a pipelined manner. Each chunk takes 0.8 ms to process. The available machines each have 8 GB of RAM. You are required to serve 600,000 queries per second. Assume there is perfect compute and communication overlap, and no additional intermediate memory usage during execution. What is the minimum number of machines required to support this throughput? You are free to assume pipelined execution of chunks for this.

Consider the Modigliani and Miller (M&M) tradeoff theory of capital structure. Assume there are taxes and bankruptcy costs. Which of the following statements is / are correct? I. Firm value always increases as more debt is addedII. Firm value stays constant as more debt is addedIII. WACC always decreases as more debt is addedIV. WACC always increases as more debt is addedV. WACC stays constant as more debt is added

Los Pollos Hermanos is considering Projects S and L, whose cash flows are shown below. These projects are mutually exclusive, equally risky, and not repeatable.  Your boss, Gus Fring, asks you which project will have the higher NPV. What do you respond?   Year                           0                1                2                3                4     CFS                        -$1,100        $600          $500          $300         $100 CFL                        -$1,100        $100          $300          $500         $600

Simon Software Co. is trying to estimate its optimal capital structure.  Right now, Simon has a capital structure that consists of 0% debt and 100% equity, based on market values.  The risk-free rate is 6% and the market risk premium, RM – Rrf, is 5%.  Currently the company’s cost of equity, which is based on the CAPM, is 12% and its tax rate is 40%.  What would be Simon’s estimated cost of equity if it were to change its capital structure to 50% debt and 50%