In Vivek Pai’s paper on the Flash webserver, the authors eva…
In Vivek Pai’s paper on the Flash webserver, the authors evaluated several webserver designs using two traces – the Owlnet trace, whose dataset size was 100MB, and the CS trace, whose dataset size was 7.8GB. The webservers they compared include Flash (which followed the AMPED model), SPED (the Zeus single-process event-driven server), MT (a multi-threaded webserver), MP (a multi-process webserver, Apache (an unoptimized Apache webserver). The tests were performed using a 333MHz PentiumII CPU with 128 MB of memory and multiple 100 Mbit/s Ethernet interfaces. Give the correct answer to the following questions:1. In the event of the CS trace, which of the following will perform better: a. Flash vs. MT b. Flash vs. MP c. Flash vs. SPED d. SPED vs. MT e. MT vs. MP 2. In the event of the Owlnet trace, which of the following will perform better: f. Flash vs. MT g. Flash vs. MP h. SPED vs. MT i. MT vs. MP j. MP vs. Apache 3. Without modifying the implementation described in the paper, when running on a more recent quad-core platform with 12GB of memory, which one would you expect to perform better: k. Flash vs. MT l. MT vs. MP
Read DetailsIn a file server, the following steps have to be performed o…
In a file server, the following steps have to be performed on each request: (1) parse request, (2) find and read file, (3) respond with file or error message. Assume steps (1) and (3) take 1ms and step (2) takes 2ms. Assume any delays due to inter-thread communication, synchronization, context-switching, etc. are negligible. Assume you have infinite resources. Consider a boss-worker and a pipelined model designs, each with a total of 4 threads. In both cases, you need to process a total of 6 requests. Answer the questions below. Please try to follow these rules:- Do not write the units. – Do not use a comma in your answer (e.g., write 1000 instead of 1,000 or write 0.75 instead of 0,75). – Write the leading zero in decimal fractions (e.g., write 0.75 instead of .75)- Show the results with up to two decimal places. Questions:For the boss-worker model, where the boss simply accepts requests and passes them onto a shared workers’ queue: a) How many requests can be processed concurrently? b) How long does it take to process a total of 6 requests in ms? ms c) What is the average throughput for this workload, in req/ms? req/ms d) What is the average response time for this workload, in ms? ms For a balanced implementation of the pipeline model: e) How many threads are allocated to the first stage? f) How many threads are allocated to the second stage? g) How many requests are concurrently being processed? h) How long does it take to process a total of 6 requests in ms? ms i) What is the average throughput for this workload, req/ms? req/ms j) What is the average response time for this workload, in ms? ms
Read Details