130 likes | 241 Views
Prototype Presentation Replication Process Simulator. Academic Advisor: Dr. Eitan Bachmet Technical Advisor: Mr . Assaf Natanzon. Project Team : Adiel Ashrov Etai Hazan Benny Michali. http://replicationsimulation.wordpress.com/. Prototype Presentation Storage Replication Simulation.
E N D
Prototype PresentationReplication Process Simulator Academic Advisor: Dr. EitanBachmet Technical Advisor: Mr. AssafNatanzon Project Team: AdielAshrov EtaiHazan Benny Michali http://replicationsimulation.wordpress.com/
Prototype PresentationStorage Replication Simulation Prototype Goals Conclusions Hands On Prototype Demonstration
Prototype – Goals • Development of the prototype was designed to overcome the risks we discovered with our requirements. • We identified the following points as risks : • How can we make sure simulator is modeling the system accurately? • Making the system perform similarly to the real-life. • Statistics handling and transfer between different layers. • How to represent calculated statistics in the Presentation level • How to model I/O behavior and how to recognize behavior • How to operate on different algorithms/policies?
Prototype – Conclusions Simulation results verification: • In order to ensure our simulation produces accurate statistics we have sought for an expert opinion • AssafNatanzon (EMC) inspected the results and approved our simulation results are coherent with real world behavior. Statistical data Handling • We’ve experimented with the idea of using CSV files in order to pipeline the results from the application to the presentation layer.
Prototype – Conclusions Data plotting and GUI • We have explored several external libraries solutions for plotting the data and displaying GUI. • We are now writing the GUI in C# and plot the Graph using a sample graph template we will embed as a part of the UI. I/O behavior modeling and recognition • We model the bursts with a random choice of time slots, each burst will be set randomly on time slot i, so all I\O operations scheduled before slot i will be sent at slot time i. • We use Entropy in order to recognize each host’s level of I\O activity.
Prototype – Conclusions Different Flow Control Alog’ Support • Flow Control algorithm: • We thought of and wrote a small flow control algorithm which can regulate the rate in which the cache is increasing capacity and the bandwidth utilization. • Dynamic Algorithm selection: • Switching a flow control algorithm dynamically will destroy the results of a simulation. • we decided to neglect this idea. • The flow control algo’ is chosen before the simulation starts(a given parameter) and doesn’t change.
Prototype – hands on I/O behavior modeling and recognition • We have modeled the burstinessbehavior with a parameter received from the user [0..100]. • Time slots concepts: • Say for example burstiness is 40:
Prototype – hands on I/O behavior modeling and recognition cont. • How did we identify a volume/host in a burst: • Entropy and time slot data capacity: • Let’s look at time slot and the amount of data transferred: • . • This is the probability that an I/O event happened in .
Prototype – hands on I/O behavior modeling and recognition cont. • Entropy measure of uncertainty • ) • The lower the entropy ,it is more likely that the current Volume is in a burst.
Prototype – hands on Different Flow Control Alog’ Support cont. • The main purpose of our project is allowing the user to compare between different flow control algo’. • After we have the ability to identify trends/burst in I/O behavior we now have the ability to regulate the flow of data and cache allocation. • Let’s see an example of such algorithm
Prototype – hands on Different Flow Control Alog’ Support cont.
Prototype – hands on • wait(((X-80)/20) * Derivative);//Version 2 Flow Control Algo’ examples: when(received(NEW_IO){ If (X < 80){//Threshold is 80 send(TransferReady); } else{ wait(((X-80)/20) * LatencyParam); send(TransferReady); } }