A Trace Detection Method Based On Sequential Pattern Mining

Server-side programs running on the network environment,users request to the server are usually generated from the dynamic,the server response interpreted programming language of the user requests are dynamically generated,cannot compile optimization in advance,can only be run to explain the way of the execution,such as Python,JavaScript,NodeJs,etc.In order to reduce the program execution time,improve the server response speed,JIT compilation technology can be used in the process of program execution to dynamically discover the program execution frequency higher code fragments,and to compile the code snippet optimization,thereby improving the efficiency of the program execution.The JIT compilation technology can be divided into two classes,which are compiled using methods as compilation units and trace as compilation units,depending on the compilation granularity.Method based JIT compilation,the whole method is compiled,even though the program fragment that isn't frequently executed by the method needs to be compiled and has a large granularity.Trace based JIT compilation,compiles only code snippets that are frequently executed,with a smaller granularity.Contrast method based JIT compilation technology,trace based JIT compilation technology because of the identification code of the hot spot of high precision,small size,can undertake a more detailed program optimization,and it can reduce the overhead caused by the program with less execution times in the compilation method.The core issue of trace based JIT compilation technology is how to identify a frequent hot trace.Traditional trace detection strategies such as the method based on counting,to identify the first anchor point in the program(trace starting basic block),and then record the subsequent implementation of basic block as a trace,statistics the number of each trace is executed,exceed the threshold trace is identified as hot trace.However,it is very time consuming to detect the anchor points based on the counting method of trace detection.At the same time,the traditional trace detection strategy is executed on the single execution of the program,ignoring the characteristics of concurrent execution of the server-side program.Sequential pattern mining is an important study of data mining.When the program interprets execution,the basic block in sequence can be considered as sequential data.A server-side program that is executed concurrently can be seen as a number of basic block sequences,the basic block sequence database.The paper puts forward a server-side program of trace detection on the basis of sequential pattern mining method,in view of the basic block sequence database using sequence pattern mining algorithm,identify the hot trace in the server-side program.This method can make use of the characteristics of concurrent execution of server-side programs,and more efficient identification of trace,so as to improve the execution efficiency of the program and improve the user response speed.The main work of this study includes receiving the basic block of the interpreter and generating the basic block sequence database.In this paper,a sequence pattern mining algorithm named Pisat is proposed for the basic block sequence,and the hot trace in the basic block sequence is identified by the Pisat algorithm.At the same time,the recognition of repeated trace is recombined to reduce compilation overhead.Compared with the traditional trace detection strategy,the method in this study does not need to define anchor points and is more suitable for server-side programs.The experimental results show that the trace detection method has high availability and effectively improves trace detection efficiency.