Sri Harsha Vardhan Reddy Chevuru

Student-Id: 010277104

Web Server Software Architectures

Daniel A. Menascé , George Mason University

menasce@cs.gmu.edu

One of the important features affecting the website’s scalability is the Web server software architecture. Software architecture schemes are broadly characterized in two dimensions: the processing model and the pool size behavior.

Processing models generally are generally characterized as

• Process-based model: It consists of multiple single-threaded processes each handling one request at a time. It is stable i.e. crash of one process doesn’t affect the other. In the event of crash a process needs to be killed and restarted which leads to excess address space usage and also the context switching overhead in case of high volume websites.
• Thread-based model: It consists of single multithreaded server in which each thread handles one request at a time. As new threads doesn’t need additional address space, spawning threads is much more efficient than forking the process. A single malfunctioning thread can bring entire web server down.
• Hybrid model: In this architecture, p processes exist where each process handles n threads each. Incase of a thread failure the process of that thread comes down whereas (p-1)*n threads perform usual functions. This model overcomes the disadvantages of above processes.

Pool-Size behavior is again characterized as static and dynamic.

• In static pooling, the fixed number of processes or threads is created during the startup time. As the arrival rate of requests increases the queue length also increases and if the server load is low then the processes or threads will be idle.

• In dynamic spooling, the number of processes and threads varies as a result the pool size is also varying in order to process more or less number of requests depending on the server load. In order to reduce the amount of lost or unusable memory that accumulates through memory leaks a process’s lifetime is limited.

Performed of a Web server can be monitored by using parameters like

• Service time: Total time which a request spends in a physical resource level.
• Physical contention: Total time a request spends waiting to use web server’s resources.
• Software contention: Total time a request spends in queue waiting for a thread or process to free to handle the request.

In order to evaluate the architecture’s performance characteristics, the results of a combined Markov chain model and Queuing Network (QN). The QN model computes the web server’s throughput and the Markhov chain model is used to represent the software architecture.

The average number of idle processes is directly a function of the average arrival rate of the requests.

Finally, it is said that software contention and architectures plays an important roles in affecting the Web server’s performance.

Strengths:

• The hybrid architectural model discussed is a very practical architecture that has real time applications.
• Importance given to software contention in deciding web server’s performance.
• Memory leaks are also stressed and looked upon.

Weakness:

• The steps in order to minimize the software contention (which is also one of the main features affecting Web server performance) are not discussed.
• Nothing is discussed about the restoration of activities performed by the thread that crashes in the Hybrid model.
• In dynamic spooling, varying the number of processes is difficult certain times.
• The context switching between processes in hybrid model is still not negligible.

Critical Questions:

• What is the installation cost of a hybrid architecture mentioned above?
• What if the parent process crashes in dynamic spooling?
• Does only software contention affects the web server’s performance?