1.4 The Advantages of Servlets Over "Traditional" CGI
Java servlets are more efficient, easier to use, more powerful, more portable, safer, and cheaper than traditional CGI and many alternative CGI-like technologies.
With traditional CGI, a new process is started for each HTTP request. If the CGI program itself is relatively short, the overhead of starting the process can dominate the execution time. With servlets, the Java virtual machine stays running and handles each request with a lightweight Java thread, not a heavyweight operating system process. Similarly, in traditional CGI, if there are N requests to the same CGI program, the code for the CGI program is loaded into memory N times. With servlets, however, there would be N threads, but only a single copy of the servlet class would be loaded. This approach reduces server memory requirements and saves time by instantiating fewer objects. Finally, when a CGI program finishes handling a request, the program terminates. This approach makes it difficult to cache computations , keep database connections open , and perform other optimizations that rely on persistent data. Servlets, however, remain in memory even after they complete a response, so it is straightforward to store arbitrarily complex data between client requests.
Servlets have an extensive infrastructure for automatically parsing and decoding HTML form data, reading and setting HTTP headers, handling cookies, tracking sessions, and many other such high-level utilities. In CGI, you have to do much of this yourself. Besides, if you already know the Java programming language, why learn Perl too? You're already convinced that Java technology makes for more reliable and reusable code than does Visual Basic, VBScript, or C++. Why go back to those languages for server-side programming?
Servlets support several capabilities that are difficult or impossible to accomplish with regular CGI. Servlets can talk directly to the Web server, whereas regular CGI programs cannot, at least not without using a server-specific API. Communicating with the Web server makes it easier to translate relative URLs into concrete path names , for instance. Multiple servlets can also share data, making it easy to implement database connection pooling and similar resource-sharing optimizations. Servlets can also maintain information from request to request, simplifying techniques like session tracking and caching of previous computations.
Servlets are written in the Java programming language and follow a standard API. Servlets are supported directly or by a plugin on virtually every major Web server. Consequently, servlets written for, say, Macromedia JRun can run virtually unchanged on Apache Tomcat, Microsoft Internet Information Server (with a separate plugin), IBM WebSphere, iPlanet Enterprise Server, Oracle9i AS, or StarNine WebStar. They are part of the Java 2 Platform, Enterprise Edition (J2EE; see http://java.sun.com/j2ee/), so industry support for servlets is becoming even more pervasive.
A number of free or very inexpensive Web servers are good for development use or deployment of low- or medium-volume Web sites. Thus, with servlets and JSP you can start with a free or inexpensive server and migrate to more expensive servers with high-performance capabilities or advanced administration utilities only after your project meets initial success. This is in contrast to many of the other CGI alternatives, which require a significant initial investment for the purchase of a proprietary package.
Price and portability are somewhat connected. For example, Marty tries to keep track of the countries of readers that send him questions by email. India was near the top of the list, probably #2 behind the U.S. Marty also taught one of his JSP and servlet training courses (see http://courses.coreservlets.com/) in Manila, and there was great interest in servlet and JSP technology there.
Now, why are India and the Philippines both so interested? We surmise that the answer is twofold. First, both countries have large pools of well- educated software developers. Second, both countries have (or had, at that time) highly unfavorable currency exchange rates against the U.S. dollar. So, buying a special-purpose Web server from a U.S. company consumed a large part of early project funds.
But, with servlets and JSP, they could start with a free server: Apache Tomcat (either standalone, embedded in the regular Apache Web server, or embedded in Microsoft IIS). Once the project starts to become successful, they could move to a server like Caucho Resin that had higher performance and easier administration but that is not free. But none of their servlets or JSP pages have to be rewritten. If their project becomes even larger, they might want to move to a distributed (clustered) environment. No problem: they could move to Macromedia JRun Professional, which supports distributed applications (Web farms). Again, none of their servlets or JSP pages have to be rewritten. If the project becomes quite large and complex, they might want to use Enterprise JavaBeans (EJB) to encapsulate their business logic. So, they might switch to BEA WebLogic or Oracle9i AS. Again, none of their servlets or JSP pages have to be rewritten. Finally, if their project becomes even bigger, they might move it off of their Linux box and onto an IBM mainframe running IBM WebSphere. But once again, none of their servlets or JSP pages have to be rewritten.
One of the main sources of vulnerabilities in traditional CGI stems from the fact that the programs are often executed by general-purpose operating system shells . So, the CGI programmer must be careful to filter out characters such as backquotes and semicolons that are treated specially by the shell. Implementing this precaution is harder than one might think, and weaknesses stemming from this problem are constantly being uncovered in widely used CGI libraries.
A second source of problems is the fact that some CGI programs are processed by languages that do not automatically check array or string bounds. For example, in C and C++ it is perfectly legal to allocate a 100-element array and then write into the 999th "element," which is really some random part of program memory. So, programmers who forget to perform this check open up their system to deliberate or accidental buffer overflow attacks.
Servlets suffer from neither of these problems. Even if a servlet executes a system call (e.g., with Runtime. exec or JNI) to invoke a program on the local operating system, it does not use a shell to do so. And, of course, array bounds checking and other memory protection features are a central part of the Java programming language.
There are a lot of good technologies out there. But if vendors don't support them and developers don't know how to use them, what good are they? Servlet and JSP technology is supported by servers from Apache, Oracle, IBM, Sybase, BEA, Macromedia, Caucho, Sun/iPlanet, New Atlanta, ATG, Fujitsu, Lutris, Silverstream, the World Wide Web Consortium (W3C), and many others. Several low-cost plugins add support to Microsoft IIS and Zeus as well. They run on Windows, Unix/Linux, MacOS, VMS, and IBM mainframe operating systems. They are the single most popular application of the Java programming language. They are arguably the most popular choice for developing medium to large Web applications. They are used by the airline industry (most United Airlines and Delta Airlines Web sites), e-commerce (ofoto.com), online banking (First USA Bank, Banco Popular de Puerto Rico), Web search engines/portals (excite.com), large financial sites (American Century Investments), and hundreds of other sites that you visit every day.
Of course, popularity alone is no proof of good technology. Numerous counter-examples abound. But our point is that you are not experimenting with a new and unproven technology when you work with server-side Java.