Suresh Arena

Delivering software with quality software design is always a challenge for a software engineer and more so for those with less experience. A good software design is expected to deliver software that is simple, extensible and maintainable. What are the factors that influence quality in software design?

Here are the ones that I can think of:

1. Understanding of business requirements - this includes a solid understanding of current business requirements and a good understanding of business/technology vision
2. Design skills - this includes the quantity of design skills that is used for the design activity - number of designers, skill-levels of designers etc.
3. Extent of validation - this includes frequent peer reviews, formal and informal design reviews etc.
4. Time spent on design - this includes the total/aggregated time spent on design

Also, some of the cross-cutting aspects that could influence any of these factors could be:

1. Motivation - the motivation of all those involved in the design activity and also the stakeholders.
2. Organizational support - the organizational support for the design activity, design decisions etc.

If we assume that there is a solid understanding of business requirements and a good knowledge of business/technology vision (if one exists), then the software design quality is directly proportional to the other three factors: design skills, extent of validation and time spent on design.

In the above diagram, the area of the triangle corresponds to the quality of software design, the bigger the area, the greater the design quality. If we assume that we could assign x number of designers to solve a problem, then we have three options to improve the design quality - improve the design skills or increase the validation mechanisms or increase the time spent.

In reality, a team cannot afford to do any of these three beyond certain levels because of the associated costs. For instance, one cannot assign too much time or too many skilled designers for budgetary reasons, customer commitments etc. As a result, improving the design quality entails a delicate balancing act between these various factors within the available tolerance bands.

This diagram may lead to other questions like how is ‘time spent on design’ related to minimalistic design - which essentially means distributing the design activities among different iterations and not doing the big design upfront? Or in other words, how does this deal with agile development paradigm? I have a week to think for my next blog.

It is everybody’s knowledge that XML is a chief enabler of SOA and semantic integration. But one of the major technology roadblocks for the adoption of XML is performance - XML processing is resource intensive and time-consuming which makes one think that software-based XML processing may not be the best way of dealing with XML. This has been a major challenge for ESB vendors who deliver xml-based integration solutions.

Yesterday I came across this company called Terari that claims to have the world’s first silicon-based XSLT engine. Terari claims that its silicon-based RAX-XSLT is five times faster than a software XSLT engine. For the last couple of months, there has been a few developments in the hardware/networking area that is related to XML processing: CISCO’s AON strategy, IBM’s acquisition of Datapower and Intel’s foray into XML routing. These vendors have XML appliances that could do things like message-based routing, content acceleration, and XML security.

All these developments seem to have direct impact on SOA and specifically ESB. Are we getting closer to a silicon-based ESB?

A XML-based ESB needs a xml router, xml transformation engine, standards-based adapters that would let service providers connect the services to the bus and optionally an orchestration mechanism. It becomes obvious that hardware implementations for the first two are already available or about to be available. All that needs to be done - some vendor may be already doing this in a top-secret way - is to develop one or more ASICs that would

If somebody comes up with such a solution, then we will see a highly-performant ESB solution that would take the entire middleware market by storm as it may be the next big thing and as this article claims, the next big innovation in enterprise computing.

On my flight from Chicago to San Francisco, I was reading an article in Information Week on SOA and web services which had a survey on SOA/web services. According to this survey, 26% of the respondents said SOA/web services introduced more complexity into IT system. 33% of the respondents said SOA/web services cost more than expected. These problems could be due to excessive service-enablement without proper business process analysis.

Also, the article talked about a few adopters of SOA and their experiences. As it did so, it sought to answer some questions.

Who should identify services? When IT identifies and builds services, it does so from components that are convenient to put together. Whereas the grander vision of SOA is automating the right business processes to fulfill business needs. So, IT should work with business-process analysts to identify services.

How do we get SOA into business process analysis? Highmark, the health insurance company, is considering a role called “SOA business-process analyst”. Creation of such roles may address one of the biggest challenges in SOA adoption - getting the business analysts to think about reusable services. By getting SOA into the titles of some business analysts, identifying services is made an important part of job responsibility. Eventually when there are enough business analysts thinking in terms of services, this role may go away.

I was talking to this person who was working in IT architecture. I asked him “Do you use SOA in your company?” He said, “Oh yes. we have web services”. Alright I have heard that before. But here is the highlight. Yesterday, I was listening to this webcast from a respected web services security expert. Somebody asked him at the end of the show, “what is SOA?”. He says, “It is a means by which you can access your applications from remote processes and different systems”. I was stunned.

SOA has been a buzzword in the IT field for quite a while and still there is a lot of IT experts who cannot distinguish between SOA and web services. SOA is still a long way to go..

If anything, SOA and service-enablement should start with business analysis. SOA without business analysis may lead to cost overruns and increased complexity in terms of service management. On the other hand, a good business analysis will become a selling point for SOA.

Last week I was reading an article when I ran into JSR 227. JSR 227 was introduced by Oracle a couple of years ago. It seeks to introduce a declarative binding and loose coupling between the application/business services and the view layer in J2EE applications. The motive of this initiative to standardize a drag-drop feature of the Application Deverloper Framework in Oracle JDeveloper. This JSR mandates a standardized way to describe J2EE-based application services and a standardized way of consuming them. BEA and IBM voted against this JSR citing concerns about the extent and scope of this JSR especially when it comes to describing application services in a standard way.

However, this JSR brings out an interesting void in service-enablement where most of the focus has been on enterprise services that are consumed over the web. Loose coupling among fine-grained application services will bring about agility and reuse of services among lower-level business processes. Most of the times, these lower-level business processes are controlled by one or two business units and hence the opportunity to share these services exist without a process-oriented organizational setup that SOA mandates. And demonstrating success at lower levels may make upper-level management realize the value of SOA if implemented for enterprise-wide business processes.

What does all this mean technically?

As a side note, Java-based Spring framework provides a nice way to decouple service clients from services through inversion of control (IOC) and dependency injection. The injection is configurable which is a nice departure from the code-centric decoupling approach promoted by J2EE through business delegates and service locators. Moreover, inversion of control provides for improved testability.

While browing the iTunes podcast directory, I found SOA expert David Linthicum’s podcasts on SOA. I find the podcasts useful to get an up-to-date aggregated information on the developments in the SOA world. David releases a 20-minute podcast every couple of weeks. Apart from giving updates in the SOA field, David answers questions from the audience that he received via email or voice mail. Moreover, he delves deeper into a topic in SOA in every podcast. One can subscribe to his podcasts at:

http://www.soaexpertpodcast.com

A Service is a business function with a published definition specifying the contract for supported operations, exchanged messages, protocol bindings and access information. A service can have several implementations. A well-designed service is expected to possess certain quality attributes like performance, reliability, scalability, autonomy etc. some of which may be guaranteed by the service provider or the service broker.

A service provider provides a service with a published definition. Some companies like Amazon.com, Google provide services in the web that could be programmatically accessed. A service broker is a marketplace hosting services provided by different providers. StrikeIron
is an example of a web service marketplace. A service consumer consumes the service.

How are services related to components in the OOP world? I got the same question and found the answer in one of the web sites that made sense. Service is a deployment and runtime concept. Component is a development and deployment concept. Component is generally a unit of code for compilation and a service is an implementation of an interface that could be realized by one or more components. A service has pre-defined quality attributes and service level agreements. A service has message contracts and published definition. A service may be discoverable.

A reverse proxy is a proxy server that proxies the requests to a group of web servers. It is called a ‘reverse’ proxy because a proxy server typically proxies the requests from a group of clients to the internet while the ‘reverse’ proxy proxies the requests from all the clients to a group of web servers. Most of the common web servers including the popular apache web server can be configured to function as a reverse proxy.

A nice article on reverse proxy can be found at:

http://www.sans.org/rr/whitepapers/webservers/302.php

Can’t believe it? Me neither. Javaone conferences have been used as the stage for Microsoft-bashing by the Java pundits. It will no longer be the case with Microsoft jumping-in as a sponsor of the 2005 edition of the conference. On May 13th, Scott McNealy and Steve Ballmer appeared in a press conference and announced, among other things, the single sign-on feature for applications that involve both Solaris and windows platforms. Furthermore, Sun gets to access Microsoft’s Remote Desktop protocol that will boost their Sun Ray offering. More details about the announcement can be found here:

http://news.yahoo.com/news?tmpl=story&u=/sv/20050515/tc_siliconvalley/_www11646916_1

This is the second positive development for Sun Ray in the same week - the acquisition of Tarantella earlier last week gives Sun the ability to enable the Sun Ray appliances to access Microsoft products.

One of my colleagues pointed me to Google suggest. As you type the keyword, you get suggestions with the number of matches. Looks impressive - it asks the question ‘why not client-side javascript to do serious things like firing a query to server and maintaining the result cache?’ quite convincingly. I am not going to dissect it as someone else has already done it.

http://serversideguy.blogspot.com/2004/12/google-suggest-dissected.html

Have you ever come across the storage jargons like SAN, NAS etc. and wondering what they are? I was wondering myself and I sought refuge in a book.

I read a book that had introductory concepts on storage technologies. The book addressed the evolution of technologies for connecting the storage media to the computer. The first step in the evolution was server attached storage where hard disks were directly attached to computers. SCSI (Small Computer Systems Interface) is a protocol (the most popular) for connecting storage to computers directly. SCSI bus is the circuit that connects the devices - computers and disks. Computers connected to a SCSI bus are the initiators since they initiate the request for data and disks are the targets. SCSI comes in different flavors - narrow, wide and ultra - single and differential.

SCSI-1 was the first in the scene and it supported a data speed of 3-5 MB/sec and only one disk. Then came SCSI2 with all the optimizations to handle data upto 20 MB/sec. SCSI-3 is called ultra SCSI. Ultra-3 SCSI is the latest ultra SCSI that can support data speeds upto 160 MB/sec. FCAL (Fiber channel arbitrated loop) uses fiber channels and encapsulated SCSI as the protocol. FCAL can support data speeds in GB/sec.

Since the enterprise data need has grown into terrabytes, technologists have started looking beyond the server attached storage into technolgies like SAN and NAS (they are not the same things written in reverse..). SAN is Storage Area Network and NAS is Network Attached Storage. SAN is a network of disk arrays connected to one another using fiber. The SAN is then connected to the servers using fiber. NAS is a network appliance that has scaled-down OS with disk arrays attached. The servers access NAS appliance (also called filers as they serve files as opposed to data) through technologies like NFS.

SAN and NAS help us centralize the management of storage like backup, mirroring/failover and administration in general. However according to Gartner and other industrial analysts, only about 10% of the storage market is on NAS, SAN etc. Still 90% of the storage is on the server attached storage like the hard disks.

Caught between SAN and NAS? Here is a good link on how to decide between SAN and NAS.

http://www.onlamp.com/pub/a/onlamp/2002/03/14/sansnas.html

Also, O’reilly has a book on SAN and NAS.