Software modularization should not be considered in isolation, but rather seen within the context of several inter-related trends that have occurred over the last decade. Indeed, software modularity is merely the latest visible facet of a much larger and more fundamental technology shift, with an impact at least equivalent to the move from mainframes to client server computing in the 1980′s.
These related trends include:
- Service Oriented Architecture (SOA) – Enabling previously rigidly coupled business systems with proprietary protocols to be expressed as “services” which can be accessed via common protocols. However the business systems themselves remained opaque and monolithic.
- Cloud Computing - Decoupling applications from the underlying compute resources upon which they run, allowing more efficient resource utilization and scaling.
- Software Modularization – Most recently, replacing the opaque monolithic applications and stove-piped middleware with dynamically assembled alternatives composed from re-usable software components.
As commented on in “The Rise of the Stackless Stack” (see http://www.redmonk.com/jgovernor/2008/02/05/osgi-and-the-rise-of-the-stackless-stack-just-in-time/); these trends collectively shift the industry away from rigidly coupled, static, opaque environments; towards adaptive, loosely coupled systems which are dynamically assembled from well-defined software components that run across a fluid set of compute resource.
Modularity and Assembly – It’s not a new idea!
The concept of assembling a product from a set of well-defined re-usable components is not new. Indeed its roots can be traced back to at least 250BC with emperor Qin Shi Huang and his commissioning of the Terracotta Army (see http://en.wikipedia.org/wiki/Assembly_line). Whatever the product, the driver for modularity and subsequent assembly is to; increase and maintain quality, reduce cost and increase output. The modern archetype for modularity and assembly is the automotive industry, where extensive use of standardization, modularity and assembly results in affordable automobiles.
Likewise, the computer industry already extensively uses hardware modularization in the form of standardised CPUs, memory and disk subsystems; leading to affordable computer hardware. These concepts are also well understood by software engineers. Extracts from http://en.wikipedia.org/wiki/Programming_in_the_large_and_programming_in_the_small include:
“Small programs are typified by being physically small in terms of their source code size, are easy to specify, quick to code and typically perform one task or a few very closely related tasks very well”,
“… programming in the large, coding managers place emphasis on partitioning work into modules with precisely-specified interactions…”
“… one goal of programming in the large involves setting up modules that will not need altering in the event of probable changes. This is achieved by designing modules so they have high cohesion and loose coupling.”
Coarse grained modularization of business processes (via SOA) and the subsequent re-assembly (via BPEL) has been underway for some time, however the individual applications have remained monolithic and opaque. Further progress was not possible until a strong, widely endorsed, industry standard for enterprise software modularity was available. OSGiTM provides this modularization standard.
Since 1999 the OSGi Alliance (see http://www.osgi.org) has provided standards, reference implementations and guidance for modularization and assembly best practices. Recently, the OSGi Alliance has succeeded in:
- Recruiting the majority of enterprise software vendors.
- Encouraging those vendors to migrate their product portfolios to OSGi.
- Producing the OSGi Enterprise standard which provides mechanisms to integrate OSGi based applications with legacy JEE & Spring.
Today OSGi is rapidly becoming a cornerstone for any successful business system re‑engineering effort.
While software modularity and dynamic application assembly are inevitable industry trends; OSGi faces the usual adoption challenges: see Innovators Dilemma for an exploration of this theme – http://www.businessweek.com/chapter.christensen.htm.
‘The Business’ will never ask for OSGi based systems, it’s an implementation detail. Yet in the fullness of time ‘The Business’ will complain when the internal IT organization is seen as; too inefficient, too expensive and /or no longer agile enough to meet tactical, let alone strategic, business objectives.
Yet, from an IT management perspective revisiting a complex but working enterprise software stack is an immediate cost and quantifiable risk. Something to be avoided. The alternative, a slow death seems almost preferable, or at least somebody else’s problem. As the environment degrades, make-shift tactical solutions are implemented to ease immediate pain points caused by strategic issues. This ongoing tactical ‘reactive’, behaviour results in an explosion in operational complexity and associated OPEX. The organisation eventually becomes paralysed by its own management’s inability to ‘grasp the nettle’ and identify and address the fundamental issues.
It is suggested that this organizational behavior parallels the well known ‘Tragedy of the Commons’ (see http://en.wikipedia.org/wiki/Tragedy_of_the_commons).
Developer attitudes and/or their working environment may also act as an impediment:
- Just Get It Done. At each point in time, the developer is pursuing, or is forced to pursue, the shortest / lowest effort / route to the immediate deliverable at the expense of longer term maintainability.
- The Artisan: The software developer “knows better” and pursues his own approach in defiance of industry standards and accepted best practices.
- The Luddite: The current opaque, undocumented spaghetti of code within the organisation ensures employment. Increased efficiency, self-describing dependencies and code re-use sound like a recipe for smaller more flexible development teams.
Each can be a powerful brake on change.
Yet history tells us, those who effectively embrace and leverage change, succeed. OSGi migration is neither free, nor in itself, a quick ‘silver bullet’. However OSGi is of fundamental strategic importance to any organisation whose core business involves the processing of information.
Quite simply, modularity localizes the impact of change, which directly leads to increased maintainability.
- As long as the module boundaries don’t change, one can change the functionality of the module freely, without concern for breaking the wider system; i.e. the impact of any local change is prevented from leaking into the wider system.
- Modules that perform a few, or a single function, are much easier to test exhaustively than an entire monolithic system.
- Smaller pieces of the system can be independently versioned.
- Specific knowledge is only required for the particular module being worked upon, along with its relationship to other modules in the system; other modules can be treated as “black boxes” that perform specific functions, without worrying about how they perform them.
The large scale structure of the migrated application, which in all likelihood was previously unknown, is now completely defined by the dependencies between the set of versioned modules. Hence:
- It is understood which modules are and are not required!
- By replacing or re-wiring modules, the behavior of a composite system may be rapidly changed.
From an ongoing maintenance perspective it is now possible to re-factor individual modules, or even the overall system, to systematically drive out accidental complexity (see http://theit.org/publishing/books/prof-app/19261.cfm) and so contain, or even reverse, design rot.
While these arguments are understood, many organizations require demonstrable real world examples. The challenge is that those organisations that are first movers have no real incentive to act as references, thereby losing competitive advantage.
For several years, Paremus have been working with a number of organizations who are at various stages of OSGi migration. While Paremus cannot discuss specific details, or disclose the organizations by name, we are able to reference some common themes that demonstrate real world benefits for OSGi in general, and the Paremus Service Fabric in particular.
When working with traditional applications, a lack of information concerning required libraries results in developers having to load every possible library into their IDE. From experience, Paremus have seen this drive memory requirements as high as 40GB per developer desktop. However, once dependencies are understood, and mechanisms are in place so that only the required components are loaded, Paremus have also seen the number of artefacts reduced by an order of magnitude with corresponding machine memory savings.
For an organization with several hundred developers the saving is considerable. The saving in reduced memory footprint in Production is of course correspondingly larger.
Developing against a monolithic application requires the developer to work and test against the complete code-base. Yet for large applications it may not be possible to test changes in the local IDE, as compile times may be hours. As a result, developers are forced to rely upon unit and integration tests that run during the nightly build cycle. The result is that the whole bug detection and rectification cycle can take days, with an increased likelihood that some issues are not found and leak into production.
In contrast, rapid testing of OSGi modules is easily achieved in the developer IDE, and Paremus have seen this test / fix cycle reduced from days to minutes.
A modular system also lends itself well to many hands being involved in its development and maintenance. It’s not necessary to understand the whole system inside-out, each individual can independently work on small well-defined and decoupled modules. This directly translates to Increased project delivery success rates as smaller, well-contained projects have a higher success rate than larger and more poorly constrained projects.
Most modern build systems allow build-time dependencies to be specified and obtained from various repositories. However, different types of build system repository specify dependencies with different formats; e.g. pom.xml, ivy.xml.
OSGi provides runtime dependency meta-data in its Require-Bundle and Import-Package bundle manifest headers. This provides definitive industry standard dependency information that can be consumed regardless of the build system you use. Hence OSGi decouples your runtime from the source build systems; thereby avoiding meta-data lock-in and allowing different types of build repository to co-exist.
In addition, OSGi supports package-level dependencies. These are finer-grained than the artefact-level dependencies supported by most build systems and can be programmatically determined directly from source code using tools such as Apache Felix Sigil . This avoids the errors that occur when trying to manually maintain dependency data.
Finally, OSGi supports its own repository API: OBR. A current priority for the OSGi Alliance EEG working group; OBR allows searching for package-level dependencies as well as the artefact-level dependencies used by other repositories.
Production Stability, Availability & Agility
Unless you happen to own applications that never change; stability, availability and agility are closely related concerns. The more agile the business service, the easier it is to change from one well-defined state to the next. This may be to introduce new business functionality, apply a patch, or roll back to a previously good version.
The internal structure of monolithic opaque applications is poorly understood; hence upgrades are complex and high risk. This, coupled with the long developer diagnostic / fix cycle, can result in repeated production outages and instability that spans several working days. Operations typically respond by maintaining a number of isolated horizontal silos; attempting to ensure service availability by releasing new software one silo at a time. This issue isn’t just common, its almost universal.
In contrast; with an OSGi-based runtime like the Paremus Service Fabric, each application is self‑describing; meaning that the application structure is now fully described in-terms of dependencies between versioned modules. A system may be deployed in seconds. A running system may be upgraded on-the-fly (i.e. within seconds); and may be returned to a previous well known-state just as rapidly.
Operational Risk & Governance
The implication for operational risk should be immediately apparent:
- From regulatory and business continuity perspectives, the structure of an application is precisely known at each point-in-time; allowing all versions of the application to be rapidly re-constituted.
- Key structural information is no longer locked within key members of staff. In principle the organisation can employee any OSGi literate developer / systems architect who can rapidly navigate the structure of the organisation’s software systems.
From a governance perspective, it is now simply task to rapidly answer the following types of question:
- Which software license types are used within which production applications?
- Which production applications use third party modules with an identified security vulnerability?
This information is readily available courtesy of the metadata embedded in each OSGi module and the dynamic deployment & assembly mechanisms provided by the OSGi runtime environment.
An OSGi ROI?
Anne Thomas Manes (Gartner) estimates that ongoing maintenance accounts for 92% of the total lifetime cost (TCO) of each application. Whilst hardware accounts for <10% of TCO, software maintenance accounts for ~70% of TCO; the remainder being the initial cost of developing the application; see slides 9 & 10 – SOA Symposium: Berlin, October 2010.
Given this, the current fashion for virtual machine based Cloud computing seems somewhat perplexing. The deployment of traditional opaque software stacks as virtual machine images does NOTHING to address the issue of application maintainability.
Closer to home, a group of Financial Services engineers recently attempted to quantify the potential return realised by migrating their production environment to the Paremus Service Fabric. They conclude that OPEX savings of 60% were possible. While Paremus do not know the details of this analysis, it is likely that the following three contributing factors were considered.
- The Paremus Service Fabric is a cost effective replacement for Application Servers, Compute Grids, CMDBs and provisioning tooling and requires less operational resource to manage it.
- The Paremus Service Fabric, being a Private Cloud, achieves the resource utilisation and efficiency savings alluded to by many virtualisation vendors, but without the management overhead and operational risk associated with ‘virtual machine sprawl’.
- However, given the Gartner TCO analysis, the bulk of the identified OPEX saving most likely results from the ongoing maintainability of OSGi based business applications running upon the Service Fabric runtime.
A large Financial Service organisations may have 1000+ applications, each with an average annual running cost of ~$1,000,000. This equates to an potential annual saving of $600,000,000! A significant medium term bottom line saving that surely warrants investing in a multi-year OSGi based application transformation program?
Hopefully a compelling set of arguments for adopting OSGi have been presented, but how do you actually migrate an organisation with a decade of legacy applications to a new OSGi based world?
Organisations considering this invariably start with the following questions:
- How do we determine and then untangle dependencies in our current environment?
- How do we move to an OSGi-centric build/test/release cycle within minimum impact on the majority of developers?
- What level of modularity / granularity should we pursue?
In response, Paremus advise an iterative approach; the precise details dependent upon each organisation’s starting point and business objectives.
1. Assemble a small high-caliber team of engineers with appropriate OSGi / modularity skills. Ensure that this team have senior management backing and representation.
2. Set up tools to determine and monitor dependencies within the organisation’s existing code base. Provide automated reporting, remove superfluous, and fix unsatisfied, dependencies.
At this early stage, the organization may have already achieved as much as an order of magnitude simplification in the code base dependencies. This in itself will improve developer productivity by reducing compile times and help increase the success of production releases.
Stage II – Tooling & Metadata
3. Review organisation’s standards for IDE tooling. Does current tooling support OSGi metadata and enable simple management of exposed dependencies? Review organisation’s current repository standards; will these support the OSGi Alliance OBR standard? If required, select new tooling.
4. Set up OSGi metadata for all organisations projects; this to be maintained by project developers. This metadata need not yet be used at runtime, but it can be used during the build process to monitor the progress of the modularisation effort.
It should now be possible to run OSGi and standard Java variants of an application side-by-side. This allows migration to progress without creating large parallel branches in source control, which are difficult to subsequently re-merge. CAPEX savings may be realised at this point, as only the required artifacts will be loaded into the developers IDE, decreasing the amount of resources required and increasing testability further.
Stage III – Runtime
5. Select candidate applications for migration based upon agility and re-usability considerations:
- A set of applications that share a high degree of functionality / code and require frequent functional updates are excellent candidates for early migration.
- A standalone application, which seldom, if ever changes, and shares little or no functionality with other applications, is a very low priority – and may never be migrated.
6. Create working runtime bundles using existing libraries as the level of granularity.
7. Create integration tests for use during and after migration to assert that parallel development streams do not break any modularisation efforts as the code is migrated.
8. Test OSGi version of candidate application in an OSGi runtime environment; e.g. Paremus Service Fabric.
9. Create integration tests to ensure fixes for hidden gotchas are caught during development rather than in production; i.e. Development and UAT fabrics.
The organization should now be in a good position to deploy OSGi based applications to production and accelerate OSGi migration. The new development, build, test. release lifecycle will enable considerable improvements in developer productivity via further reductions in time to run unit and integration tests.
Stage IV – Iterate and Reward
10. For each application take an iterative approach to modularisation; break down the deliverable into smaller modules (improving modular boundaries), test, deploy, release to production. Then re-visit modularity, break down further if appropriate, test, release.
11. For each migrated application, using the dependency information now available, publish ‘composition reports’; listing degree of re-used of in-house modules, certified third part modules and alerting on uses of non-certified modules.
12. The modularisation process will now be running across many candidate applications in parallel; during which the core OSGi team will continue to advise each participating application group on appropriate levels of modularity and opportunities for re-use.
13. Reward application teams; not just for meeting an initial delivery objectives, but also on:
- Use of agreed third party OSGi libraries, where appropriate.
- Achieve a high degree of re-use in a re-factored applications.
- Delivering and maintaining in-house modules which are re-used in many in-house applications.
The ‘composition reports’ and associated incentives provide the development teams with powerful feedback mechanisms.
In this manner, the organisation typically starts with an isolated application, building the required skills & processes. Driven by initial successes and cost savings, adoption naturally flows from this initial seed point through the organisation.
Whether OSGi is destined to be the next IT ‘fashion bubble’ (once the current hysteria on ‘Cloud’ has waned) or it will grow organically via initial adoption by the most sophisticated IT organizations, is unclear.
However, the software industry is a manufacturing industry; no more and no less than manufacturers of disk drives, steel or automobiles. While the raw materials are low cost, the ongoing effort required to craft and maintain flexible, high quality software, is not.
Necessity will drive the software industry, and those organisations with large in-house development teams, towards modularization, dynamic application assembly and so OSGi.
The author has had direct experience of this behaviour in an enterprise environment.
A backup solution that was initially designed for a hand full of servers was nearing collapse. Each time a new backup server was added to address over-runs the reduction in backup time was less than the 1/n expected as loads could not be exactly balanced across backup real estate. The situation degraded to the point where the backups extended to a full 24 hours. The correct solution was to refresh the enterprise network and consolidate backup servers into centralized high speed silos. Yet, despite the operational risk, some management refused to sign off on what was perceived to be a high risk project. Only when no ‘last’ silver bullet could be found was sign-off achieved to progress the correct strategic solution. The situation had by then become so serious that this solution had to be implemented as rapidly as possible.