Oct 10 ’07

Distributed Linux in a Virtual World

by Editor in z/Journal

Corporate IT is growing its deployments of Linux and actively seeking out areas to virtualize it, but distributed Linux systems still have their place in enterprise architectures.

In some cases, systems and architectures are firmly entrenched and a distributed deployment of Linux seems to be the best option. In other cases, there are application- driven arguments for a true, distributed solution. For the majority of IT shops—and for vendors—enterprise IT architecture is sufficiently varied that everyone understands all these   platforms must work well together—and there’s no single solution that addresses everyone’s needs.

“There has been a substantial workload transfer to Linux over the past few years,” says Carlos Montero Lucque, vice president of Product Management for Novell. “Just a few years ago, Linux was largely concentrated in areas of network management and edge computing. It’s now being used with mission-critical applications such as ErP [Enterprise resource Planning], CrM [Customer relationship Management] from companies such as SAP, and on platforms such as WebSphere and Oracle.”

Montero Lucque says that a growing number of sites are deploying distributed Linux using IBM xSeries and pSeries servers. Sites are taking these servers and migrating Unix applications to Linux. The selling points are lower costs and the robustness and stability of Linux. There also are abundant Linux technical skills available in the marketplace— and the Linux operating system can be tailored to a specific IT environment.

Why Distributed Linux?

While sites are taking a serious look at Linux virtualization on platforms such as the System z, there are still industry and application niches where a distributed Linux approach excels.

Retail

One example of significant distributed Linux deployment is in-store retail transactions.

“Retail is a big example where enterprises like the feeling of having servers close to the actual operations,” says Vince re, senior vice president and chief software architect for CA. “Most of them have a server in every store for in-store transactions.”

Several factors favor distributed Linux systems in retail. First, retail business managers have historically felt they had better response time for in-store transactions from a local server than from a geographically distant central server. With bandwidth improvements, that argument has gone away—but store managers still like the comfort of having local failover capability for in-store transactions.

Second, most large retailers are deployed with distributed Linux systems that easily integrate with their instore Point of Sales (POS) applications and with their daily retail operations. Virtualizing a distributed architecture that spans thousands of locations in different geographies is a huge project that requires a convincing rOI for upper retail business management, which largely came up through the organization by starting as store managers with their own in-store servers. Third, multinational retailers employ distributed Linux computing because there are different server purchasing, support, and configuration standards for each country that must be followed, even on an “open” platform such as Linux.

Banking

Also very transaction-sensitive, banks and other financial institutions are accustomed to using distributed servers in local branches as failover mechanisms for mainframe-provided processing. “When a central processing unit goes down during the day or in a disaster recovery situation, the procedure at bank branches is to manually process transactions and then post the transactions afterward, when the system is up,” says one Seattle bank CIO. “This is irritating and troubling for customers, who want their financial records posted right away. Paper ledgers of transactions also invite user error.”

Because of this, most bank processing software vendors offer server-based, inbranch systems that can step in with “store and forward” transaction services when the mainframe is down. Store and forward allows bank personnel to log transactions and serve customers without using paper. The transactions are temporarily stored in local servers. Once mainframe service is restored, the transactions are then forwarded to the mainframe.

The arguments in banking for a distributed approach are similar to those of retail. Bank managers feel they are in control of their transactions and their customers with localized failover support from an in-branch server. It’s also a very serious perception issue for financial customers to see their transactions logged onto paper ledgers when the system is down, with the teller informing them that the transactions will be keyed into the system when the system is back up. Few bank managers are comfortable with “virtualization,” which they perceive as putting all their eggs in the basket of a central processor at headquarters that could fail.

The Manufacturing Floor

Automotive parts depots, airplane and automobile assembly lines, and other manufacturing applications frequently rely on distributed servers for specialized parts and drawing lookups. In these instances, a feeling of local control over critical data and the fact that access latency may be reduced by having a local server are critical driving factors. “In these cases, the factory flow simply can’t tolerate the extra network loop that communications with a central server would create,” says CA’s re. “If a manufacturer has shop floors with a lot of robotics, distributed computing with proximate physical servers is the best solution.”

One large U.S. jet manufacturer places all its engineering drawings and bills of materials on a local server in the factory. Factory staff can maintain these drawings, and the investment in a single server that factory users “own” is less expensive than a mainframe corporate resource that’s more business- than engineering-oriented, and that must be shared with other users. With the Linux operating system and open source solution sets, cost is further reduced.

Research Laboratories and Universities

The intense mathematical computa- tions and modeling performed by university science departments and public and private research labs demand separate, distributed servers that can serve as dedicated department or project “workhorses.” As separate entities, these servers, like their mainframe counterparts, can easily link into “grid” architecture in which enterprises and institutions can cross-communicate with each other and share resources.

Both mainframes and distributed servers deliver high availability through clustered or grid computing. However, above and beyond this, many science and research departments require supercomputers to solve certain scientific and business problems. In this arena, Linux is the driving software behind most systems.

This has encouraged many universities, laboratories, and businesses to gradually migrate from distributed Unix to distributed Linux, since similarities between the two operating systems facilitate migration. In making the move to Linux, organizations know they will be able to control the Linux distributions they use, as well as the end applications. Finally, Linux is an operating system that can be very easily customized down to the bit level—and research labs and university science departments usually have the staff on board who can perform this customization.

Multi-Location Customer Applications

Companies also are finding it to their competitive advantage to deploy distributed Linux for value-added services for customers. Song Airlines features a Linux-based, in-flight, multi-media system with Internet connectivity for passengers, email messaging, GPS tracking, 24-channel television broadcasting, access to an in-flight MP3 playlist library, in-flight shopping, connecting gate information, and an interactive network trivia game. Each commercial jet has its own Linux server.

Traditional “Best Practice” Distributed Linux

Most organizations already deploy distributed Linux servers for applications such as Web and email administration, Domain Name Service (DNS), Dynamic Host Configuration Protocol (DHCP), Lightweight Directory Access Protocol (LDAP), file and print administration, network application drivers, and localized storage repositories for Network Attached Storage (NAS) and backup. Over time, some will choose to virtualize many of these functions on a monolithic platform such as the System z, running instances of Linux.

The decision will be determined by whether the applications can be run more economically and simplistically at a central physical source.

The Management Challenges of Distributed Linux

Any distributed server scenario offers predictable challenges to IT, and distributed Linux is no exception.

“The challenge in retail or any other business when you are using Linux servers at thousands of different locations rests with central IT,” says CA’s re. “It’s a monumental challenge to keep up with patches, system upgrades, routine maintenance, and repairs. The workflows behind this process are complicated, because items such as scheduled maintenance have to be adapted to different scenarios. A good example is a holiday weekend in one country that isn’t a holiday in another. This is a convincing argument for server consolidation and a process-centric methodology for IT.”

Security and storage also are difficult to manage. For example, how does IT really know what’s going on with remotely situated servers? And if storage is distributed on servers throughout the enterprise, how do you balance capacities? “It’s the physical realities of distributed servers that complicate the overall service picture for IT,” says re. “When you have tens of thousands of servers to cover, and some of these service calls must be onsite, the task is daunting. In such an environment, the asset management piece alone is major.”

IBM and System z Positioning

It’s no secret that IBM has been advocating the adoption of Linux system virtualization on the System z. Few System z shops believe there’s a better platform for virtualization, given the z’s high reliability and excellence in resource management. System z also provides excellent I/O performance and can quickly scale large numbers of virtual Linux servers. Because of this and other factors, there are now historically non-z shops that are seriously contemplating System z acquisitions for their virtualization initiatives.

IBM also recognizes the heterogeneous nature of IT, and knows that virtualization can’t address every issue and application. Customers may choose to virtualize certain Linux servers and applications on the System z, while using the System z as a centralized management resource to oversee other physical Linux servers on the network. For this purpose, IBM, TIBCO and other vendors have strong middleware for distributed Linux servers that support open source solutions and position distributed Linux for Service-Oriented Architecture (SOA).

“Diversity is a good thing in our view, and so we have a variety of ways to get there,” says Jim Stallings, IBM general manager of Linux. “We don't say, ‘here’s our architecture, take it or leave it.’ We take the time to understand our customer's needs, their business problems, and then we work with them to choose a solution accordingly.”

Understanding the Benefits

There are industry niches and applications that are likely to always make a strong case for distributed Linux, despite the impetus in most organizations today to flock to virtualization. IT shops with a System z in their “mix” enjoy an inherent advantage with the z’s already proven capabilities and assets, whether their Linux scenarios are virtualization, distributed servers, or both.

“There will be greater understanding of a virtualization strategy once upper management sees and understands the cost savings,” says CA’s re, “But there also can be a strong end-user reaction when there’s a threat to take a physical server away. In one Wall Street firm, an executive said: ‘That server runs my business operation. It’s really important to me and I don’t want it commingled with anyone else’s computers. What if VMware has a bug, and it affects me? Can you guarantee me that it won’t?’ ”

This is why working with upper business management to understand the benefits of virtualization is one of the top items on CIOs’ lists these days. At the same time, the message needs to be tempered for those “niche” areas that really do run better with a distributed Linux deployment. Z