 | |
A prototype demonstrates that multi-site transactional Grids are possible -- and valuable
| |
 |
|
"...applying the grid to transactional computing will forever change the course of computing and bandwidth demand."
- GRIDtoday, 10/07/02
"The Internet is going to become a persistent transactional grid."
- Gartner as quoted by Infoworld, 04/18/02
The challenge Grid computing helps customers tap into the unused processing cycles on networks of computers across distances that can span the world. Grid computing is already successfully being used for scientific applications that require large amounts of processing power as well as computationally intensive financial applications, such as portfolio optimization and risk analysis. Applying that processing power to everyday transactions and core business processes is the next step. This would bring the advantages of Grid -- flexibility, adaptability and reliability as well as reduced computing costs and capital expenditures - to businesses. IBM Research is already working on many of the key technologies needed to make transactional, multi-site Grids a reality.
The approach Transactions are day-to-day business computing that supports core business processes. To have atomicity (indivisibility), consistency, isolation and durability (the so-called "ACID properties"), transactions must be completed in their entirety or have the ability to be cancelled. As an example, think about the transaction for getting money out of an ATM machine. You wouldn't want the first part (a request for cash and the deduction from your account) to be finalized without completion of the second part (getting the cash). The other variable is timeliness - waiting ten minutes for your cash is too long.
The possibility of multi-site Grids opens up the vision of ubiquitous computing power - always there, always on, infinite. It also means that transactions could be processed on your local computer, in another business unit or in another country. Offering reliability and response no matter where processing is done is one of the goals being addressed by IBM Research.
The solution developed by IBM uses Service Level Agreements (SLAs) and autonomic controls. SLAs define the transactions that have priority and the controls detect (or predict) when an SLA is not being met. Additional autonomic controls are poised to adaptively reassign computing resources to help meet demand.
When there aren't enough resources for high priority transactions at a local data center, resources at remote sites can be negotiated and configured, and lower priority transactions can be routed to the remote Grid. This helps meet SLAs for higher priority transactions without stopping the low priority transactions.
Figure 1: Transactional Grid Vision
Using a Grid built on open standards that IBM helped develop, peak demand is handled by using resources at local or remote sites instead of having the full capacity on hand. Since the peak demand can be many times the average, companies are expected to realize large savings by sharing resources across sites connected to the Grid. This way, available resources or "white-space", can be used as needed.
The solution may sound simple, but it involves complex real-time coordination between applications that serve as routers, providers, monitors, balancers and managers, all working across computers that could be half a world away.
A working prototype of this solution has demonstrated that a multi-site transactional Grid can support overload at one site by dynamically off-loading peak traffic to other sites in an intra-organizational Grid. It has also shown that this can be done to keep transactional response times within reasonable limits.
Next steps Right now, some of the monitoring and load balancing processes developed for the transactional Grid are being used for clustered systems that share resources to focus on computationally intensive jobs and transactional applications. Some clients are beginning to realize the benefits of:
- Configuring for average, not peak loads
- Improved availability
- Extended capacity
- Flexible pricing
As single site clusters extend into multiple clusters and multiple sites, Grid solutions will be attractive to enterprises with computing resources in different time zones with the need to access distributed data and Web services or to optimize how computing resources are used. Taken all together, the Grid solutions on the drawing board can be expected to offer great benefits for companies in all industries.
| |
|
|
Related case study | 
A large national brokerage firm A financial services company achieves new levels of customer service with Grid computing. |
More on the issue | IBM Grid computing: Grid computing powers e-business on demand Explore introductory and advanced Grid computing and IBM’s role in this field. |
More on the team | Daniel Dias For the past 20 years, Daniel Dias has worked at the edge of parallel computer processing, pushing the limits of speed. |
More on research | IBM Research: Distributed & fault tolerant computing An update on IBM Research's latest work in this area. |
|
