By Simon Duddy
From scientific niche to oil and gas, grid computing is starting to go mainstream. Eliot Beer looks at the advantages of grid for regional enterprises. From scalability to cost cutting the benefits are clear but installation and compatibility concerns remain.
|~|Bayado,-Haze_m.jpg|~|Bayado: Any economies an organisation might get from purchasing a grid could evaporate pretty quickly if it has to pay for 256 separate licenses for the OS.|~|The concept of a global grid of computers, networked together to handle massive processing tasks, has long been a dream of the scientific community. Out of this academic vision, though, has come the concept of enterprise grid computing, which is creating a quiet revolution in how heavy-duty processing tasks are handled.
The original aim of grid was to take advantage of under-utilised processors both within and outside an organisation, such as employee workstations, or computer rooms in universities. Some companies carry this idea straight over to their IT environment, especially for projects, which may only require intensive processing power at infrequent intervals.
John Foster, volume systems product manager for south-eastern EMEA at Sun Microsystems, says, “We have a customer in the financial services sector which has actually assembled a grid using various PCs and employee workstations within the company. The firm uses this grid to perform risk analysis tasks, which is something grid is very suitable for. When the PCs aren’t in use, their processing power can go towards carrying out the analysis, instead of sitting idle.”
But for enterprise applications, a much more common approach is to employ a dedicated set of servers in a central location, thus ensuring security and homogenous technology. While this means many organisations cannot utilise spare processing power on the desktops of its workers, it does open up possibilities for different types of cost saving.
“Instead of having to buy very high-end, very expensive servers, an organisation can purchase a number of low-end servers and use them in a grid; potentially this can save a huge amount of money,” says Yasser Ragaei, marketing manager for business critical systems at HP Middle East. “In addition, there are large savings to be made in maintenance and downtime as well, if the grid is configured correctly. If this is the case, it can be as easy as pulling out one server and putting in a new one, with no loss of data and no need to take down the whole system.”
Sun’s Foster says that although industries such as banking and finance, and companies with large databases are starting to take up grid computing, by far the biggest users of grid in the Middle East are the oil and gas firms. The sector has long been at the forefront of computing technology in the private sector, due to the highly complex calculations required to process seismic data or reservoir simulations.
Masoud Al-Mughairy, team leader (IS/IT) at the petroleum development division at the Abu Dhabi Company for Onshore Oil Operations (ADCO), says grid computing has allowed his company to triple its available processing power at one quarter the cost of its previous solution. ADCO runs software such as Eclipse from Schlumberger and VIP from Landmark on its 128-CPU grid from Sun Microsystems, to perform reservoir simulation tasks.
Al-Mughairy says, “The drive for us was performance, specifically price performance, and grid also allowed us to get away from the proprietary lock, where we had to use processors and software from a specific vendor. Now we use Intel Xeon processors, open-source Linux operating systems, and standard interlinks, so we’re not tied to any particular system.”
Grid computing, then, has a great deal of potential to multiply the raw processing power of IT systems for a given budget, while at the same time reducing downtime and maintenance requirements. But there are still several aspects of grid, which make it a challenging technology to implement, along with other factors such as application suitability which can affect the usefulness of grid to an enterprise.
While the basic concept of grid computing is reasonably straightforward, the actual implementation is potentially quite complex, certainly during the initial set up. One of the critical factors in making a grid work is selecting the most suitable management software, according to Michael Grayston, business engagement manager for oil and gas at Sun.
He says, “Because grid has come out of the scientificcommunity, much of the management software available is open-source, and this can be very attractive to organisations wanting to minimise the costs of implementing grid computing. But the management application is the element, which controls how tasks are distributed over the grid, as well as other vital aspects of the processing. While free software can be very good, many companies opt for packages which have support, so that if something goes wrong they’ve got more than an e-mail address to rely on.”
Two other important aspects of grid to consider are the interlinks between the machines, and the operating systems (OS) which run on them. The network connections within the grid must be able to support the data throughput of the processing tasks; for situations where the grid is based on a central cluster of servers, this is less of an issue, but for organisations which are considering splitting a grid between two sites, this could create challenges.
“Many enterprises look at using a grid not just in one single cluster, but over a local area network (LAN) or even a wide area network (WAN),” says HP’s Ragaei. “Sometimes there’s even a need to use a grid through virtual private networking (VPN), and with these increased distances can come issues if the applications are not configured correctly.”
Another issue Ragaei raises is the problem of multiple OSs on a grid, and making sure applications can run across them. While the networking issue only comes into play when using a disparate grid, OS considerations are important even in a centralised cluster grid, according to Hazem Bayado, operations director at Novell Middle East, whose Suse Linux OS is designed for use on computing grids.
He says, “When you’re looking at running an OS on potentially hundreds of CPUs, it’s vital to look at the licensing side, to make sure the agreement is appropriate. Any economies an organisation might get from purchasing a grid could evaporate pretty quickly if it has to pay for 256 separate licenses for the OS, and the same point is also true for any software to be run on the grid.”
While these issues are essentially challenges and considerations to deal with when implementing a grid computing solution, the problem of applications is one which currently presents a serious barrier to the wider uptake of grid. Because of the distributed nature of the processing, applications traditionally have to be specially coded to take advantage of the grid, thus putting the majority of mainstream applications out of the running.
This is starting to change, though, with companies such as HP offering management software which can allow standard applications to utilise grid computing power, usually by virtualising the processing resources. Database applications are also being adapted for use on grid; Oracle’s 10g software has been specifically designed for the technology — the ‘g’ stands for grid, according to Mohamed Al Ojaimi, principle product manager for the Middle East and Africa at Oracle.
“Grid support is a major advantage for intensive database applications, as it gives greater availability and higher performance at the same time,” says Ojamei. “Grids for Oracle could range from four CPUs to hundreds, but typically we see relatively small grids, at least compared to oil and gas applications, of eight to 32 CPUs. But the beauty of grid is it can pay dividends from a very low number of processors, and it is very easy to upgrade if you need higher capacity later.”
The scalability of grid is a major advantage of the technology. In principle any additional machines, once configured with the necessary OS and software, can just be connected to the grid and become available immediately. In practice, the process of upgrading a grid is pretty much this simple, as ADCO’s Al Mughairy explains.
“We are in the process of acquiring an additional 256 CPUs, and we expect to have another 256 processors towards 2007,” he says. “When we bring these new CPUs into our existing grid, we expect the process to be mostly plug and play. We don’t build the machines here, so the vendor will carry out the stress testing and other preliminary procedures before they arrive at ADCO.”
For industries such as oil and gas, then, where massive processing power is a necessary business requirement, there is a strong incentive to make grids work, mainly because the advantages over alternative processing solutions are overwhelming. But for applications where grid is still emerging as a useful technology, there is still a degree of wariness over making the leap from more established systems.
Raed Hudaib, senior database administrator at Fastlink, a telecoms company in Jordan, says his company has recently adopted an Oracle 10g solution, but has not brought in a grid computing solution along with it at this stage. He says he still has some concerns about the technology, which Fastlink has looked at in the past.
“We haven’t adopted a grid solution yet, because we still have some issues about the potential stability of the technology,” says Hudaib. “But it is maturing, and I think we will look again at grid in six months or so, because it does have some definite advantages. At the moment, though, we feel it is still a maturing technology for our sector.”
As the processing power of standard CPUs increases, though, and vendors move away from supporting proprietary processors, grid is set to become more attractive to any company requiring major reserves of processing power as time goes on. Sun’s Grayston says the processing advantage of grid will be multiplied as dual-core CPUs from both Intel and AMD become available, and Al-Mughairy thinks the future for many regional enterprises lies with grid, based on his experience.
He says, “If we’re talking about number crunching, the areas with such requirements are limited in this region, where we don’t have military research and similar requirements. So mostly the manufacturing industries we can focus on are oil and gas, that’s where you see most of the number crunching requirements. But I’m sure grid is becoming applicable to other disciplines, such as finance, banking and data transactions, although obviously I’m not an expert on those areas. But the stories about the use of grid in these areas are everywhere, and we can see the alliance between Oracle and Sun on grid computing as more evidence of this.”
The advantages of grid for Middle East enterprises are becoming clear. The ability to use low-end servers or existing processing resources gives cost-conscious organisations the opportunity to increase their computational resources for a relatively low capital expenditure, and the reliability and ease of maintenance of grid means enterprises will need fewer expensive technical staff to oversee the system. And as the technology matures for non-specialist sectors, allowing more general applications to benefit from multiple processors, organisations trying to plan for rapid expansion may find that grid’s scalability makes it a viable option for their IT infrastructure.
But the key in grid’s success in the region may lie with vendors ensuring they implement the technology in an appropriate way, according to Al-Mughairy. He says when ADCO brings in new equipment he and his staff ensure they test their own software on it first, to make sure it is suitable for their applications.
“We’re not looking for just a machine, we’re looking for a solution, and we’d like companies to be in tune with that,” he says. “Any vendor can provide performance benchmarks for their product, but we are doing our own testing using our own models, to determine what the best configuration is, what the best interlinks are, and so on. That is one of the key criteria we use in our technical evaluation, to make sure any product fits our intended solution.”||**||