Hot and clustered

Clusters are not just things you find in your cereal bowl - they arefast becoming the hottest way to soup up your servers. Annie Gurtonexplains

The Internet and the huge amount of data that organisations have to handle are fuelling user demand for mainframe power on the desktop.

Combined with an increasing expectation to be able to run mission-critical applications on distributed PC platforms, this has put vendors and resellers under pressure to deliver clustered technology.

Clustering is simply the linking of several servers to provide high-performance, fail-safe protection and reliability with improved network management.

The crucial element is the software which manages the cluster and treats it is a single entity. Clustering is not new - Digital developed the technique more than 20 years ago to help Vax computers compete with IBM mainframes - but it has taken on a new lease of life and become more affordable.

Compaq and Microsoft, among others, are making plenty of noise about clustering, and the arrival of the quad processor Pentium Pro boards from Intel has led many vendors, including Tandem, to announce clustering solutions.

ON THE NODE

Now resellers and Vars can offer their customers the benefits of high-end clustering - improved scalability, performance and power. The prices are high enough to offer good margins and allow Vars to provide solutions which take over the management of customers' most valuable commodity: their data. It's not surprising that clustering is proving attractive to the channel.

Stratus Computers was one of the first firms to bring high-end clustering technology to the desktop. Several vendors can link two Pentium servers, but Stratus is delivering a multi-node NT cluster using Pentium servers, called Radio, which will connect up to four computing servers and two storage servers within a single cluster. By the end of the year the company plans to offer any combination of up to 24 computing servers and storage servers in a single clustered system, as well as support for the Pentium Pro.

Gary Haroian, president and CEO of Stratus Computers, says: 'The advantages of flexibility and scalability are very attractive to resellers and their customers. The benefits of being able to run several applications in one environment, with all the computing availability that is required, means that solutions can become more sophisticated without compromising integrity.'

Stratus users can replicate Microsoft's SQL Server, Internet Information Server and Exchange Server in a single clustered environment, as well as running existing or off-the-shelf applications.

A WILD BUNCH

Clustering has not previously been available at the desktop because of the software it requires. Managing the aggregate power of two or more coupled servers is greater than the challenge of managing individual machines.

'There are not a lot of good clustering software solutions available, and the simplest are still prone to fall over,' says Haroian.

But the need for mission-critical power on the desktop has focused developers' minds. Database giants such as Oracle and Informix are working with hardware vendors to enable their software to work with clustered NT servers, although critics doubt that the software has achieved the necessary degree of reliability.

The PC operating systems and NT have yet to be proved to be sufficiently mission critical to deliver all the promise which the vendors are hyping up.

Haroian says Stratus is working on technology which will manage instant fall-over using its own software called Isis. And Microsoft is concentrating on boosting the reliability of NT so it can compete in the financial services market, where it so dearly wants to dominate.

HOT PROPS

Although margins are excellent on clustered systems, there is also good margin to be had in support. But not all resellers are geared up to provide the zero error level of support which is necessary. When a customer has committed its entire data warehouse to a clustered system, it is vital that the system is robust and any support or maintenance requirements are reliably met.

Paul Brennan, general manager of Netframe, says that although a growing number of hardware vendors are offering clustered systems, the level of support available to users and resellers is mixed. 'First-class support is even more essential, if that is possible, than with simple networked servers,' he says. 'But not all resellers have that level of expertise or the interest in providing it.'

All Netframe's clustered systems offer remote support and hot maintenance, which means the user's system is connected direct to the reseller, or to a Netframe technical partner if the reseller is unwilling or unable to provide the technical expertise to manage the system. 'Our systems are able to provide remote diagnosis and get a crashed system up and running quickly, with minimum interruption to the business,' says Brennan.

It is that element of continuous computing that users are looking for when they turn to clustering, says Brennan, and it is critical that resellers can deliver it. 'Clustering offers so much to users and software providers that are looking for increased data throughput and improved reliability.'

He says resellers are looking for ways to differentiate themselves and being able to provide first-class remote support on proven clustered systems is a good move. 'Some vendors are just moving into clustering, but we have been providing it for seven years and have satisfied clusters and reference sites around the world.'

SILVER SERVERS

Netframe has just won the Interop award for server of the year at the recent Interop Show, and Brennan has high hopes of picking up more awards in the near future. 'Netframe has the established credibility in a sector where there are lots of entrants,' he says. 'Resellers are under pressure to provide measurable improvements to cost-of-ownership figures, and we can demonstrate a 44 per cent saving when servers are consolidated at a ratio of three to one.'

Paul Craddock, net servers marketing manager of Hewlett Packard, says resellers need to be prepared to answer a lot of questions from users about clustering. 'Users raise queries about availability and performance,' he says, 'so resellers can offer a checklist which will help the decision process.'

THINK ABOUT IT

When considering clustering, says Craddock, the reseller should first go through the customer's needs for available processing power. 'These should be examined and analysed,' he says. 'It makes sense to take full advantage of existing, proven high-availability solutions and to ensure that the current server is stretched as far as it can be without threatening integrity or reliability.

'The customer and reseller should ask: are the fastest processors already being used? Are they taking advantage of multiprocessing within the servers?

Do the servers have enough memory? Do they make effective use of cacheing?

Do they have robust, high-speed network connections? Do they make use of multiple I/O buses for greater throughput?'

It is important that the links between clustered servers are standards-based not proprietary. 'You don't want to find that the server vendor's clustering software has to be revised by that vendor every time the OS or NOS vendor issues a new release,' says Craddock.

'HP has wide experience of providing complex systems into demanding enterprise environments. Our clustering strategy is to couple that expertise with industry standards. Clustering standards are still emerging, so resellers and users should go carefully. The first move for resellers is to analyse the existing system and question whether it has been extended as far as possible.

'Clustering is the only way to meet customers' demands for availability and performance, both of which are fundamental to HP's Netserver strategy.'

TRY BEFORE YOU BUY

The final word comes from Jack Lewis, Amdahl chairman and chief executive.

He says: 'Support and confidence for clustered system users is crucial, and with that in mind we have announced an enterprise computing centre which offers users and resellers a seven-point migration methodology to simplify the implementation of a clustered system.'

Amdahl is offering a clustered system called Parallel Sysplex, which customers and resellers can test and evaluate at the new centre. 'Users need an opportunity to evaluate their migration strategy and we aim to give them a road map which matches their business and computing needs,' says Lewis. He is in no doubt that clustering is the way forward.

'Clustering offers lower operating and software costs compared with traditional multiple systems. Users are increasingly looking for sysems which deliver faster throughput.'

NUTS ABOUT CLUSTERS

Unisys positions clustering between network computing and massively parallel processing. Martin Sexton of Unisys says: 'Clustering is a set of symmetric multiprocessors (SMPs) that are connected by high-speed interconnection so the cluster acts as a single system. In most clustered implementations, peripheral resources and data are shared among the nodes, creating the term shared data.'

Clusters offer several advantages over networked computers, says Sexton.

'A cluster can present a single image to applications and users and offer an ideal environment for implementing a high availability solution.' But growing a single platform SMP system into a larger configuration is not a simple a matter of connecting multiple servers with an interconnect and adding some software to control the interbox communications. 'Several issues arise,' says Sexton, 'both at the administration design level and at the site administration level.'

Resellers are perfectly capable of dealing with these issues, adds Sexton, but the complexity has two effects. 'On the one hand resellers have to be technically competent, but on the other hand the value of clustering in terms of margin and value add is increased,' he explains. Resellers providing software solutions which would benefit from a clustered hardware approach can work with vendors so that the customer gets a fully supported comprehensive solution.

CACHE THE BUS

Netframe's Paul Brennan explains that it is the single sequential bus architecture of PCs and servers which has created the demand for robust clustered technology and systems. 'The bus has always been a headache and bottleneck,' he says. 'We've seen memory and cacheing and everything else about a system change, but the bus is the same. It means that systems cannot be scaled online, which is a big weakness.'

Brennan explains that any user who wants to add an element like a new storage device to a system has to take it off-line to effect that upgrade.

'The system is down for however long it takes the engineer to effect that upgrade,' he says, 'and for many users it is just not good enough. PC servers run out of steam at a few hundred users, but clustered servers can support a thousand or more users.'

Questions about data integrity are common with ordinary PC servers, he says. 'There is no form of parity checking, but we can deliver that with a clustered system. The problem is that a single sequential bus locks up and kills the server, but that is not a problem with a Netframe clustered system. By distributing the computing away from the central I/O,' says Brennan. 'We split the computing on to separate cards or clusters creating a distributed workload model, which avoids the problems of the single sequential bus.'