Buyers' Guide
Xeon Buyer's Guide: Why downtime isn't an option
Published: 30 Oct 2008, 12:01am
Everyone needs IT to run their business efficiently and effectively. That requirement is steadily increasing and, as IT equipment becomes more indispensible, it also needs to become more reliable - we simply can't afford to be without it.
Servers, and other hardware, have to fail less often and be quicker to fix. And these benefits have to be provided at ever-lower costs, so reliability is available to all.
"Five years ago, we could live without email, even for days at a time," says Intel solutions architect William Crowe. "Now, even a few minutes is too long."
Email is one of the many applications that have crossed the divide, moving from being something that wasn’t business-critical to joining tasks such as stock control, customer accounts, and inventory as one of the things almost no business can now do without.
As a result of our increased dependence on IT, reliability is an increasingly important demand put on IT managers. Service outages, even for planned maintenance, have become simply unacceptable in many organisations. Millions of pounds of business may hang directly on applications being continuously available. Even if it doesn't cause a direct loss, a break in service may result in customers losing patience and going elsewhere.
Hardware has risen to this challenge, formulating the requirements as Reliability, Availability and Serviceability (RAS).
Fundamentally, reliability is the result of good engineering: making sure that devices are used within their working limits and making sure any element which might fail has some redundancy or duplication. Servers can have redundant disks as well as regular back-up plans, and other aspects of the system such as network connections can be duplicated.
The move has come through a conscious drive to build more reliable systems and also as a beneficial spin-off of architectural changes. For instance the move to parallel architectures, using multiple processors, is both delivering greater reliability and fuelling the demand for it.
Multi-core adds to reliability
Running applications on multiple processors gives better performance for a given price. It also means, if the architecture is designed properly, the system can survive the failure of one of those processors and continue working.
The flip side of this coin is that more powerful servers support more applications, as users consolidate their IT onto fewer machines. If one server goes down now then 20 applications can be lost - big servers simply become not only essential, but business-critical.
"The increased awareness of business criticality requirements is driving a rapid shift towards four-socket servers as opposed to two-socket servers," says Crowe. "Multi-processing machines have inherently greater RAS features that enable them to recover from system hardware faults."
Features that were once the preserve of fault-tolerant systems have now come through into lower-end servers. Modern four-socket Intel Xeon machines have similar levels of RAS capability to those traditionally available in higher-end mission critical Intel Itanium-based systems.
Intel took a decision some years ago that, instead of increasing clock rates to produce more powerful processors, it would multiply the processing cores on a given chip. This has kept reliability on track because higher clock-rates mean higher temperatures, which makes the engineering tolerance of the system much tighter, so reliable systems are harder to build.
Virtualisation adds strength
Other factors come into reliability, such as manageability and serviceability. If faults are flagged quickly then components can be replaced before their failure has a serious impact on the overall system. Indeed, automatic systems can monitor the health of components like disks, so they are swapped out before they fail.
And hot-swappable components means that there is no downtime when the inevitable happens and those components need changing.
Virtualisation - the technology which allows multiple virtual "servers" to be run on one physical server - massively adds to reliability, and the benefits are increasing as the technology evolves.
In particular, virtualisation now includes the ability to move virtual hosts from one server to another without shutting them down - using tools such as VMware's VMotion. This function can be the basis of fully-fledged disaster recovery strategies. If one server fails, the virtual machines it is supporting can be brought up on another server (or even in another data centre) to continue where it left off.
This function has been of limited value so far, as it has required the source and destination processor to be exactly the same. Now the newest processors from Intel include FlexMigration, a feature which will maintain the ability to move virtual machines, even across different generations of Intel processors.
"Depending on whose statistics you read, a large percentage of downtime - up to 80 per cent - is associated with planned migration," says IDC analyst Ken Cayton, in a White Paper, Choosing the Right Hardware for Server Virtualisation. "Because of the large percentage of downtime associated with planned outages, IDC believes this live migration capability can really begin to impact the availability of a customer’s applications."
Without this increase in reliability, certain kinds of business, such as outsourcing would be simply impossible. For instance, Affiliated Computer Services (ACS), a global outsourcing company has been building a consolidated virtual environment using VMware running on IBM xSeries servers that use Intel Xeon 7300 processors.
“The Intel Xeon processor 7300 series and large memory capacity of the Intel Server Board deliver the performance we need to support multiple virtualised applications on a single physical server,” says ACS's information management manager, Alton Ingram.
Without high reliability, no outsourcer would put multiple customers' applications on a single server.
Digg
Slashdot
Del.ici.ous
Stumble
Reddit
