Home > Cover Story

Servers shape up for scalability, availability

Though mainframe computers continue to exist, servers rule the roost in today's businesses, and borrow from mainframe designs the threat to Unix servers has increased now that Intel has launched its 64-bit processor. Itanium's new architecture has attracted manufacturers of RISC-based servers

Today, IT managers look for multi-processor servers, starting with a single or dual-processor system, and scaling up to several processors

Decades ago, the mainframe
computer hogged floor space in a computer room, and was the heart of a corporation's IT infrastructure. Mainframes (now much smaller in size) continue to chug along in large corporations, but smaller and more powerful servers are more widespread. The computing environment in enterprises has evolved from a single central processing unit to distributed computing units (servers) spread out over a network. Today there are servers dedicated to specialized tasks like primary storage, Web hosting, telecommunications, file & print services, application and database hosting etc. The demand for servers has risen in the past year as more businesses got e-enabled and began transacting over the Internet. This resulted in the formation of Internet Data Centres (or server farms) and the expansion of enterprise networks. Anyone shopping for a server today has to make important decisions about processors, brands and other technical specifications. But the key requirements for today's fast expanding businesses are scalability, availability, capacity, processing power, uptime, and ease of management.

Mainframes or servers?
Not long ago, IT managers thought that RISC-based (Unix) servers would replace ageing mainframes. However mainframe computers are still manufactured today, although they have shrunk in size considerably. Some enterprises continue to hang on to legacy systems and even use middleware to tie these in with Web front-ends or new IT infrastructure. While new mainframes are sold mainly to large public utility and financial organizations, Internet driven businesses favor servers.

Meanwhile server manufacturers are taking the best from mainframe designs and putting these features in servers. An example is the hot-swappable feature that lets administrators swap components like memory modules, disk drives or CPUs without switching off the machine.

The battle between Wintel and RISC-based servers continues, though Wintel systems are fast approaching the speed and performance of RISC systems.

RISC Vs Wintel
Wintel servers (servers that use Intel processors and the Windows operating system) are seen as a threat to RISC-based (Unix) servers. Intel's server processors get more powerful each year and are fast approaching the performance levels of RISC-based systems. Wintel servers also cost far less than Unix servers but analyst say cost alone won't give Wintel an advantage. This is because manufacturers of RISC-based processors have proven technologies and well-defined roadmaps. For intensive processing tasks, IT managers are comfortable using Unix-based servers that approach mainframe performance.

RISC-based servers continue to hold out with HP, Sun and others introducing new models. Last year Sun took the lead in server sales thanks to demand for its Netra servers. HP recently launched its 9000 Superdome Unix server that was well received in the market.

But the threat to Unix servers has increased now that Intel has launched its 64-bit processor, Itanium. Itanium's new architecture corrects the deficiencies of RISC systems and has attracted manufacturers of RISC-based/Unix servers. SGI, NEC and Hitachi already offer Itanium servers pre-loaded with Linux; HP is porting its HP-UX Unix variant to Itanium servers manufactured by NEC; Sun has already ported Solaris to the Itanium platform. IBM has also ported its AIX 5L (a Unix variant) to Itanium.

Now that Intel has gained support from traditional Unix server majors, it is only a matter of time before Itanium processors match mid-range 64-bit RISC servers in performance and grab market share from RISC servers.

Whether RISC-based or Wintel powered, manufacturers continue to design servers keeping in mind key business requirements that were mentioned earlier. Here are a few changes that we can expect in the next generation of servers that are already hitting the market.

Ultradense servers
Space has always been an issue at server farms and data centers. The problem was circumvented with the introduction of super slim rack mounted servers. These servers are just 1.75 inches thick and are about the size of a pizza box (a form factor known as 1U). Processing power can be scaled by adding more servers to a rack. But this calls for additional air-conditioning and power requirements, making expansion an expensive affair.

Server manufacturers are now contemplating a new design dubbed 'Ultradense' or 'Blades' a design that packs in more computing power within the same form factor. Such Bladed servers come in the form of single boards, one-eighth the size of a typical 1U (1.75-inch-high) server and consume up to 12 times less power.

The RLX System 324 (from RLX Technologies) packs in 336 bladed servers in an industry-standard rack currently designed to house 42 1U servers.

Among the first server manufacturers to roll out Bladed servers are RLX Technologies, FiberCycle, and Amphus all use Transmeta's Crusoe processor, which consumes less power than mainstream PC (IA-32) processors.

But the server heavyweights are not standing by as mere spectators Compaq, IBM, Dell, HP and Sun are furiously working on bladed/ultradense designs and are expected to roll out products by the end of this year. Look out for HP's Powerbar and Compaq's QuickBlade servers (among others).

Ultradense servers will be targeted at communications companies, IDCs and ISPs who run data centres packed with servers.

Availability
'Big iron' servers are designed for 24x7 operation and should ideally offer 100 percent uptime. Further, data stored on servers should be protected against power loss or system failure. In reality, server components do fail and need to be replaced with minimal downtime. But servers had to be powered down for maintenance resulting in downtime, which affected business. According to the Standish Group, server failures that lead to downtime can cost companies as much as $27,000 per minute.

Hence, today's servers are designed with fault-tolerant, redundant and hot swappable components. Dual processor servers with hot swappable power supplies, disk drives and memory modules top an IT manager's shopping list.

The Compaq Proliant L380 and ML370 servers have a backup memory bank to take over immediately from one that's failed, letting the defective part be replaced on a more leisurely schedule. Memory can be upgraded in Compaq's 500 series of Intel servers without shutting them down. And the 700 series will have memory technology similar to the RAID data-protection system found on hard-disk arrays.

Scalability
IT infrastructure must keep up with growing business, which means processing power, memory and storage on servers should be scalable. Today, IT managers look for multi-processor servers-starting with a single or dual-processor system, and scaling up to several processors.

The HP 9000 Superdome server for instance, can scale up to 64 CPUs, each operating at 550 MHz. Its memory can scale up to 128 GB.

Sun's Enterprise 10000 server can scale from 4 to 64 CPUs. Sun offers a choice of processor clock speeds for this server: 250, 336, or 400 MHz. The memory of this server can range from 2 GB to 64 GB. There are an array of options for mass storage, which can scale all the way to 20 terabytes with the Sun StorEdge A7000 system.

Appliance servers
These are single function servers used for dedicated tasks like Web caching/acceleration, Web hosting, networked attached storage, load balancing etc. Appliance servers are designed for quick installation and simple maintenance. These systems come with a pre-loaded operating system and application software (often Web server software) that simplifies deployment so servers can be plugged into networks as easily as desktop PCs.

All server majors offer appliance servers: Compaq offers its TaskSmart series, Sun sells Cobalt Qube servers, Dells products are called PowerApp, and IBM offers eServer appliances.

Self-healing servers
When servers are used for mission-critical applications and businesses depend on them, maximum uptime becomes imperative. Over the years vendors have been selling fault-tolerant systems that are largely redundant in nature. But self-healing and self-monitoring systems look for signs of breakdown and immediately take corrective action. All this happens in the background without affecting system performance.

Compaq has been researching and developing self-healing server technology for years and now IBM is doing the same through Project eLiza. IBM is investing billions of dollars into eLiza, which will also extend to storage and other sub-systems.

One facet of eLiza is the 'software rejuvenation' feature that IBM has incorporated in its xSeries servers. The self-healing software detects a process called 'aging,' in which abnormalities build up over time and cause memory errors and consume computer resources. Special software with artificial intelligence can sense when a server is about to fail and then shut the server down systematically, preventing crashes or data loss. The software closely monitors server activity, determines the stress the server is under, and will reboot the machine if necessary.

Soon other server manufactures will be building such features into servers and there might even be an industry standard set for self-healing server software.

Conclusion
In the not too distant future, servers will truly be the powerhouses of network infrastructure, working tirelessly round the clock and addressing the demands made on network resources. On the maintenance front, humans will be required only for physically changing components, as servers will be able to update their software automatically and heal themselves.

Brian Pereira can be reached at brianp@rediffmail.com