ACADEMIA
What Users Want From Servers
By Jennifer Mears, Network World -- Performance, reliability and virtualization capabilities top the list of what users want from their servers. They're also looking for an easier way to manage this pool of resources that are the cornerstones of business. "Simply put, I want control of everything from anywhere regardless of the means of connection," says James Barry, CIO of OneUnited Bank in Boston. "Further, I want to be able to manage below the operating system, and I want to be able to dynamically allocate resources on the fly where needed." 
What follows is a wish list for server vendors compiled from discussions with network professionals:
1. Standard architectures.
Companies are looking for the easiest, most cost-effective way to deploy servers without having to worry about training staff or integrating with existing systems. That means standard boxes.
"I am asking my vendors for a 'commodity' server. I don't want any special differentiating functionality other than reliability and cost - performance is a given," says Ulrich Seif, senior vice president and CIO at National Semiconductor in Santa Clara. "We evaluate our vendor offerings against our requirements and then select the best solution that can be deployed worldwide as a standard."
2. Distributed computing capabilities.
Users are tired of provisioning too much hardware and getting less than optimal use from their infrastructure. They want to be able to share resources on the fly, manage groups of servers and run applications with the best performance possible.
"Servers should become more bladed in nature, and those blades should be able to be combined into a single system," says an enterprise systems architect at a large financial organization who asked not to be named. "Say they are dual-processor blades: I could have a two-, four-, eight-, 16-, 32-, 128-processor system with the operating system actually virtualized across those blades."
To do that, servers will have to support technologies such as InfiniBand, a high-speed I/O switching fabric, he says.
"The data center itself would be nothing more than a grid of compute, storage and network resources," he says.
3. Virtualization.
More businesses are looking at virtualization as a way to consolidate servers, but some complain that Intel-based operating systems aren't mature enough when it comes to handling virtualized resources. W.L. Gore & Associates in Newark, Del., for example, is looking at VMWare as a way to virtualize its Intel-based machines.
"We always want to get more for what we paid for in a box," says Richard Sun, network systems engineer at W.L. Gore. "If the VMWare stuff works, it will address these [Intel] issues."
Virtualization will help network managers make more efficient use of their data centers, users say.
"The ability to allocate resources on one big box is appealing because it simplifies many issues of server availability, electrical and wiring management, as well as possibly streamlining test and staging environments," OneUnited Bank's Barry says.
4. Clustering.
Many network professionals might think of clustering only for reliability and fault tolerance, but it's also being used to distribute workloads and improve performance. Lower costs and increased scalability are reasons some users are looking to cluster standard servers rather than run expensive, proprietary machines.
"This has been a slow transition, and as the technology continues to mature and become more standardized we expect to dramatically increase our use of these smaller clusters," says Scott Mastre, server engineer at Wells Fargo Trust Operations in Minneapolis. "For our needs, using smaller, industry-standard servers and clustering based on common network [operating systems] such as Win2K and NetWare 6 provides an excellent value proposition over high-end, more proprietary machines in terms of cost of the equipment."
5. Dynamic resource management.
Building on the ideas of clustering and virtualization, users are looking for a way to dynamically manage servers as one pool, adding and subtracting resources such as processing power and memory on the fly.
"Crossmark's long-term desire is for a more organic model of back-office resource allocation. The industry is slowly moving toward this model in mass storage virtualization. We would like to see processing power follow this trend more rapidly," says Jason Robohm, director of technical services at Crossmark Holdings in Plano, Texas.
6. 64-bit computing.
The move to this pumped-up processing power is a slow one. Many users say 32-bit servers satisfy their needs today. Even those that use 64-bit capabilities say they'd like greater vendor adoption of a variety of 64-bit offerings.
"The availability of 64-bit computing on the x86 platform will allow me to migrate Sun-based design applications to a lower-cost platform," National's Seif says. "I'll evaluate adopting the [32-/64- bit] Advanced Micro Devices chipset when major server vendors provide it in a standard offering. The use of 64-bit computing is a very specialized segment for us."
7. More resiliency.
Users want servers with more resiliency through failover and redundancy. OneUnited Bank's Barry, for example, says servers should have built-in redundancy.
"In a perfect world, I would like to see a server that never went down because it has redundancy throughout its design. This would mean that a component could fail and the server and operating system would be able to adapt, even if at a diminished capacity," he says. "Current server technology allows this only after a reboot, and every clustering solution has at least one scenario where the solution does not work or it is overly complex."
8. Better support for third-party hardware.
Servers are more networked than ever, and that means third-party hardware is often used to connect servers to each other and to other systems, such as storage. Users want to see vendors provide better support for third-party devices that build out server capabilities.
"Certification for a broad range of hardware is important, and in some cases we'd like to see them be more aggressive in certifying third-party hardware running with their systems," Wells Fargo's Mastre says. "For instance, while we use HP servers exclusively, we are using Xiotech for our SAN disk storage, QLogic for our host bus adapters, and Brocade for our fabric switches."
9. Lower power consumption.
Consolidation and rack-optimized servers and blades can lead to overheating and power drains. Users say they'd like to see servers become less power hungry.
"The cost of electricity and air conditioning is significant for us," says Jeff Wenger, vice president and CTO of Tax Technologies in Bradenton, Fla.
"Because we use rack-optimized equipment, we are susceptible to hot spots within our rack, and this can cause long-term problems. Clearly, this would need to be addressed before blade or other condensing technologies can succeed. Reduction in power consumption also can greatly reduce our monthly costs," he adds.
10. Standard management specifications.
Users say they want to see vendors adopt server management specifications such as the Intelligent Platform Management Interface (IPMI), which lets users manage the physical components of heterogeneous servers.
"National runs large clusters with disks being automirrored as soon as a server is swapped into the cluster," National's Seif says. "I look for uniformity and not functional differentiation. Uniform features I'd like to see adopted consistently and universally are [IPMI], [Baseboard Management Controller] and [miniBaseboard Management Controller] to give me better diagnostics and a leg up on manageability."
What follows is a wish list for server vendors compiled from discussions with network professionals:
1. Standard architectures.
Companies are looking for the easiest, most cost-effective way to deploy servers without having to worry about training staff or integrating with existing systems. That means standard boxes.
"I am asking my vendors for a 'commodity' server. I don't want any special differentiating functionality other than reliability and cost - performance is a given," says Ulrich Seif, senior vice president and CIO at National Semiconductor in Santa Clara. "We evaluate our vendor offerings against our requirements and then select the best solution that can be deployed worldwide as a standard."
2. Distributed computing capabilities.
Users are tired of provisioning too much hardware and getting less than optimal use from their infrastructure. They want to be able to share resources on the fly, manage groups of servers and run applications with the best performance possible.
"Servers should become more bladed in nature, and those blades should be able to be combined into a single system," says an enterprise systems architect at a large financial organization who asked not to be named. "Say they are dual-processor blades: I could have a two-, four-, eight-, 16-, 32-, 128-processor system with the operating system actually virtualized across those blades."
To do that, servers will have to support technologies such as InfiniBand, a high-speed I/O switching fabric, he says.
"The data center itself would be nothing more than a grid of compute, storage and network resources," he says.
3. Virtualization.
More businesses are looking at virtualization as a way to consolidate servers, but some complain that Intel-based operating systems aren't mature enough when it comes to handling virtualized resources. W.L. Gore & Associates in Newark, Del., for example, is looking at VMWare as a way to virtualize its Intel-based machines.
"We always want to get more for what we paid for in a box," says Richard Sun, network systems engineer at W.L. Gore. "If the VMWare stuff works, it will address these [Intel] issues."
Virtualization will help network managers make more efficient use of their data centers, users say.
"The ability to allocate resources on one big box is appealing because it simplifies many issues of server availability, electrical and wiring management, as well as possibly streamlining test and staging environments," OneUnited Bank's Barry says.
4. Clustering.
Many network professionals might think of clustering only for reliability and fault tolerance, but it's also being used to distribute workloads and improve performance. Lower costs and increased scalability are reasons some users are looking to cluster standard servers rather than run expensive, proprietary machines.
"This has been a slow transition, and as the technology continues to mature and become more standardized we expect to dramatically increase our use of these smaller clusters," says Scott Mastre, server engineer at Wells Fargo Trust Operations in Minneapolis. "For our needs, using smaller, industry-standard servers and clustering based on common network [operating systems] such as Win2K and NetWare 6 provides an excellent value proposition over high-end, more proprietary machines in terms of cost of the equipment."
5. Dynamic resource management.
Building on the ideas of clustering and virtualization, users are looking for a way to dynamically manage servers as one pool, adding and subtracting resources such as processing power and memory on the fly.
"Crossmark's long-term desire is for a more organic model of back-office resource allocation. The industry is slowly moving toward this model in mass storage virtualization. We would like to see processing power follow this trend more rapidly," says Jason Robohm, director of technical services at Crossmark Holdings in Plano, Texas.
6. 64-bit computing.
The move to this pumped-up processing power is a slow one. Many users say 32-bit servers satisfy their needs today. Even those that use 64-bit capabilities say they'd like greater vendor adoption of a variety of 64-bit offerings.
"The availability of 64-bit computing on the x86 platform will allow me to migrate Sun-based design applications to a lower-cost platform," National's Seif says. "I'll evaluate adopting the [32-/64- bit] Advanced Micro Devices chipset when major server vendors provide it in a standard offering. The use of 64-bit computing is a very specialized segment for us."
7. More resiliency.
Users want servers with more resiliency through failover and redundancy. OneUnited Bank's Barry, for example, says servers should have built-in redundancy.
"In a perfect world, I would like to see a server that never went down because it has redundancy throughout its design. This would mean that a component could fail and the server and operating system would be able to adapt, even if at a diminished capacity," he says. "Current server technology allows this only after a reboot, and every clustering solution has at least one scenario where the solution does not work or it is overly complex."
8. Better support for third-party hardware.
Servers are more networked than ever, and that means third-party hardware is often used to connect servers to each other and to other systems, such as storage. Users want to see vendors provide better support for third-party devices that build out server capabilities.
"Certification for a broad range of hardware is important, and in some cases we'd like to see them be more aggressive in certifying third-party hardware running with their systems," Wells Fargo's Mastre says. "For instance, while we use HP servers exclusively, we are using Xiotech for our SAN disk storage, QLogic for our host bus adapters, and Brocade for our fabric switches."
9. Lower power consumption.
Consolidation and rack-optimized servers and blades can lead to overheating and power drains. Users say they'd like to see servers become less power hungry.
"The cost of electricity and air conditioning is significant for us," says Jeff Wenger, vice president and CTO of Tax Technologies in Bradenton, Fla.
"Because we use rack-optimized equipment, we are susceptible to hot spots within our rack, and this can cause long-term problems. Clearly, this would need to be addressed before blade or other condensing technologies can succeed. Reduction in power consumption also can greatly reduce our monthly costs," he adds.
10. Standard management specifications.
Users say they want to see vendors adopt server management specifications such as the Intelligent Platform Management Interface (IPMI), which lets users manage the physical components of heterogeneous servers.
"National runs large clusters with disks being automirrored as soon as a server is swapped into the cluster," National's Seif says. "I look for uniformity and not functional differentiation. Uniform features I'd like to see adopted consistently and universally are [IPMI], [Baseboard Management Controller] and [miniBaseboard Management Controller] to give me better diagnostics and a leg up on manageability." 