IDEAS Insights

Datacenter Too Hot? Open the Windows

In the not so distant past, computers and other IT equipment were very expensive. To protect that multi-million dollar investment, everything was locked up in a ‘glass house’ with filtered air and a massive air conditioning system to keep everything running cool. Electricity for cooling was cheap and servers were very expensive. As a result, companies tended to upgrade and repair rather than replace servers. IT managers could not afford a hardware failure because that would mean downtime and the associated loss of access to business critical applications and data.

Now fast-forward to 2008. The price of electricity is rising as rapidly as the price of hardware is plunging. Most datacenters are now based on inexpensive industry-standard servers made from off-the-shelf commodity components. Even the most expensive processor only costs a few thousand dollars and devices like muffin fans can be obtained for under $20. Rather than a monolithic server, groups of smaller servers are interconnected to form a computing grid and sophisticated software moves applications around the grid as needed. Storage functions in a similar way with SANs. Instead of keeping servers and storage forever, IT equipment is replaced every five years or so with new models that are infinitely faster and more efficient.

With all of this cheap equipment that is only used for a relatively short time, why are we still cooling datacenters? Why not turn off the air conditioning, open the windows, and just let the datacenters run? That would save a huge amount of money on electricity. Warm air could be blown outside with a fan in the summer and recycled to keep office space warm in the winter. When the question was posed to multiple IT vendors around the industry, the surprising reply came back "We are looking into just that." Obviously, the issue with IT equipment is the mean time between failure (MBTF) goes up when the temperature goes up and few of us want our servers and storage device to be more prone to failure. Let’s take a look at what happens when the air conditioning is shut off. Temperatures inside the box generally climb to around 40C (104F) and stay there, and local fans keep hot components from getting too hot. The MBTF of a server, based on the combined MBTF of all components, drops from an average of seven years to five years. But if servers are only being used for five years, that’s not bad.

Of the individual components, the processors are the most robust components at high temperatures. Virtually all processors today have over-temperature monitors that will throttle back the clock or shut down cores when a temperature threshold is exceeded. Memory DIMMs are likewise quite robust in the heat, but there will be slightly more DIMM failures. Current memory fault-tolerant methods appear to be sufficient to keep servers running when a DIMM fails, and replacement DIMMs are much cheaper to purchase three years later. Power supplies can also take the heat and back-up supplies will take over transparently when one fails. Disk drives are an area that could see a drop in MBTF, although it may not as severe as one would expect. It may be possible to put the drives into a SAN and provide enhanced local cooling through something like a large refrigerator. Solid state drives may be a solution for the drives within a server. There is also the option of redesigning the disk drives to push the MBTF out to 7-10 years. Like memory failures, disk failures are not catastrophic because of RAID protection. That leaves the lowly muffin fan as the most likely component to fail within the first five year of use. With redundant fans, a simple fan failure is not likely to bring down the server.

When the numbers are finally crunched, the result shows the energy savings from turning off the air conditioning are significant and the risk to uptime is relatively small. There is a corresponding benefit to the environment as well because the electricity for cooling does not need to be generated. It may be possible to take existing servers and put a "service by date" sticker on them. When that date is reached, they either get pulled and discarded, or pulled and rebuilt with a few new components. Of course for many applications such as web serving, you could just let the servers fail in place and replace them. Even mission-critical applications could be maintained in such an environment using virtualization and automatic failover. Just think about it. Servers may only require a new set of fans at 48 months to keep them humming for their entire useful life, just like servicing your car at 60,000 miles. That’s exciting news for everyone!

Convergence, Openness, and Eco Computing Are Key to Sun’s Storage Strategy for 2008

Analyst conferences for 2008 are in full swing and one of the most recent was Sun’s analyst summit held in San Francisco earlier in February.  As with all analyst conferences, this is an opportunity for vendors to tell the research and consulting community that things are going well.  For Sun, this was an opportunity to cover all its major business offerings including servers, storage, software, and networking.   Pertaining to storage, several technologies were discussed, including virtualization as well as storage specific products such as disk arrays and tape libraries.  But the subject of this post is not so much the events of the summit, but what Sun has identified as key underlying components to its overall strategy: Convergence, Openness, and Eco Computing.  These areas are certainly not exclusive to Sun, but Sun has identified them as the fundamental elements of its plan, not only for storage, but throughout its product line.
   
Convergence is an interesting area.  It was only ten short years ago when vendors were talking about the separation of the server and storage purchases.  Where direct attached storage from a single server/storage vendor was the precedent, customers were beginning to make their purchasing decisions based more on the requirements of their specific applications.  This often meant discreet purchases of servers and storage from multiple vendors.  The benefit of this method is that customers could purchase the best technology available.  If their server vendor did not have state of the art storage capability, they could go to another vendor who could offer what was required.  This also gave customers an advantage to negotiate price, which they didn’t have previously.  But today, the trend is showing signs of convergence between servers and storage.  Some vendors have begun integrating processor and software technologies to cover both servers and storage.  Sun’s X4500 (Thumper) product is an example of a converged server/storage offering.  The theoretical advantages to the converged model are exponentially increased scalability and simplified management.

A second area that Sun has identified is openness.  Sun has differentiated itself in a very competitive market by leading the charge to provide non-proprietary, open computing platforms.   The plan is to deploy an open storage software platform with OpenSolaris and non-proprietary software such as ZFS and MySQL (of which Sun has just completed its acquisition).  The benefit to customers is that they can break away from the traditional vendor lock-in of proprietary implementations.  This should also provide a significant pricing advantage since support will be available from a wide variety of vendors as well as ISV and SI channels.  This plan for Sun is not without its issues however.  There is pending lawsuit by Network Appliance demanding that Sun cease its use and distribution of ZFS software until properly licensed from NetApp.  Stay tuned for more information on that subject.

Finally, Sun has not neglected to emphasize green computing, or as Sun refers to it, “Eco-Computing”.  Sun plans to be an active member of the eco community and provide products, services, and general information for the most energy efficient methods of running a data center.  Of course, Sun is not the only vendor to emphasize green computing – every major storage vendor has laid out product and service plans for allowing data centers to conserve precious energy supplies.
 
There is certainly a significant benefit to customers if Sun is successful with its plan and other vendors follow suit.  Traditional storage technologies cannot meet the future demand for increased data growth.  An open storage platform that is tightly integrated with server technology should provide the scalability, performance, and economics to meet high demand environments such as Web 2.0 and High Performance Computing.  Sun has certainly identified the right areas on which to focus, but as always, the key will be execution. 

IBM Teaches Its 800 Pound Gorilla to Sprint After New Workloads

Although the mainframe is acknowledged as the undisputed "800 pound gorilla" of traditional data processing, it has not been perceived as able to challenge RISC cheetahs in a race. But with the speed boost of the z10, IBM is training its heavyweight to chase after new workloads.

With great fanfare, on Tuesday IBM unveiled its latest mainframe – the z10 EC. And, for the first time in many, many years, the mainframers actually showcased technology, instead of their typical approach to avoid such discussion. This time around, System z and System p engineers jointly designed circuitry used in both the POWER6 and the z10 processor. Certainly the mainframe design retains its unique, admittedly complex, instruction set and is not built on top of RISC instructions. However, basic functions such as arithmetic units, cache, memory control, etc. use common implementations in both POWER6 and z10. Leveraging the fast clock rate design point of POWER6, IBM highlights that the quad-core z10 chip runs at 4.4 GHz, more than 2.5 times faster than the 1.7 GHz of the z9 predecessor. Of course, to achieve such clock rates, less “work” is accomplished per clock cycle. The result is that z10 processors offer about a 50% performance boost over z9. Nonetheless, operating at 4.4 GHz allows IBM to describe the z10 as state-of-the-art to reassure customers who held a perception that mainframes could not keep up with the performance of other more “modern” designs.

Let’s be clear, z10 is not going to out-perform UNIX systems on the computational and transactional workloads, and benchmarks, where RISC designs excel. Mainframes were created for “data processing” and still shine for real-world applications that access vast amounts of data and value extensive I/O bandwidth as much as computational prowess. Of course, when it comes to reliability and security, the mainframe is unmatched. But, such “traditional” computing is only growing slowly, often growing slower than the technology advances commonly identified as following Moore’s Law growth rates.

So, IBM has a major focus on attracting “today’s” workloads to the mainframe. IBM promotes its specialty engines as cost effective solutions to consolidate Linux and Java workloads. Naturally it would be expected that a vendor will select its best customer stories to present to its customers and analysts. Nonetheless, the customer testimonials were impressive, especially from Nationwide. Dissecting the customer stories to understand superior aspects of the mainframe (hardware and operating environments such as zVM) takes detailed examination that can’t be accomplished quickly. We hope to probe further to understand just where mainframes are the best choice.

Tackling Virtualization Management with Open Source

As basic virtualization functions become a standard part of infrastructure both in hardware and software, many vendors have concluded that virtualization management will offer their greatest opportunities for adding value, and they are rushing to stake out positions with a variety of offerings in this space. Virtualization management solutions are undergoing rapid innovation, and users face a growing set of product choices as a result.

At the head of the pack is VMware with its VirtualCenter platform, which plays by default as a result of VMware's leadership in the x86 virtual machine market. Microsoft, which now claims to be the fastest growing major systems management supplier overall, is in pursuit with its System Center Virtual Machine Manager (VMM). Novell also continues to build out its ZENworks management platform for managing virtual infrastructure, as shown by the recent acquisition of PlateSpin. Systems providers like IBM, HP, and Dell are extending their hardware management frameworks for virtualization, but are treading more carefully so as not to antagonize their key software partners. Finally, a profusion of startups have emerged peddling point products targeting various aspects of virtualization management, and established systems management platforms suppliers such as CA and BMC are gradually evolving to embrace the mass virtualization era as well.

Users typically have to acquire these virtualization management offerings in the manner of traditional software, licensed in some controlled fashion based on hardware footprint or some other metric. But as with any software category, virtualization management is potentially an opportunity for disruption by open source software. Until now, the most successful open source efforts have targeted relatively mature areas like operating systems and databases, and the emergence of open source-based virtualization management solutions will be an interesting opportunity to observe how open source can potentially take hold from the ground up in a major new growth area involving critical infrastructure. Two such efforts to deliver open source virtualization management have recently materialized.

Sun announced that it has released a production-ready version of xVM Ops Center, its new datacenter automation tool that is designed to simplify the management of heterogeneous IT environments. xVM Ops Center is based on a combination of Sun’s N1 System Manager infrastructure management framework with the Aduva technology that Sun acquired in 2006, which is optimized for patch and dependency management. The xVM Ops Center console is designed to automate routine system administration tasks, covering such tasks as discovery and inventory management; firmware and bare-metal server provisioning; patch management & updating; managing and monitoring system operations; and compliance reporting. xVM Ops Center supports Solaris, Red Hat Linux, or SUSE Linux operating systems, and while the initial release only works with physical servers, Sun describes the platform as a critical milestone in its xVM virtualization platform. xVM Ops Center is intended to work equally well with two of Sun’s virtualization technologies: a Solaris-based implementation of the Xen hypervisor which is about to be released, and Logical Domains (LDOMs) in Sun's CoolThreads servers. Sun is releasing the Ops Center code under the GPLv3 license, and it is available at http://www.openxvm.org/.

Red Hat recently revealed an open source virtualization project call oVirt, which includes a new web-based console to manage virtual machine platforms. The goals for oVirt are to provide a secure way for owners of virtual machines to manage their environment without granting them access to the host hardware itself. oVirt is also intended to automate virtual machine clustering, load balancing, and Service Level Agreement (SLA) maintenance; and generally simplify the management of large numbers of machines. oVirt uses a Kerberos/LDAP server for authentication and authorization, and it is tightly integrated with the freeIPA project, so that administrators will be able to authenticate, authorize, and audit their virtual resources across the enterprise. oVirt is also designed to work across platforms and architectures. The current release of oVirt is based on Red Hat’s Fedora 8 Linux distribution and the Kernel Virtual Machine (KVM) virtualization technology, but it uses the libvirt API to control virtualization functions, so it could be adapted to work with any libvirt-compatible platform, including the Xen implementation in Red Hat Enterprise Linux (note that Sun is also embracing libvirt in its virtualization platform). oVirt is currently at an earlier stage of development than Sun’s Ops Center. However, Red Hat has promised to aggressively pursue virtualization opportunities as part of its ambitious expansion plans, so expect to hear more about oVirt in the future.

As with other open source offerings, users will have free access to the code itself for these projects, and they will have the option of paying for subscriptions that provide support and a feed of updates. Consequently, these systems will potentially have lower barriers to entry than traditional software packages, since users will be able to "kick the tires" as long as they want before opting to establish a formal commercial relationship. Neither of these solutions is yet functionally competitive with heavyweight products like VMware VirtualCenter for Microsoft VMM but over time, their open source basis could help them mature rapidly, depending on the momentum that their respective development communities build.

There are many possible motivations for adopting open source software, including cost, practical benefits, and/or ideology. In some cases, users may wish to avoid entrusting their most critical IT functions to a proprietary solution. Indeed, some of the decisions that organizations make now as they start to deploy virtual infrastructure will have a long-term impact. As virtualization management policies are defined, they will become the junction at which virtualization is woven directly into business processes. While service and support are essential for such a critical system component, users may be wary of committing to a long-term dependency on a single vendor to obtain their virtualization management solution. There are few better ways to minimize the risks of such dependency than using open source software. Sun and Red Hat appeal to this reasoning, and they are betting that the open source nature of their offerings will give them a significant advantage over competitors who are using a traditional software commercialization approach.

Dell's New Global Service Offerings

Dell has retired its old "Enterprise Services" portfolio (Silver, Gold & Platinum Plus) as well as some others such as Business Helpdesk, Gold Technical Support, and Dell On Call and replaced them with a single new ProSupport offering. The announcement revolves around Dell's commercial support. Consumer support is not affected.

The announcement signals a broadening of Dell's enterprise-level support scope, a standardization of Dell's commercial services globally, and a lowering of the bar to access higher-end proactive services.

• Unlike the old Enterprise Support offerings that were focused on mid- to high-end servers and storage, ProSupport has been broadened to the desktop and even mobile products, for commercial users. This consolidation simplifies the relationships between Dell and its customers both at the contractual and support levels, where previous multiple contact points can now be rationalized.
• Dell is also using ProSupport to standardize its offerings worldwide. Prior to this announcement, users may have seen significant per-country variations in Dell's service portfolio.
• The re-packaging has allowed Dell to "lower-the-bar" to access some proactive services, previously the domain of Dell's high-end Platinum Plus offering, which was aimed at larger enterprises with around 100+ servers. With the relaxation of this minimum size "rule,"  smaller organisations can now get access to these higher-level services.

In many ways, key aspects of the services delivered (hardware maintenance, collaborative support, proactive services) haven't changed, and it is more about how Dell has changed its packaging and hence, its go- to-market strategy – including its now global consistency, which is different. This means that Dell doesn't have a major logistical task to reengineer itself in order to roll out these new products. As a result the worldwide rollout is expected to happen simultaneously with the US in EMEA and Asia Pacific. The similarity of features also means less disruption in terms of migrating clients from old to new service contracts.

Dell has announced two types of support, ProSupport for End Users and ProSupport for IT. The former is aimed at clients with little in-house expertise, while the latter is for the IT professional with a certain degree of expertise.

A brief overview of features and options for ProSupport are as follows:

ProSupport – Common Features

• Collaborative Support for Hardware and Software
• 24x7 Expert Center Phone and Online Support
• Escalation Management
• Next Business Day onsite support

ProSupport for End Users

• Virus and Spyware help
• Application Assistance (How To) on a focused group of products
• Getting Started Advice
• Remote Assistance – Dell technicians connecting directly to system to enhance the troubleshooting process.

ProSupport for IT

• FastTrack Dispatch via IT Professional Technical Training and Certification – allows qualified users to order their own replacement parts directly, bypassing Dell's telephone queues and hardware diagnosis procedures where the user is confident he/she knows what the problem is and what part is needed to fix it.

Options (applicable to either ProSupport for End User or ProSupport for IT)

• Mission Critical Option – faster onsite response and critical situation processes
• Enterprise Wide Contract – Dedicated Service Delivery Manager (SDM) , proactive planning and reporting
• Proactive Maintenance Option – Health check services
• Remote Advisory Options – remote assistance consultation including installation, configuration, and best practice recommendations
• Computrace Theft Recovery Service – Helps track and recover stolen notebooks and remotely deletes sensitive data
• CompleteCare Accidental Damage Service – Repairs or replaces accidentally damaged systems
• Keep Your Hard Drive – Enables users to control their sensitive data by retaining their hard drives

Products Included

ProSupport for End Users and IT is supported on Dell products aimed at the commercial marketplace and includes select: Dell PowerEdge, PowerVault, PowerConnect, Dell | EMC Enterprise Storage systems that are in a standard configuration, as well as OptiPlex, Precision, Latitude, Vostro computer systems and select Dell printers that are in a standard configuration.