IDEAS Insights

Performance & Energy Shootout – SSDs vs. Traditional Disks

We all know the performance superiority of solid state drives compared to traditional spinning disk technologies, but how does their energy usage compare? With the recent release of a new storage benchmark from the Storage Performance Council (SPC) that looks at not only performance, but real world energy use, and the release of the first two tests, we have a chance to find out.

The first two results under the SPC’s SPC1-C/E benchmark were released along with the announcement of the new benchmark. One was a Seagate 300GB 10k rpm SAS drive subsystem and the other was an IBM 69GB Solid State Disk SAS configuration. As luck would have it, one result uses traditional rotating disk drives and the other uses solid state drive technology. This gives us an initial insight into the power consumption characteristics of these very different technology types, along with the more traditional price and performance aspects.

The above chart is analysis by IDEAS based on data supplied by the Storage Performance Council (SPC) in the two Executive Summaries of the two test results so far. Note that the segment for each bar with a green label is better than the red label. e.g. The lower cost per GB is better, even though it is the smaller segment of that bar on the chart.

Bearing in mind that the two subsystems have significant differences, such as the total addressable capacity for the IBM offering is just 548GB in contrast to the 3,600GB of the Seagate solution, an analysis of these first two results is interesting all the same.

• The outcomes certainly back up the perception of the speed of solid state technology over traditional spinning disk.
• But as the Cost per IOPS shows, the relatively higher cost of SSD technology means that the price performance of the two is on par.
• And if capacity rather than performance is your key requirement, then from an energy use perspective at least, spinning disk technologies would appear to still be in play.

The SPC1-C/E benchmark and the current results have now been added to the Competitive Profiles CPStorage module and as new results are released they will be added to the coverage. The full set of results can also be found on the SPC-1C and SPC-1C/E Benchmark Results page of the SPC web site.

The SPC is a non-profit corporation founded to define, standardize and promote storage benchmarks and to disseminate objective, verifiable storage performance data to the computer industry and its customers. The organization's strategic objectives are to empower storage vendors to build better products as well as to stimulate the IT community to more rapidly trust and deploy multi-vendor storage technology.

A more detailed analysis of these first two results will be featured in the July edition of Tech Trends Monthly, a free on-line newsletter produced by Ideas International.

VMware's vSphere Pricing is Attractive at Entry, but Premium Looms at High End

VMware's new virtualization platform, vSphere 4, introduces some significant changes in the way the company prices and packages its software, compared to the previous version Virtual Infrastructure 3 (VI3). To determine the impact on of VMware's revised pricing on various classes of servers, IDEAS calculated the equivalent cost per processor for deploying VI3 and vSphere 4 on various hardware configurations (see details in the June edition of our Tech Trends Monthly newsletter). The analysis found that VMware's pricing for vSphere 4 will be attractive for entry users in SMB environments, but that VMware has increased its premium for users at the high-end who are likely to exploit its most advanced features.

vSphere 4 Standard, the entry version of VMware's virtualization platform for single servers, is clearly similar in features to VI3 Standard, and it is a lot cheaper in all three configurations. With vSphere 4 Standard, the two- and four-socket configurations are 47% cheaper than VI3 Standard, while the one-socket configuration is 73% cheaper. The new vSphere Essentials offerings (see datasheet) create a new entry price point for SMBs users with up to three servers. These users can obtain the same level of functionality as VMware's previous entry bundle, VI3 Foundation, for around one third of the cost.

At the high end, though, users will ultimately pay more to get the most advanced level of functionality. vSphere 4 Enterprise is priced the same as VI3 Enterprise at two- and four-socket configurations, and costs 50% less on one-socket configurations, while including many new features. However, VMware will discontinue vSphere 4 Enterprise after December 15, 2009, at which point users will have to buy vSphere 4 Enterprise Plus. vSphere 4 Enterprise Plus costs 21% more than vSphere 4 Enterprise, for which users get VMware's most advanced features, including VMware vNetwork Distributed Switch and VMware Host Profiles.

Further, some of VMware's new packages vary in terms of the number of cores per processor socket that they can be deployed on. While most versions of vSphere can be deployed on processors that contain up to six cores per socket, vSphere 4 Advanced and vSphere Enterprise Plus editions will be required for deployment on processors that contain up to 12 cores per socket. Most 4-socket servers today have 4 cores per socket, and Intel and AMD have each announced processors with 6 cores per socket. However, Intel has pre-announced processors with 8 cores per socket, which are expected in late 2009, while AMD has previewed processors with 12 cores per socket, and it is promising 16-core processors in the future. When systems with these processors become available, they will require the installation of vSphere Advanced or vSphere Enterprise Plus.

VMware's pricing has come under some pressure lately. Microsoft includes virtualization with its Windows Server 2008 operating system, and Citrix recently starting allowing use of its virtualization functions for free. With its vSphere pricing, VMware has clearly responded to these pressures. In an effort to blunt the appeal of bundled and low-cost alternatives, VMware has reduced the premium for users who may just be getting started with virtualization, and have limited funds to invest in virtual machine software. At the same time, VMware is betting that enterprise users who value VMware's high-end functions will not be as price-sensitive, and are willing to pay more to step up to the next level of VMware's vision for virtual infrastructure.

Keeping Afloat or Keeping Abreast – The SMB Dilemma

Recently I attended a briefing held by Dell Australia that focused on services and small to medium businesses and the impact of the economic downturn. Dell in Australia defines a small to medium business (SMB) being from 3 to 500 employees. Dell sent out a survey asking them about the challenges they face as a result of the current economic conditions. The survey consisted of responses from about 200 organizations with 80% of those being companies with less than 100 employees.

The respondents said that the most significant challenges to their businesses were seen as maintaining the quality of customer service, retaining and growing customers, keeping revenues stable and cash flow. The things that are less important include financing business expansion and keeping up with current trends. It is this last point that I found of most interest.

The Dell survey revealed that the overwhelming majority (90%) of people responsible for IT operations and purchasing decisions are company employees. The survey didn’t say whether these were dedicated or part-time resources, but it is probably fair to assume that many are people doing this part-time. The IT marketplace is full of technologies and acronyms and can be very daunting to someone who has only a part-time interest in the subject, whilst they concentrate on running a business. It would seem that the sheer complexity of the issues and topics is an inhibitor to innovation for smaller companies.

But you don’t need to be a subject matter expert to make savings!

There have been great strides made in server performance and the reduction in power consumption, combined with the compaction of the processing footprint with the introduction of multi-core processors allows you to put more grunt into smaller packages. So, the simple act of replacing a server just a few years old, one that is just about to run out of warranty for example, can have a big impact on your costs. It’s the same logic that we all use with mobile (cell) phones. So why treat servers any different? In fact, imagine if your phone bill would go down when you got a new phone too. With servers that is what can happen, only your running cost reductions are not airtime minutes, but power consumption. Food for thought.

IDEAS offers the Server CAR tool to help users better quantify savings they can make when looking to swap out old servers for new. The Server CAR was also recently awarded the AIIA’s iAWARDS 2009 New South Wales State Winner in the Sustainability and Green IT category.

 

Gary Burgess
Senior Vice President, Research & Operations

 Follow Gary on Twitter

IBM's Stealth Launch of POWER6+

IBM has historically updated some part of its product line, hardware or software, in the Spring and Fall seasons, keeping the "new and improved" banner fluttering to generate opportunities for its feet-on-the-street sales force. IBM's 2009 April update (Spring here in the US, but Fall for our Australian co-workers) included a number of new features, many of them significant, announced under IBM's Dynamic Infrastructure umbrella. Hidden among the multiple coordinated announcements (most of which will be reflected in IDEAS' Competitive Profiles) was the sheepish acknowledgement that POWER6+ has arrived. 

Intel's "tick-tock" schedule of introducing chips formalized what IBM had already been practicing – it is less risky if fabrication technology advances and processor logic-design advances are separated into independent chip releases.  But first, forgive some reminiscence. 

In the old days, when I was a mainframe processor designer, too often management's "scheduling of invention" unsuccessfully bet that breakthrough fabrication advances would conveniently coincide with innovative processor design patents. Back then, the repercussion of misaligned fabrication/design advances left the customer to suffer the delays. Since there were few serious alternatives, mainframe customers were stuck with IBM-brand systems and the delayed introduction of new servers typically resulted in a frenzied "first-day-order" lottery, along with proverbial rumors of bodegas winning early ship dates for multi-million dollar behemoths and profiting immensely by selling their spot in line. 

In the RISC space, starting with POWER4 (and perhaps even before), IBM instituted a more disciplined approach, which Intel later popularized. The milestones:  

• POWER4 (new dual-core design at 180nm),
• POWER4+ (incremental design with faster clocks, 130nm),
• POWER5 (new design with SMT, still 130nm),
• POWER5+ (incremental design with faster clocks, 90nm).  

Then came POWER6. To deliver its leapfrog competitive performance, at least on a per-core basis, POWER6 not only incorporated a new logic design, but also moved to 65nm. That upped the ante for POWER6+. To keep up with Intel, POWER6+ needed to be delivered at 45nm or smaller, and incorporate at least 4 cores per die.  However, IBM seems only to have added some incremental design enhancements (better storage keys, etc.).

Apparently, POWER6+ slipped out, unheralded, in the second half of last year with the Power 560 and Power 570 enhancements.  Now with the updated Power 520 and Power 550 employing POWER6+, IBM has reluctantly admitted that POWER6+ is not the headline-grabber it had once hoped.  (Acknowledgement -- Timothy Prickett Morgan uncovered this a few days before the rest of us.) 

In its heyday, Digital Equipment Corp was infamous for its stealth marketing (parodied with the oft-repeated analogy that Digital's attempt to market sushi would be as "cold, dead fish.")  But while Digital often couldn't figure out how to properly market some of its wonderful technology, IBM has usually performed superbly in marketing its own technological breakthroughs. Could it be that POWER6+ didn’t quite live up to IBM’s dreams, leaving IBM for once with the task of downplaying its introduction? 

New Release of HP-UX Demonstrates Appeal of UNIX Reliability

While most of the activity in the IT market seems to be driven by industry-standard platforms based on x86 hardware running Windows and Linux, servers running the more mature and proven UNIX operating system remain the workhorses for business critical enterprise computing. HP's Integrity server platform recently got a boost with enhancements to the HP-UX operating system, and HP’s Serviceguard High Availability (HA) clustering package.

The new release of HP-UX, called HP-UX 11i v3 Update 4, includes improvements to the process of upgrading the OS, support for erasing sensitive data, and improved usage of power optimization functions. Upgrades from earlier versions of HP-UX can now be performed with about half the downtime as required before. The HP-UX Online Operating System Update now upgrades system software entirely online, with downtime incurred only for the reboot itself. The new release also introduces a secure disk erase tool that is compliant with Department of Defense specifications. Update 4 also takes better advantage of the power management functions in the Itanium processor on which HP’s Integrity servers are based. While earlier versions of HP-UX were able to achieve up to 10% power savings on idle processors, the new release can achieve 8% power savings on active processors by optimizing their power consumption based on application workloads.

The new release of Serviceguard Solutions, HP Serviceguard Solutions for A.11.19, introduces 100% online reconfiguration, 83% faster failover that does not decrease as the cluster grows, and integrated workload balancing to ensure predictable performance after a failure. Most common administrative actions in Serviceguard can now be performed while the clustered applications continue running. Serviceguard integrates workload balancing to ensure that services are moved to the appropriate nodes during a failover without incurring any performance degradation. The approach makes sure servers do not become overloaded after a failover by allowing administrators to define a "weight" for each clustered package, and a "capacity" that each cluster node can carry. When a failure occurs, Serviceguard will automatically move affected packages to nodes that can accommodate their respective weight. This new workload balancing functionality also is fully integrated with the new enhanced package dependencies in Serviceguard A.11.19. Using a graphical display in Serviceguard Manager, administrators can preview how each package depends on others so that they can correctly control potentially complex dependencies between applications, without scripting, saving up to 25% on cluster administration time and costs.

Other enhancements in the new release of Serviceguard include the following:

• The new version can restore clustered services 83 percent faster than previous versions, with failover time reduced from 30 seconds to 4 seconds.
• A new feature of the Serviceguard Extension for SAP called HP Hot Standby liveCache reduces the recovery time for SAP liveCache from hours to 2 minutes.
• Serviceguard is now fully compliant with IPv6, as well as mixed IPv4 and IPv6 environments, which is required by certain government agencies.
• HP offers several new Serviceguard Toolkits for databases, including IBM DB2, MySQL, and Sybase ASE. 
• HP now offers a lower cost licensing option for Metrocluster for RAC disaster recovery, with the ability to license the package either per core or per cluster, which will appeal particularly to smaller organizations. Additionally, SONET/SDH site interconnects offer a less expensive alternative to DWDM, and multiple subnets within a single cluster spanning data centers provide simpler clusters that cost less to manage.

HP was one of the first major systems vendors to adopt UNIX as its strategic business operating system, and it became a leader at driving UNIX solutions into commercial data-processing environments. Derived from a code-base that first began shipping in the early 1980s, much of HP-UX’s reputation in the market rests on stability, familiarity, and well-tested innovation. Complementing HP-UX (and since 2001, Linux), HP’s Serviceguard is recognized as one of the most proven High Availability and Disaster Tolerant solutions in the industry, with hundreds of thousands of licenses sold worldwide to date. Since introducing HP-UX, HP has followed a path of enhancing its UNIX system in a conservative and highly integrated manner, rather than rushing to add exotic new functions to the UNIX domain. Though still feature rich, as part of this approach, HP has optimized HP-UX primarily for criteria such as stability and investment protection, emphasizing business-oriented factors such as quality and support.

HP now issues major updates to HP-UX every 3-4 years, and minor updates such as this one approximately every six months. With the latest release, HP targeted three design objectives: enabling rapid recovery from conditions causing downtime, delivering a quick return on IT investments, and simplifying operations. These objectives accurately reflect some of the most pressing concerns that customers have today. Simplifying operations and active processor power savings can be essential to reducing current costs in the current economic climate, while the new Serviceguard Toolkits and improved licensing terms will help customers respond more rapidly and cost-effectively to emerging business opportunities once the economy starts to recover.

Moreover, the uptime improvements delivered by the updates to HP-UX and Serviceguard will be particularly appealing to UNIX users, for whom reliability has become one of the most valued operating characteristics that the UNIX platform provides. Based on the average weightings collected from 35 validated end users visiting the IDEAS Collaborative Product Evaluation (CPE) on UNIX Operating System functions in 2008-09, reliability functions carried the greatest importance of the five major functional categories in the evaluation (see Figure 1).

Figure 1: Average priority of functional capabilities in UNIX operating systems
Source: IDEAS Collaborative Product Evaluation of UNIX Operating Systems - http://ideasint.eval.com/unixos

The decision to deploy any operating system usually starts with the requirements of an application. In this context, there are many reasons to deploy operating systems such as Windows and Linux. However, for workloads that require the highest levels of reliability, users continue to be drawn to UNIX systems. With the latest release of its HP-UX platform, HP has directly responded to that demand.