AdvancedTCA – Green conferencing for data centers

Introduction

The complexity of the global economy has led to new business models where companies match the most efficient locations and human resources with the value adding task. Business is therefore not defined in a traditional company building as it was 20 years ago. Its boundaries are blurring, and business now includes close partners, remote experts, and employees on the road.

In this highly distributed world, team collaboration is a key success factor. Many companies rely on regular face-to-face meetings to create strong teams, but recent changes in the global economy make this approach impossible to scale and sustain. First, oil prices increased and travel became expensive. Second, the environment became an important social issue, and companies imposed travel restrictions to support their green initiatives. On top of that, a confluence of issues from terrorism to operational and financial issues among airlines to extreme weather has made travel more time consuming and unpredictable, making it less dependable and efficient.

In response, high-performance teams started to rely on remote collaboration technologies to work more effectively across distances. Audio conferencing provides the foundation for team interaction, but now an emergence of high-definition video with high-quality audio is delivering on the promise of lifelike collaboration for geographically dispersed workers.

Not only can this new communications technology decrease carbon dioxide emissions, reduce the business travel carbon footprint (Figure 1), and save oil for other uses, but it also delivers a wide range of business benefits:

• Connectivity is instant
• Productivity is not impacted by long security lines or jet lag
• Meetings do not get cancelled because of delayed or cancelled flights

Figure 1: (click graphic to zoom by 3.5x)

However, the processing-intensive conferencing equipment is a power hog, generates plenty of heat in the data center, and needs a lot of cooling. How do we select a hardware architecture that can meet the high-performance requirements for current and future conferencing and collaboration while delivering on the promise for green data center technology?

Performance requirements for conferencing and collaboration

As mentioned above, only high-quality audio and visual communications can deliver the experience required for efficient interaction of high-performance teams. Therefore, one of the key market trends is towards higher quality audio and video.

Video quality can be improved in two ways: improving compression techniques and increasing the available network bandwidth. Compression technology moved over the last 10 years from ITU-T H.261 to H.263, then to H.264; each of these standards delivered higher compression and required higher DSP performance. The DSP manufacturers responded with a new generation of chips that allows real-time encoding and decoding of higher quality video streams.

H.264 is implemented in most video endpoints today. Figure 2 shows the bandwidth required for transmitting video of certain quality compressed in H.264 format. The Y axis shows the video quality. Bandwidth requirements for transmitting video with this quality across the network is shown on the X axis.

Figure 2: (click graphic to zoom by 3.5x)

So as resolution has increased, so have the requirements for processing and bandwidth. But in the area of video quality improvement, there is no end in sight. HD 720p (1280x720 pixels) will be followed by 1080p (1920x1080 pixels) and then 4K (3840x2160 pixels). There have also been initial experiments with three-dimensional (3D) images that create an even more realistic illusion of a face-to-face meeting.

Three trends

This trend towards higher video quality means that conferencing and collaboration equipment has to be able to support higher input/output data rates, host a larger number of more powerful DSPs, and process larger data volumes.

The second major trend is that standard video calls become visual collaboration sessions that include shared multimedia content. While rudimentary content sharing allowed participants to see still images or presentations, todays multimedia content sharing includes X-ray and microscope images, full-motion video, architectural CAD drawings, digital animation, and the like.

The third trend is that visual communication is moving out of the conference room and reaching the enterprise users desktop and will next move to mobile devices. The result is that enterprises that had several dozens or several hundreds of video endpoints are now looking to scale to thousands and tens of thousands of systems. This trends leads to the requirement for much higher scalability and availability of the conferencing and collaboration equipment.

Multipoint conferencing

Point-to-point conferencing is the basic communications scenario in which two parties are connected on a call. Even more performance is required when multiple parties have to be connected through a conferencing server into a multiparty conference.

Conferencing and collaboration servers today have to support dozens and sometimes hundreds of users simultaneously and, if the trend towards desktop video continues, will need to support thousands of users in the future. In the multipoint scenario, each audio-video stream goes into the server, is processed there, and then goes out of the server. Figure 3 summarizes the multipoint conferencing and collaboration concept.

Figure 3: (click graphic to zoom by 3.5x)

The conferencing and collaboration server touches every stream and processes the audio and video for every user, so its performance directly impacts the end-user experience. It therefore must be reliable, scalable, and support high-speed input/output interfaces.

Consider this example of a realistic scenario for the not-so-distant future: if the server supports 100 simultaneous users, and each of these users is connected at 4 Mbps (3 Mbps for live HD video and another 1 Mbps for content sharing), the total bit rate going into the server is 100 x 4 Mbps = 400 Mbps. Even the highest quality audio, for example, Siren 22 Stereo, does not need more than 128 kbps, and is therefore left out of the calculation. Assuming symmetric connections (4 Mbps going back to each user), the server must support another 400 Mbps outwards.

Impact on data centers

The natural location for a conferencing and collaboration server is in the corporate or service providers data center. What is the best way to quantify the impact of these applications on the data center?

A recent survey of data center managers by Emerson Network Power indicated that power and cooling are the most critical resources in data centers today[1]. The survey showed that more than 96 percent of the data centers will run out of power and cooling capacity by 2011. Additionally, 40 percent of the respondents said that even today their main problems are around power and cooling. The steep increased power consumption in data centers was noticed even on the federal level. In July 2006, the House of Representatives passed H.R. 5646, which calls for the EPA's Energy Star program to conduct a study on power use in data centers. The bill refers to the current spending of $3.3 billion per year for power in data centers.

Looking at power consumption is the best way to estimate the impact of conferencing and collaboration on the data center. Lets assume an 8U server chassis with 14 blades. Based on current technology, one blade usually draws ≤ 200 W. 14 blades therefore need 2800 W (2.8 kW) or the equivalent of two standard household irons. Now lets assume five of these chassis are stacked in a rack-mount or server rack arrangment; the result is 2.8 kW * 5 = 14 kW (equivalent to 10 household irons). Then we will need to modify that by the number of racks in the room. But it gets worse! The power necessary for powering up the equipment is only 30 percent of the power necessary to run a data center. Figure 4 depicts the power consumption split.

Figure 4: (click graphic to zoom by 3.5x)

The IT equipment in the data center consumes only 30 percent of the power. Another 42 percent is used for climate control: 33 percent for chiller (chillers feed cold liquid to the data center) and 9 percent for air conditioning. From the remaining 28 percent, 18 percent is used by Uninterruptible Power Supply (UPS), and 5 percent by Power Distribution Units (PDU). The data center needs therefore more than three times the energy used by the servers themselves; consequently, any reduction of the power consumption in servers leads to 3x power savings in the data center.

Blade servers and power efficiency

Blade servers have been around for a while but did not deliver serious power savings over regular rack mountable 1U servers in the past – up until 2003-2004. Only recently have they reached the technical maturity that allows them to clearly surpass other server technologies with regards to power use efficiency.

Recent studies show that blade servers allow four-fold increase of the processor density and for 20-30 percent decrease of power consumption. This is a very new technology in the beginning of its adoption curve. According to Gartner Dataquest, 2006 sales of blade servers were 850,000 units or 10 percent of the server market. By 2011, Gartner expects the sales to grow to 2.3M units or 22 percent of the server market. AdvancedTCA, as the only standard for blade servers. will be a substantial part of this growth. Figure 5 shows the Polycom Media Processing Module (MPM), an AdvancedTCA blade that is inserted into the Polycom RMX 2000 conferencing server.

Figure 5: (click graphic to zoom by 3.5x)

The technology improvements in blade servers allow for less power/Hz than standalone servers. The use of a DC power option provides the greatest savings because data centers usually have large-scale DC power distribution throughout. There is no need for conversion from 110/220V AC to 48 V DC in each chassis, and this increases power use efficiency. Consolidation at the rack level provides higher cooling efficiency, compared to cooling separate 1U servers with the equivalent of one blade. Other AdvancedTCA power management features include:

• Dynamic optimization of airflow and power consumption
• The ability to stay within a predetermined power budget
• The ability to monitor temperature levels and energy use at blade, enclosure, and rack levels.

All of the factors just noted assure continuous operation of the AdvancedTCA based system under optimum power consumption.

AdvancedTCA versus closed box architecture

In the video communications space, there are still some attempts to implement high-performance servers in a proprietary closed box. While there is no argument that this approach is less efficient from power saving perspective, we have to also consider the impact on the environment when rapid advancements in DSP technology lead to rapid product obsolescence. The result is what is referred to as a forklift upgrade, and that has very high true cost to customers and society as a whole. Table 1 compares the two hardware architectures.

Figure 6: (click graphic to zoom by 3.1x)

Lets look first at the impact on the customers business. In addition to the cost of the new server, customers experience disruption in service. Customers have to assign employees to assist the installation team and grant access to the facility. Configuration has to be copied from the old server to the new one.

The closed box approach is also extremely damaging to the environment. As soon as the DSP technology changes and the server models designed for higher resolutions and speed hit the market, the older servers become obsolete and have to be replaced. Recycling these boxes is difficult – everything including power supplies, backplanes, and fans has to go.

However adding a blade with new functions into the AdvancedTCA chassis does not interrupt operations, thanks to AdvancedTCA support of hot swappable blades. In addition, all components that rarely change and are built into the chassis – power supplies, fans, and the like – continue to be used after the blade has been replaced.

Conclusion

When designing equipment for the data center, vendors should select hardware architectures not only based on technical requirements but also based on requirements for reduced power consumption, upgradeability, and other factors limiting the impact on the environment.

Polycom has taken steps in this direction through the introduction of AdvancedTCA hardware architecture in its product line. AdvancdTCA allows for substantial power savings in the data center, simplifies upgrades, and reduces the need for recycling.

Stefan Karapetkov is Emerging Technologies Director at Polycom, Inc., where he focuses on visual communications market and technology analysis. Prior to Polycom, he spent more than 10 years with Siemens in product management, new technology development, and enterprise product definition. He has been involved in Voice over IP since 1997. Stefans MBA is from Santa Clara University and he holds an MS degree in Engineering from the University of Chemnitz (Germany). He is fluent in English, German, Russian, and Bulgarian.