Will HEVC/H.265 Kill the Data Center?
The advent of HEVC (aka H.265) has generated equal parts hype and promise, but one company is questioning whether the use of HEVC in the near term will kill the data center.
QuickFire Networks, which makes transcoding platforms for integration with third-party platforms, sees the uptick in mobile video demand—especially around H.265—as a potential "murderer of data centers" when it comes to required computational capacity.
"We think mobile video is the killer app," said Mike Coward, QuickFire's CEO, "but we think it comes at the expensive cost of having to build more data centers. We see a 50 percent bandwidth reduction possible with H.265, but a 1,000 percent increase in the computational needs to achieve that bandwidth savings."
Coward used an infographic featuring the company's bold claiim (click thumbnail below to view the full infographic) to illustrate his point: Consumers will consume more videos, which drives engineers to suggest scaled-up computational requirements, which in turn puts chief technology and financial officers in the position of having to scramble to find data center space.
Click image above to see full-size infographic
"According to some estimates, current data center capacity is oversold," said Coward, "and there are even projections that we'll run out of current data center capacity in the U.S. within the next year."
While Coward admits the 1,000 percent computational increase requirement is only for a few scenarios, he points back to the fact that H.264 is baked into silicon while H.265 is not yet.
"When we worked with the reference HEVC encoder, we found that it took an average latency of 20-30 seconds for a single frame of video to be transcoded," said Coward. "While we know that will be optimized over time, and the broadcast world may delay the use of HEVC for a number of years, we think data centers will face a bottleneck much sooner, perhaps within the next 9 months."
Coward's point is well taken: Having worked on MPEG-1, MPEG-2, and MPEG-4 telco-grade solutions in the past, I've seen firsthand that there was always a window of time between the standard being stabilized in software before it's committed to silicon.
That window could last a year or more—and QuickFire is betting it will take longer, so the company's approach is to provide a denser computational solution platform on which existing transcoding software solutions can ride.
Call the T-Video Transcoding Platform V1100, Quickfire's single-RU box consists of multiple Ethernet connectors (3 x 1 GigE and 2 x 10 GigE SPF) and eleven quad-core Core i7 mobile CPUs.
"We use 11 of Intel's Core i7 3615QM CPUs," said Coward, "each with an integrated graphics processor unit. The combination of the quad-core CPU and the integrated GPU lets our platform dramatically decrease one of the big factors in data center cost: power consumption."
While we've not been able to yet verify the claim, QuickFire says its box can do 100 ingests via IP, simultaneously outputting up to 100 1080p outputs.
"We can bring the power consumption down from 75 watts per transcode to 7 watts per transcode," he said.
Coward says that QuickFire has focused the T-video solution on being a platform solution.
"We're not selling the V1100 as a pre-packaged encoder," said Coward. "Companies can put their own software on the box."
Coward agreed when I asked if the approach was more like that of Network Engines, which would create a separate SKU per pre-packaged build for each of its customers. In that way, the competition for QuickFire isn't around the Elementals or other purpose-built transcoders, but more a competition with Dell, HP, and other blade server manufacturers.
Vantrix is one company to take advantage of the QuickFire solution, as witnessed by a joint press release the companies published at April's NAB show.
Coward also said the power of the T-video solution is attracting file-based transcoding clients.
"We designed T-Video be a real-time IP-IP transcoder for format conversion and rate adaptation," he said, "but the density we achieved is so high that we're also winning customers for offline file-based transcoding."
In light of the file-based transcoding potential, Coward says the V1100 has both shared storage and private storage, the latter being dedicated to a single CPU.
"The V1100 has 4TB of shared storage and also has 480GB of private storage for each CPU," he said. "This private storage is intended for operating system, application, scratch files, and local storage, while the shared storage is mainly for shared media content."
"All told, we can do around 9TB (480GB x 11 + 4 TB shared) in this 1U appliance, all with SSD," he added. "This brings the cost per stream to about one-fifth of Xeon-based approaches."
Quickfire Networks is showcasing its T-video V1100 this week at The Cable Show 2013 in Washington, DC.
Company teamed with MulticoreWare to help develop and promote an H.265/HEVC codec, building on the success of the x264 codec, and already claims encouraging data rate reductions for encoding
Think HEVC will be in place in the near future? Think again. Here are three myths about the emerging standard and the sobering reality for each.
What makes the HEVC codec the future of online video? At Streaming Media East, Elemental explained it's the bandwidth savings that matter.