V-Nova: PERSEUS lowers OTT network storage costs and cuts number of ABR profiles

By Barry Flynn, Contributing Editor

Most of the focus on V-Nova’s PERSEUS video compression platform since it came out of stealth mode last April has been on its ability to deliver significant efficiency increases in the bandwidth required for contribution and distribution when compared to existing MPEG-2, MPEG-4 and HEVC technologies.

However, points out Fabio Murra, the company’s SVP of Product and Marketing for TV and Media, PERSEUS’s ability to cut OTT operators’ network delivery storage costs is equally noteworthy. “We get [storage] applications in which we see a performance increase around 40-50%, perhaps even more,” he claims.

The main reason V-Nova is able to achieve such benefits is down to two factors, explains Murra: the multi-stage nature of CDN-mediated OTT video delivery, and the widespread use of Adaptive Bit-rate (ABR) streaming.

A typical client use-case will, in simplified form, involve:

1.  A PERSEUS software plugin operating in tandem with an existing video encoder to compress a video file and deliver it to an origin server.
2. The origin server sending the file to a Content Distribution Network (CDN).
3. The CDN delivering the file to an ISP.
4. The ISP delivering the file to an edge-server.
5. The edge-server delivering it to a consumer device (containing a decoder equipped with a PERSEUS software plugin to decode the video file).

Murra points out that copies of the video file in question would generally need to be stored or cached once for each ‘hop’ in this delivery chain.

Since the addition of PERSEUS to the encoding mix means that it can compress the video file to a smaller size than existing compression systems, the storage requirement is thereby reduced – not only at the origin server and consumer device ends of the delivery chain, but at each and every storage point in between, creating a multiplier effect.

The potential efficiencies do not end there, notes Murra.

ABR technologies work by creating multiple copies of the master video file using a hierarchical set of quality profiles, and dividing each into chunks. Then, if the network gets congested, a lower-quality chunk or set of chunks can be temporarily served to the consumer’s device until the traffic crunch passes, at which point the stream switches back to a higher-quality profile. This minimises the risk of the viewer’s experience being interrupted.

In the ABR version of the use-case cited above, what would therefore usually be happening would be that the PERSEUS-equipped encoder would be sending not just one, but a multiplicity of different-sized files to the origin server for initial storage and beyond, each relating to a different ABR quality profile.

“What we’ve seen with PERSEUS is that by improving the quality, by reducing the bit-rate, you can also reduce the number of profiles,” Murra says. Illustratively, he suggests, an operator using five different profiles for its ABR streams would, with PERSEUS, be able to reduce that number to just three.

“What we are saying is that with our system you can reduce the size of the files to the point where the difference between them is not that great,” says Murra.

Fewer profiles means fewer files need to be cached across the network and at its  edge. “Even if you have a great connection and you get the top profile, all the other profiles underneath still get cached at the edge of your network, because that connectivity can go up and down. So even if you just go from five down to three profiles, that’s massive savings in the storage on the network, as well as the actual delivery bit-rate.”

The benefits depend in part on the number of ABR profiles an OTT operator already has running. “We’ve seen people with over 20 profiles – it depends on the business of the operator. If you target connected devices, mobile devices, you get one set of profiles, if you go to connected TVs you need a different set.”

The savings also depend on how much ‘correlation’ is present in the video being compressed (PERSEUS leverages the elements in a picture that ‘correlate’ with each other to avoid data repetition instead of traditional block-based compression approaches which throw away redundant information (see here for a more detailed account)).

“We’ve processed data with high correlation, that we can encode 70% better than [other] compression methods,” says Murra, “but that’s a specific case. You can find corner cases, where we’re [only] 0% better, i.e. black or white noise – but those are very corner cases, they do not represent TV in general.”