Octasic Moves Transcoding to the Core
A company with a deep background in voice-over-IP and media gateways has moved into the PCI-E boardset market, taking aim at x86 CPU-based transcoding solutions.
"Our single PCI-E board is capable of transcoding 355 channels of QCIF at 50 frames per second," says John Fry, director of sales and business development at Octasic, a Montreal-based company known for its digital signal processing (DSP) solutions.
I spoke with Fry on Monday, asking questions about the company's history, the new TPX1000 boardset, and the software stack the company makes available to developers looking to create high-density transcoding solutions.
"Our API doesn't expose the chip itself," Fry says, "but a significant amount of 'canned functionality' is offered to our customer, with all the audio and video transcoding functionality, so that they can, in turn, offload commodity transcoding to our boards and get on with the true differentiator of value-added services for their ultimate customer."
Fry says that Octasic sees transcoding as a commodity that many wireless providers need to check off their list of functionality, but that transcoding itself is not the true value-added service.
"We see a slew of applications for mobile video viewing," says Fry. "When we started looking at moving into this market, we quickly realized our customers don't see the value-add out of transcoding, but do see a value-add in creating order out of the mess of containerization."
The Mobile Value Proposition
A typical customer might, according to Fry, offer transcoding as a service to a wireless provider, with the end game being a move from an SaaS model to one that implements transcoding at the network core.
"Our strategy is that we can take off your hands the commodity processing," says Fry, "since our boards are capable of handling a sizable amount of video transcoding."
The API itself, given the board's design, looks more like a network interface card (NIC) with an embedded Gigabit Ethernet switch.
"The plug-in card looks like a bump in the wire," says Fry, referring to services that are transparent to the network itself. "In actuality, since everything is delivered via IP, transcoded, and then sent back out as IP, it's easiest to say the TPX1000 is a NIC card that also does transcoding."
Fry notes a typical customer in Aylus Networks, a company that has begun to use the TPX1000 boards in stackable 2U servers.
"Aylus is selling value-add services to wireless carriers," Fry says. "We're doing the real-time transcoding and Aylus is using that to sell generic services, such as video chat, or specialized services such as audio-note tagging tied to cell phone video clips for insurance claims adjusters. Wireless carriers can then charge a premium for these value-added services."
In addition to the stackable servers, Aylus is also offering a chassis with capacity for multiple server blades, which Fry says are robust IBM Blade Center blade and chassis combinations.
"Our offerings are built from the ground up to support mobile video services and allow the carrier to not only efficiently cope with the current tidal wave of data but to monetize this data as well," said Aylus Networks CEO Shamim Naqvi in a previous press release highlighting the configurable chassis. "Aylus facilitates the smooth transition from a voice-based to a data-based mobile economy, and the rapid deployment of revenue-producing broadband services."
The Octasic DSPs can resize video in real-time, adapting video streams to both internet and mobile device protocols as well as creating the right resolution and bandwidth for a particular mobile device's screen and playback capabilities.
More Bang for the Buck
The press release announcing the Octasic TPX1000 PCI-E board claimed that the new boards provide a way to "can cut video transcoding systems capital costs in half, reduce the power consumption by a factor of 8 and still deliver 5 times more channels per unit of rack space."
This kind of claim, and others like it, are why I have invited a number of the leading DSP, GPU and x86 transcoding solutions to undergo a set of comparative tests starting later this week, using a test methodology from Transitions, Inc., a consultancy I co-founded several years ago.
In order to better understand Octasic's claims in the meantime, though, I asked Fry to provide details on how the company makes such claims.
"The price point for one of our boards is $8,000 for a 335-channel QCIF card that's purchased as a one-off," says Fry. "We priced the boards at $4,500 each for quantities of more than 100 boards.
"By comparison, if you use move to a 3 Ghz quad core Xeon x86 solution, using dual processors for 8 total cores, the average price is around $6,000 on the Dell site," said Fry, stating Octasic had verified this type of solution was used by Envivio and other x86-based transcoding solutions. "For that amount of processing power, you gain about 150 channels of QCIF resolution transcoding. Stack 10 of these 1U servers on top of each other and you've got the ability to do 1500 channels at approximately $60,000 in hardware costs."
The TPX1000 requires a 2U server form factor, but can run with a much slower 1.5 Ghz processor, according to Fry.
"We can stack four PCI-E cards in a 2U server, and that gives us roughly 1300 channels of video in a 2U box," says Fry. "That puts us at about a $20,000 solution."
"Even if we dialed back to 250 channels per board," Fry continued, "there would be significant space savings: 4 rack units to accomplish our solution versus 10 rack units for the x86 solution."
The company also claims that the use of the low-powered clockless DSP at the heart of the TPX1000, alongside the lower-power 1.5 Ghz CPU on board its typical server, means that the company can save approximately 62% power consumption without sacrificing any channel density.
"Put another way," a company white paper claims, "compared to typical server class dual quad core Xeon setup, our solution can very easily provide a ~15x gain in channel density without increasing power consumption."
One question Fry had for me, as the transcoding testing gets underway, is why CPU-based solutions don't drop back to slower processor speeds to save both time and power consumption.
"If you've got a 2 Ghz server you're at roughly $2600," says Fry. "With 2 Ghz server, you'd lose about 1/3 of the throughput, but also drop to about 1/3 of the cost."
I think I heard a request to add another data point to the Transitions test criteria. So noted.
New OCT2200 series capable of two 1080p HD transcodes without additional hardware assistance