At SEEN, we’re within the personalised video enterprise.
We serve a variety of shoppers who have to render and stream a excessive quantity of distinctive, excessive definition movies that leverage information factors — like identify, gender, buy historical past, and so forth. — to talk on to their prospects and constituents.
For a few of these campaigns, we have to render and stream tons of of 1000’s — or tens of millions — of movies in just some days, and even in just some hours. To take action, we leverage an adaptable, scalable, and environment friendly cloud-based structure constructed on Google Cloud and Google Kubernetes Engine (GKE).
On this weblog, we’ll stroll you thru:
Why we would have liked to switch our legacy naked steel structure with a brand new cloud-based structure constructed on Google Cloud and GKE
Why we selected Google Cloud as our cloud companies companion
How Google Cloud helped us design and implement our new structure, and what it appears to be like like
The quantitative and qualitative advantages we’ve skilled leveraging Google Cloud and GKE
Let’s dig in.
Our problem: Rendering personalised movies at velocity and scale
Each startup wants an formidable purpose.
At SEEN, we wish to render and stream tens of millions of particular person personalised movies to tens of millions of particular person individuals in just some seconds.
This purpose is formidable, but it surely isn’t simply one other “moonshot” designed to sound spectacular — reaching this purpose is essential to our firm’s development and future.
Right here’s why.
Once we first launched SEEN, we attracted smaller shoppers who wanted to render and stream a comparatively low quantity of personalised movies.
To serve these shoppers, we constructed a fundamental structure based mostly on colocated machines. Our system labored — it might render and stream 1000’s of cinema-quality personalised movies despatched over the course of some weeks— but it surely was extremely guide and time consuming to function, and it slowed us down and didn’t scale very properly.
This legacy structure grew to become an actual downside as we grew and started to draw bigger firms with bigger initiatives. To serve these new shoppers we would have liked to generate and stream many extra personalised movies at a a lot quicker charge. For instance, one consumer wanted 1.eight million movies in just a few days, whereas one other — a health app — wanted 100 million movies in someday.
Tasks at that velocity and scale have been unimaginable to think about with our legacy structure, however to develop SEEN to the following degree we would have liked to discover a option to seize these enterprise shoppers and ship on a lot of these initiatives. To try this, we would have liked to rebuild our previous system from the bottom up with a contemporary, environment friendly, and scalable cloud-based structure.
This is how we did it.
Trying to find a brand new companion: How (and why) we selected Google Cloud
As quickly as we realized we would have liked to rebuild our basic structure we started to seek for a cloud companies supplier. We knew that discovering the best companion — with the best merchandise and help — might resolve lots of issues for us and speed up the method of designing, constructing, and deploying our new video rendering system.
To begin our search, we drew up a listing of what we have been on the lookout for in a really perfect cloud companies companion. Our record included:
The fundamentals, like sustainability, safety, and a neighborhood presence in Europe to assist mitigate GDPR and different considerations (most of our shoppers have been in Europe on the time).
Subsequent-gen like NVIDIA GPUs, Kubernetes merchandise, and rising technical options like GPU time sharing and with extra NVIDIA merchandise on the roadmap.
White-glove help with direct collaboration and steering to assist us resolve each normal and area of interest issues in our system.
This final level was key to deciding which service supplier we might select. On the time we had a small-but-mighty group of 4 builders. Whereas they’d vital experience with the cloud and rendering engines, on the time they simply didn’t have the bandwidth essential to carry our imaginative and prescient to life.
With these necessities in hand, we reached out to cloud companies suppliers, together with Google Cloud.
We met with different suppliers, however Google Cloud felt totally different from the beginning. We met with them in-person, the place we held productive conversations about our present legacy structure. We offered our new structure’s design, and so they offered insights on how one in every of their options — Google Kubernetes Engine (GKE) — would match properly inside it.
Trying again, the suggestions and remedy we acquired from Google Cloud was miles forward of different distributors. It felt like they actually understood our use case, and actually wished to companion with us one-on-one to carry our imaginative and prescient to life. As well as, they’d the best product suite for our wants and supplied us early entry to options that may match our scalability wants equivalent to rising GPUs from NVIDIA.
The selection was clear. We chosen Google Cloud and started working.
Our new answer: How we render personalised movies as we speak
Google Cloud’s hands-on help and personalised consideration didn’t cease after we signed our contract. Their groups have remained hands-on all through the complete means of designing, deploying, and increasing our new structure.
As we constructed our new system, Google Cloud consultants:
Continuously reviewed and commented on our structure designs and ongoing implementation course of
Pointed us to merchandise that grew to become key elements of our new structure, equivalent to GPU timesharing
Supplied enter as we solved small issues like configuring NVIDIA drivers to work with Kubernetes, and rewriting elements of our rendering engine to leverage Google Kubernetes Engine (GKE)
Continued to present us early previews of upcoming Kubernetes and GPU options, in addition to their latest NVIDIA chips and machines
Our engineers labored hand-in-hand with Google Cloud’s consultants to revamp our structure from the bottom up. Finally, we constructed on Google Cloud and created a cloud-based structure able to producing distinctive, personalised movies at a scale and velocity we by no means achieved earlier than.
Whereas our new structure is proprietary — and we have to preserve it close-to-the-chest for aggressive causes — we are able to share just a few of its key technical elements.
We created employee pods that every carry out a single rendering process at a time, and pull their jobs from Pub/Sub.
We use GKE’s auto-scaling to quickly scale from 1 to a nearly infinite variety of nodes, and tailor their efficiency on the degree of particular person workloads. This offers us granular and responsive management over the compute energy we deploy (and over our prices).
We use GPU time-sharing between pods to leverage our compute assets as effectively and successfully as potential — rising our common GPU utilization by 1.6x and reducing our prices by 66%.
By leveraging these companies, our system now runs a lot quicker, and is extra environment friendly, scalable, dependable, and adaptable than earlier than. With it we’ve reworked the merchandise we provide, the dimensions and velocity we are able to promise, and the shoppers we are able to serve — and it’s already delivered some vital technical and enterprise outcomes.
What we have achieved by partnering with Google Cloud
By partnering Google Cloud’s consultants and rebuilding our structure on Google Cloud and GKE, we’ve achieved outcomes that embrace:
Producing 8800% extra movies rendered per hour. With our previous system, we might solely render ~6,000 movies per hour. In a current check of our new system, we effortlessly rendered 540,000 movies in a single hour (and we all know it is able to rendering far, much more in that time-frame).
Growing enterprise-class capability. With our new system, we are able to now render and stream tens of millions of personalised movies at velocity and scale. With it we’ve been in a position to serve main world firms like BMW, WWF, Redbull, Purple Cross, Coop, ICA, and Motion Towards Starvation — and tackle greater initiatives, like sending 2+ million personalised movies for the biggest meals retailer in Sweden.
Clearing bottlenecks to our enterprise. Earlier than, we might solely generate movies for one or two campaigns at a time. Now, we are able to comfortably deal with as many campaigns as we’d like at one time — which implies we not want to show down initiatives and have elevated our income.
Constructing and providing new merchandise. We are able to increase our portfolio with highly effective new merchandise. For instance, we are able to now construct real-time personalised movies that stream seconds after the viewer inputs their information and presses “play” letting us scale new, dynamic merchandise.
Enhancing system visibility. We’ve elevated our logs, alerts, and general visibility into render standing for our large campaigns (one thing that was tough with our legacy structure).
In sum: By partnering with Google Cloud, now we have laid the groundwork to realize our most formidable technical and enterprise objectives — and now we have already dramatically improved how we render and stream an enormous quantity of cinema-quality personalised movies, and the way we serve our greatest and most demanding shoppers.
Deliver these outcomes to your group
At present, we’re constructing extra of our basic structure on Google’s Cloud companies, and enhancing its core efficiency so we are able to scale our personalised video manufacturing to insane ranges. Our large, formidable purpose not appears out of attain. The truth is, we now imagine we are going to quickly be capable to render our cinema-quality personalised movies in .2 seconds or much less.
To get there, we’re persevering with our shut partnership with the Google Cloud group. They’re serving to us develop new scaling methods, giving us essential finest practices to use to our structure, and guiding our improvement of recent merchandise like real-time rendering.
Collectively, we’ve reworked how our core merchandise perform — and you are able to do the identical by partnering with Google Cloud’s groups, by constructing on Google Cloud and by utilizing GKE. It’s labored for us — large time — and we all know it will probably give you the results you want too.
Particular thanks to Alina Bylkova, Lekë Dobruna, Gabor Lossos, Vigan Sokoli, John Rowley, and Michael Ivanov.