Recently, we had the opportunity to interview Satish Raghunath, Co-founder of Twin Prime, which officially launched in March. Twin Prime has introduced an innovative business model to the CDN Ecosystem that’s a first of its kind. Twin Prime is a “Mobile App Acceleration Network in the RAN”. The way the service works is simple. The service is comprised of two parts, the SDK that’s integrated into the mobile app, and a PoP infrastructure that houses the brains and communicates with the mobile app. Once implemented, their platform studies and learns the behavioral conditions of the mobile network, and then intelligently optimizes the network to accelerate content delivery over 3G, 4G and WiFi networks.
They are not a CDN, but they have built a global PoP infrastructure. The SDK is the differentiator that separates them from Rev Software, Saguna Networks and Akamai. As illustrated below, Twin Prime runs independently of a CDN, bypassing CDNs in the optimization process.
Twin Prime Interview
Bizety Team: Who is Twin Prime and how did you come up with the idea?
Satish: My co-founder, Kartik Chandrayana, and I go a long way back; we met in grad school in upstate New York. We went to RPI, where we did our PhDs in Computer Networking. Then after graduation, Kartik and I worked for a number of years on the operator side of things: phone networks, routing, security, WiFi, network optimization, you name it. All flavors of networking problems. That’s really what excites us.
I spent a couple of years at Nortel, but most of my time was spent at Juniper Networks, working on the core and edge routers, building services for large telecom vendors. With Twin Prime, our goal was really simple: to make mobile faster; but as you know, that’s quite an expansive goal. The way Twin Prime thinks about it is from a network perspective. We raised money from several blue chip investors, including Draper Fisher Jurvetson (DFJ) and True Venture.
Today’s mantra is that mobile is all about apps. Recent research showed that more than 86% of users’ time is spent on apps; whereas all of the analytics companies have been reporting that mobile connections are 3X slower compared to websites. The big question is why mobile is different? What changed in the last few years to make mobile a problem again? If you recall in the late 90s, we solved the performance problem with the wired Internet. Compared to the wired network, it’s only the last mile that changed. The last mile is the segment including the handset, the cell tower and the wireless link in between; especially with cellular networks, if you look at the latency, 70% or more of the latency is spent on the last mile network. It’s what you do in the last mile that matters.
CDNs solved the latency between the origin data center and the edge of the Internet latency problem. However the last mile is still underserved by the solutions that we have today. This is why we focus on the mobile last mile – its different enough to warrant a completely new approach.
BT: Twin Prime brings a fresh new perspective to mobile app acceleration. How is your approach different than a CDN? We understand CDNs have very little control over the RAN, so some have partnered with companies like Saguna Network, in order to get more control in the wireless last mile. What’s your thoughts on that?
Satish: It’s a great question because there are self-organizing networks which are critical in terms of how cellular operators have been able to adapt to demand and congestion. There are at least two independent parts to the overall solution. One as you mention is in the RAN itself. The other part of it is outside the operator infrastructure, i.e. once the operator/RAN has fixed the resource allocation, how does an application make the most efficient use of that resource? The reason why this is a problem is that our current solutions are used to a world that is consistent and does not have the volatility or diversity that mobile features today. The physics of the wireless channel introduces a lot of variation in metrics such as bandwidth, latency, jitter, etc.
Diversity is another factor. What we’re seeing with mobile is an unprecedented number of devices, operating systems and apps. If you think about it, there are 20,000+ device types, dozens of operating systems and then thousands of applications working all the time. They all combine to create an astounding amount of diversity.
BT: How are you addressing theses problems?
Satish: The obvious question to ask is: why can’t current solutions like CDNs help with this issue? The main reason is that current solutions mostly feature a one-size-fits-all approach – come up with a one-time optimization solution that you hope will solve for numerous performance factors. As you can see that mindset is in direct contrast to the inherent nature of mobile today. You need to do different things for different mobile users because of the inherent diversity in mobile.
Thats the problem: we know how diverse and volatile the mobile network is but just throwing resources at an infrastructure level is not enough. We need a higher level intervention that is automatically addressing the huge number of factors that we see impacting mobile performance. Twin Prime is how we can achieve this.
There are three components to Twin Prime: one is the SDK component, which is a small library that sits in the application itself. Then the other is the proxy in the cloud – what we call the Twin Prime Accelerator. Both of these components are in the path of the traffic. There’s a third component, which is the core of our intellectual property, the Global Location and context Based Acceleration Strategy (GLAS). I mentioned that there are all these problems with reading the network and performance factors, right? The job of GLAS is to look at a lot of measurements and data to churn out machine-driven hypotheses.
These hypotheses look at variables like network type, time of day, performance, signal strength, handset type, etc. There are many questions you can ask of this data source. The bottom line is that GLAS is churning out all these hypotheses, then based on the performance factors that it says are important, you then evolve strategies for those performance factors. This Automated Hypothesis Testing (AHT) framework driven by GLAS is what enables to provide superior performance.
BT: HTTP2 has just arrived and it’s significantly altering the picture when it comes to protocol optimizations over the middle mile. It has a lot of baked-in features that CDNs typically charged for, whether its DSA, FEO, etc. How does Twin Prime handle HTTP/2?
Satish: Excellent question. To make your question a little bit more generic: how does the next new technology or protocol fit into this picture? Take this particular example of HTTP2: the question you’ve asked is how HTTP2 affects a performance perspective. HTTP2 enables many long-sought advanced capabilities such as multiplexing, pipelining and prioritization of requests to a server. Now the question becomes, when is it good for my application to leverage a particular capability?
When does using a HTTP2 feature lead to a better performance? There is no easy answer to that – that is where GLAS is crucial and it enables you to answer a question such as “Hey, should I use this particular HTTP2 capability considering I’m coming on AT&T’s LTE network at 5:00 pm? Will it help me, or not matter, or hurt me?” Then GLAS uses historical information and analysis to provide guidance in the form of a simple rule-set on when you benefit the most from particular capabilities or knobs or levers.
So the question is not whether the next new protocol will solve the problem, but the question is, how do you fit the next new protocol into your set of tools? We see using the GLAS as a way to pick and choose the best tools for the right context to maximize your performance.
BT: Great answer. Your Juniper Network experience is coming out. Most CDNs aren’t thinking 4-5 years into the future. In 5 years, next generation GPON (gigabit passive optical network) and 5G are going to be major disruptors to our industry. In that time we are likely to witness 1Gbps downstream / upstream speeds on mobile networks, and 10Gbps – 100Gbps per wireline location. Do you think five years ahead?
Satish: I love that idea of 1GB up and down, I’m waiting for that myself. As you mentioned, optical networks are so mature right now that it’s not a problem to add or grow bandwidth in the near future or even medium term. The problem is multiple layers of software not being able to work with each other – they’re second-guessing each other. Take for instance, LTE. Most of our customers see better performance on LTE than on 3G.
Why? Every mistake that you make in transferring data on top of an LTE network is 10x more expensive for you than the standard network. On an LTE network, you are talking about 1 vs. 100 Mbps difference: you’re talking about the difference between a 4K video and a grainy SD video. The question is not about maximum bandwidth: it’s related to the achievable throughput – the more the variation, the more the need for intelligent software to build it and do the right thing.
BT: Do you consider yourself a CDN? It seems you have deployed CDN like PoP infrastructure globally. Also, do you require a DNS change pointing to your network in order for the service to work?
Satish: We designed our solution recognizing that the customers have invested a lot in the backend infrastructure for a reason. It gives them scale, security, geo-presence. We’re not asking them to change their backend or DNS. Twin Prime provides a simple mechanism to speed up your app whenever it makes a request to get content. The SDK automatically knows what to get and then ships that request to the accelerator. The other question is, are we a CDN? Unlike CDNs, Twin Prime is focused on a whole set of different technologies. Now that mobile is the most important screen, content delivery presents a new problem and we are going to follow that new problem. You can say that we are complementary to a CDN. We don’t want to replace them, we are bringing a whole new set of capabilities when it comes to mobile. These are capabilities that enterprises see as essential to achieve their business potential.
BT: So you offer an SDK that customers integrate into their mobile app. How does it work and do you touch the web server in any way?
Satish: Twin Prime has made it simple. Instead of touching the web server, we have a component in front of the web server which transparently does the job. We don’t touch the server at all. There are two parts. That is how we’re able to do interventions both on the backend and the handset side.
BT: How do you price out your service?
Satish: The pricing has been driven mainly by who we are talking to, which is people who own mobile apps and need to drive engagement, retention and such metrics. What we’ve seen is that they understand what drives their growth: things like monthly active devices and monthly active users; that’s how we’re pricing them. We add value to your user engagement and retention, so we price based on monthly users. We’re not thinking about it from a bandwidth or a CDN perspective.
BT: Do you see yourself as a competitor to Rev Software?
Satish: The reason we feel our approach is different is that we are framing this problem as a very broad set of hypotheses. Twin Prime asks different questions: what is affecting your performance right now? Is it a network issue? Is it a transport issue? Is it an application issue? Is it a handset issue? You need to be prepared to deal with the entire set of variables. That’s where we differ. We’re building a framework that can deal with a much larger number of variables. I think that’s reflecting the context of this whole problem. At some point, we’ll have more clarity around what are we facing in each of these directions.