Here at Internap, we pride ourselves on providing the fastest available Internet for our customers. Our roots are in the innovative Managed Internet Route Optimizer™ (MIRO) technology developed by our founders back in 1996, and we still strive to provide high-performance Internet Infrastructure that leverages the MIRO algorithm to deliver maximum uptime, speed and reliability for our customers. That’s why we’ve redesigned MIRO to deliver even better Internet performance for today’s applications.
The performance of every Network Service Provider varies based on bandwidth capacity, traffic on their network at any given time, network black-outs/brown-outs and number of hops required to reach the destination. These factors ensure no single NSP can provide the best performance all the time. The massive growth of Internet traffic (think big data, Internet of Things, mobile, social, etc.) and number of interconnections and end points exacerbates the problem; i.e., NSP capacity isn’t growing fast enough to keep up with the increase in traffic. Adding to this, Internet traffic must travel across multiple devices to reach its destination, creating the potential for failure at many points.
The graph below shows a 30-day time slice from an Internap Point of Presence (POP). The data measures performance of all providers located in the POP for existing Internap customer traffic.
Notice that no carrier provides the best performance more than 15% of the time. To tackle this problem, many companies buy bandwidth from multiple NSPs, but they have to manually route traffic. Since this is an intensive manual process, and NSP performance can change quickly, it’s next to impossible to fully optimize traffic using this method.
Internap’s MIRO algorithm dynamically evaluates performance across NSPs and routes traffic over the best performing path. While MIRO works extremely well, making millions of calibrations daily to avoid blackouts, brownouts and other problems, we wanted to make it even better. With exponentially increasing Internet traffic and an explosion of new prefixes, we need to be ready to handle future growth. The global routing table, which refers to the Internet’s total number of destination networks, has quadrupled in size since 2004 (see chart below).
We’re proud to announce the deployment of next-generation MIRO. Here’s what’s changed:
- In the next generation of MIRO, we have made the probing engine scalable. That means we can run multiple probing engines simultaneously, which reduces the time to identify the optimal routes. To accommodate growth, we can keep adding probing engine instances in a market, which maintains efficiency. This is analogous to spinning up additional cloud instances in parallel to provide faster processing and manage increased demand.
- We’ve changed the route optimization engine from a meta-heuristic model, which provides a sufficient solution but not the most optimal solution, to a linear programming model that computes global optimal solutions every single time. The benefit of the linear solver is that it is less complex and more deterministic, as opposed to more complicated rules and heuristics.
- Our new route optimization engine is much more efficient, and allows us to examine more of the prefix table at once. This has also required more computing power, so we have upgraded MIRO’s hardware as well. New MIRO maximizes performance over the entire set of prefixes in the market each time. This means that the new solver is taking a holistic view of all prefixes in the market every time it makes a new solution, and more prefixes are consistently put on the best performing path.
- MIRO technology was originally developed at a time when IPv4 address shortages were not an issue, so it didn’t initially optimize IPv6. As the internet grows, unallocated IPv4 addresses are depleting quickly, and IPv6 is in various stages of deployment. The Next Gen MIRO has been purpose built to handle IPv6. This is another way that we have future proofed MIRO.
How do these changes translate to performance?
We did some initial tests in New York and found that next-gen MIRO is making 650,000 optimizations in an hour on average. To put this in perspective, it takes about five minutes of manual effort to make one change. So we are talking about two man hours of effort reduced per customer in one minute.
Next-gen MIRO is on average about 20ms faster than the fastest single provider and 60ms faster than the slowest. Lower latency is good but consistent lower latency is better. In the graph below, notice how the next-gen MIRO (red line) is not only much lower than any other carrier latency, but it is much smoother compared to any other provider.
At the beginning of this blog, we saw that no carrier was providing the best-performing route more than 15% of the time.
Our redesigned MIRO can transfer traffic over the best route more than 99% of the time. The ability to complete a probing cycle 120 times faster results in eight times more optimizations. Because the next-gen MIRO can evaluate routing paths considerably faster, it can put more traffic on the best performing route.
The revamped MIRO also provides significantly lower latency and less volatility, ensuring consistent performance. Scalable probing and IPv6 parity make our next-gen MIRO future-ready.
In keeping with our mission to provide performance without compromise, we have future-proofed our MIRO technology. With a new algorithm, redesigned components and upgraded hardware, MIRO is more modular, efficient and scalable, ensuring the best available Internet performance for Internap customers.