Aug 3, 2012

The brains behind intelligent route control

Ansley Kilgore

The brains behind intelligent route control“In a picture, it should be possible to discover new things every time you see it.” – Joan Miró

Internap’s Managed Internet Route Optimizer™ (MIRO), the brain behind our Performance IP™, is really astounding when you think about it. It makes millions of calibrations every day to measure the Internet and route around problems. Every fraction of a second MIRO collects massive amounts of data. The algorithm then analyzes, parses and prioritizes so that it can perform routing changes that make sure our customers’ content reaches every destination optimally. It makes our Performance IP three times less likely to suffer a network brownout or blackout than commodity bandwidth providers. It reduces packet latency across destination prefixes from any one Private Network Access Point (P-NAP®) by 25 milliseconds on average relative to any single competing backbone.

I’ve been writing about MIRO for more than four years now, in earnings scripts, press releases, marketing collateral and missives to investors and clients. Our customers have spoken volumes on MIRO and the benefits it provides to their own clients. Today, there are 546 mentions of MIRO on Internap’s website. On this very blog we’ve compared MIRO’s awesomeness to everything from parts on an F1 race car to Mom’s meatloaf. Other MIRO comparisons we could have referenced but didn’t (until now) include Professor Badass, Jetpacks and Bacon.

One of our optimization patents (US6912222 for you patent bugs out there) describes MIRO’s inner workings as:

“A Private Network Access Point (P-NAP) packet switched network is provided where two customers connected to the same P-NAP exchange traffic through the P-NAP without transiting over the backbones of the Internet. In addition, a multi-homed customer connected to the P-NAP is provided with access to the P-NAP optimized routing table so that the customer will also have the ability to know the best route for a particular destination. In this way, if a multi-homed customer connected to a particular National Service Provider (NSP) to which a destination is also connected, the P-NAP customer can use the P-NAP information regarding the NSP to send the information to the destination through that commonly connected NSP in the most direct fashion.”

But inevitably, words can’t really describe how awesome MIRO truly is. With our new Route Performance Monitor (RPM), I think we’ve started to address that difficult issue. RPM has two views:

The Monitor Internet tab

Monitor Tab

Here you can see some of the significant network events that MIRO sees emerging across the Internet every day. Optimal Internet paths are determined by running trace routes between the selected P-NAP and destination points. After mapping IP addresses of each hop into a mapping database to obtain city names and longitude/latitude information, RPM ascertains the worst round trip time (RTT) to each probed prefix and then compares that path to the alternative carrier route MIRO would select (i.e., the “best” route). RPM then randomly selects 50 events per day that have more than 500 ms of difference between the best and worst route options. By clicking on any event dot, a pop-up window appears displaying information on each carriers’ latency performance from the origin P-NAP to the selected destination.

And the Compare Carriers tab –

Compare Carriers

This tab displays average latency values (using ping and trace routes) across all probed prefixes (~ 400,000 at any one P-NAP) for each carrier over a 24-hour period and compares these suboptimal routes to Internap’s latency across the same prefixes and time period (MIRO leverages only the optimal routes across all carriers). The “split-flap” display in the screen-left dialog border shows the average latency difference between Internap and the carrier with the highest latency carrier on any particular day.

Split Flap

This figure represents the average improvement you might see by using Internap’s Performance IP versus a single poorly performing carrier per packet round trip. So if you have 40 objects on a web page and the average improvement is 25 ms across all prefixes you are potentially saving one second per page load…that’s HUGE (see our page on every second counts).

We’ve posted the RPM data collection and display methodology here if you want to delve into more detail. Take RPM for a drive and if you want to see how your bandwidth provider is performing to certain geographies or how they compare with Internap, leave us your email and we’ll be happy to walk you through the details.

Explore HorizonIQ
Bare Metal


About Author

Ansley Kilgore

Read More