Nov 26, 2014
Researchers have long been interested in retailers’ ability to cope with online demands of Black Friday and Cyber Monday. Less interest, though, has been paid to the impact the surge of e-commerce traffic has on the cloud and the enterprise as whole. For IT managers interested in shifting some or all of their traffic from corporate wide area networks (WANs) that run over private data services to the Internet this is probably the more interesting question.
As we rethink how to build faster, nimbler, and more cost effective WANs, the Internet looms as the logical replacement for private data services. The ubiquity, low-cost, and short time-to-deliver makes the Internet an exciting alternative to private MPLS service. At the same time, the lack of service level agreements (SLAs) and centralized control means there’s no one provider to turn to should Internet backbone performance suddenly decline.
For non-professionals the strengths and weaknesses of the Internet are of minor importance. The characteristics of the Internet are such that users receive a fairly consistent Web browsing experience. Yes, there are times some sites may be sluggish, but those instances are too infrequent to be of serious concern for many.
But for networking professionals the Internet’s tendency to fluctuate should be of major concern. Web browsing is fairly resilient to delay, but as enterprise applications involve data transfers or must operate within specific-timing constraints (as is the case of virtual desktops, voice, video, real-time dashboards and the like), the effects of latency and loss play a larger role. The algorithms used to calculate TCP session throughput are dependent on latency and packet loss — not bandwidth. So as enterprises consider the Internet, network professionals must determine whether the shifts in latency and loss across Internet connections will interfere with their enterprise application.
And here is why Black Friday and Cyber Monday are so intriguing. They offer some of the most severe test conditions one can expect to encounter when traversing the Internet. When else can you arrange for millions of users to pound on websites?
To uncover how changes in Internet weather will impact the enterprise we collected data from two sources from the start of November. We worked closely with InternetWeatherMap.com to map latencies across the Internet and supplemented this work with results for packet loss statistics from the Internet Health Report by Keynote Systems. The report provides a real-time view into the network performance of nine major Tier 1 US Internet backbones.
InternetWeatherMap.com runs traceroutes every 15 minutes across about 85,000 hops throughout the Internet. (The specific number changes regularly and the current number is up from 60,000 hops when we last worked with InternetWeatherMap.com.) Included in the locations monitored by InternetWeatherMap.com were IP addresses from the top 30 e-tailers as defined by Deloitte’s report the “Global Powers of Retailing 2014” and IP addresses from leading SaaS providers.
Each e-tailer can in theory show dozens of external IP addresses, many of them never accessed by most users. To ensure the accuracy of the data, we reduced our dataset to only those IP addresses visited by consumers when typing in the website name and limited our list of e-tailers to those within North America.
To measure the impact on the cloud, we monitored latency to IP addresses from more than 30 SaaS providers. This list was extracted from Silver Peak’s Cloud Intelligence service, which aggregates Internet “weather” and cloud information and distributes that information to all Silver Peak Unity instances.
The outcome of all of this work are a series of forecasts, reports and campaigns before, during and after the holiday seasons. For full listing of all items visit this blog page. You can also participate in the research through our #TweetYourPing Sweepstakes (and be eligible to win a GoPro Silver Edition camera).