E-tailers weren’t the only ones grumbling on Cyber Monday. Enterprise users felt the effects of the online sales crush as North American Internet delays to leading SaaS providers increased by as much as 363 percent. This was the latest finding of an extensive research project conducted by Silver Peak and two leading Internet researchers into Internet Weather.
As we noted in our review of Black Friday, the holiday season provided an excellent opportunity to understand the impact network congestion has on the performance of the cloud and SaaS applications. This is particularly important for IT buyers in regions of highly developed Internet connectivity, such as North America, where the abundance of backbone bandwidth coupled by numerous alternative paths between locations, may inure users from traffic surges.
To test this theory, Silver Peak teamed up with InternetWeatherMap.com to gather detailed end-to-end latency statistics to SaaS and e-tailer sites across the Internet. Those results were supplemented with packet loss statistics for the major North American backbones as tracked by Keynote. (For more information about our methodology see this post.)
What we found was that while overall there was little variation in packet loss across the Internet backbones, latency was a different matter. Latencies to SaaS providers jumped on Cyber Monday even though sales grew less than expected (8 percent vs. the projected 13-15 percent).
The most extreme case within North America was a 363 percent increase in delay on a route from Ashburn, Virginia to Intuit in Atlanta, Georgia. Ashburn, Virginia was also the origin of a route to Zoho.com in San Francisco that increased by 288 percent, rising from 86ms to 334ms (See “Cyber Monday: The Most Erratic SaaS Routes”).
Normally, we might have assumed there was a problem in Ashburn, but those numbers reflected the trend across many routes. Reaching Success Factors in Los Angeles from Atlanta, Georgia required 232ms instead of the average 68ms. Getting to Box.com in Los Altos, California from Miami, Florida took 173ms, 116 percent longer than usual.
Nor were the problems constrained to North America. Reaching Marketo’s site in Virginia from Brisbane, Australia was the most extreme example, increasing by 1,217 percent. Moscow users looking to reach Dropbox in San Francisco saw latencies grow 237 percent. London to Intuit in Atlanta? 76 percent longer.
Such extreme fluctuations pose significant challenges for IT professionals as they look to shift applications from MPLS and private data services to Internet VPNs. High latency undermines application throughput, but extreme latency variations complicates the network deployment by randomly interfering with the user experience. Both need to be planned for in any deployment.