Having visibility to more than 100 datacenters and at a scale at which Zscaler operates provides invaluable insight to events that happen to impact the Internet. We service 4,000+ customers processing more than 10B transactions per day from 180+ countries daily. As a result this allows us to have a unique perspective to quantifiably measuring Internet-wide impacting events.
Before I continue I wanted to make an important preface to this post. I am always the first to say that no network, provider or infrastructure is ever immune from outages. By no means am I attempting to point out any flaws or pick on any organization. Nevertheless after hearing sensational headlines around Google’s recent 3-5 minute Internet-wide outage I wanted to investigate if those headlines were indeed true. Harnessing the power of our Nanolog platform we can observe on a second by second basis cloud wide not only how much traffic was passing through our cloud (in terms of bytes transferred) but to where it was going and all kinds of other attributes.
In short, the headlines were true. Google’s event had an astonishing effect on our network. Given the diversity of our customer base it is not unfair to extrapolate this out to the broad Internet in general either.
Beginning just shortly after 15:50 PT on Friday (Aug 16) we recorded a sharp decline of approximately 30% of all traffic flowing back and forth through our cloud globally. This decline happened in less than 100 seconds into the event. Around 15:56 we began to see a quick and noticeable increase back to levels previous to the outage.
What is so remarkable about this event is that it had a global impact on our cloud. Normally events that can move the needle on our cloud are regional – never global in nature. The other interesting item is that to novices if Google goes down they equate that as the Internet in general is down. I am sure we have all watched those who will “Google” a fully qualified URL instead of just typing it into the address bar. As a result when events like this happen it becomes an education process to inform our customers that we are still up and operating.