Key Criteria to Move Cloud Workloads to the Edge
in a remote location . A UHD camera ( 3840 x 2160 resolution ) running at 30FPs and using the H . 265 HVEC codec with high quality settings consumes a network bandwidth of 13.4Mb / s 9 . At this rate , this flow would transmit a gigabyte of data to the cloud in about 10 minutes , or about 144GB / day . Some cellular networks charge $ 10 per 50GB of data overage . So , at that incremental rate , the cost of hauling this one camera feed to the cloud for processing would be over $ 800 per month . Obviously , running the video analytics in the cloud is cost prohibitive because of the cost of the network bandwidth consumed . Some other access modes ( satellite internet , for example ) can have even higher bandwidth cost .
Excessive network bandwidth use has more problems associated with it beyond the monthly cost of the bandwidth . It also overloads networks and radio spectrum to the point where other users of the network experience delays or service reliability problems . If many high bandwidth endpoints are located close to each other , local network congestion may drastically slow down the networks , or prevent additional users from successfully connecting .
Moving the analytics algorithms from cloud data centers to edge computing nodes largely eliminates the need to send this high bandwidth traffic across wide area networks , saving those large bandwidth-related charges and preventing capacity issues . A local edge node is directly connected to the camera in the above surveillance example with a cable or short-range unlicensed wireless link that does not create monthly bills and performs the image analysis very near the camera . Then , only the results of that analytics (“ This camera didn ’ t detect any intruders during the last minute ”), which are orders of magnitude smaller in bandwidth can be sent to the cloud for action .
Data Gravity
Data gravity is the property of networks where certain datasets are optimally stored or processed on specific network nodes . The preference for data location can be due to many factors , including performance , geographic considerations , user policies , and government regulation . If all data must be processed and stored in the cloud , challenges can arise .
As a concrete example , consider wearable devices that record the medical vital signs from warfighters , and transfer that data to a command , control , communications , computer , and intelligence ( C4I ) networks . There are certain data gravity considerations associated with this system . The data is most useful to the local chain of command near its source and becomes diminishingly less valuable as the geographic distance between the wearable device and the data processing or storage location increases . It would make very little sense to move this sort of data to a cloud server thousands of kilometers away for processing and storage . There may be policies against transporting this data across certain boundaries ( outside a base , outside a theater of
9
IIC Journal of Innovation - 7 -