Key Criteria to Move Cloud Workloads to the Edge
with potentially thousands of other application clients , there is always the risk that private data may be disclosed , either unintentionally or through deliberate hacking attacks , to someone not entitled to receive it . Privacy is especially critical for personally identifiable data , or healthcare data covered by HIPPA and similar laws . Since edge computers are more distributed , more local , used by fewer tenants , and generally smaller than major cloud data centers , these privacy concerns can be reduced using edge techniques .
Taken as a whole , these five aspects of trustworthiness require careful attention to system architectural considerations , and deliberate partitioning of workloads between cloud data centers and edge nodes to optimize the system .
Why can ’ t we run everything in intelligent IoT devices / endpoints ?
Let us investigate the converse of running everything in cloud data centers , namely running the computation and hosting the storage functions directly on intelligent IoT devices , without significant contributions from the cloud or edge . This approach would seem to have promise in terms of attributes such as latency , network bandwidth , and scalability . But , as shown in the following discussion , there are some significant drawbacks to the intelligent IoT device approach .
Energy
Many IoT devices are energy constrained . Edge computing applications , especially those making heavy use of video analytics or AI , can require significant power dissipation in their processors and related hardware . Many classes of IoT devices are expected to run for years on reasonably sized batteries . Certain IoT devices would cause problems if they dissipated excessive heat to the environment , especially if they required noisy , failure prone cooling fans . So , many classes of IoT devices simply cannot dissipate more than a few watts of electrical power , and that severely limits the sophistication of the processing they can perform .
By moving the energy-intensive portions of the processor workload from the IoT devices to the layer of edge computing immediately adjacent to them , we can offload the high-power dissipation from a highly energy and heat constrained IoT device to an edge computing node without those constraints . These edge computing nodes can support multiple instances of different types of high-capacity computational resources , such as CPUs , GPUs , TPUs , specialized accelerators or FPGAs , some with the equivalent of tens of thousands of processor cores 13 .
Edge nodes can dissipate thousands of watts of power and be cooled by advanced forced air or liquid cooling systems . Since these edge computers are connected to the electrical grid , and often have backup power sources , they are not constrained by batteries as many IoT devices are .
13
See the “ Heterogeneous Computing in the Edge ” article in this issue .
IIC Journal of Innovation - 11 -