“Each cloud is a little different, does things its own way, speaks its own language and of course, brings joy. Sometimes, however, clouds can also be hard to work with.” (via VMWare’s blog)
As cloud computing continues to evolve, there are numerous cloud service providers in the market offering variety of services and newer capabilities. And, enterprises typically tend to pick and choose best of breed clouds to meet its overall needs resulting in hybrid / multi-cloud architectures.
As enterprises move into their hybrid / multi-cloud architectures and reap its benefits, they inevitably face challenges due to multiple ‘cloud silos’. Here, we discuss few of the common challenges enterprises face with managing workloads across ‘cloud silos’ in typical hybrid / multi-cloud deployments.
- Cloud native lock-in for workload provisioning and migration
The interfaces to manage and provision workloads are very specific to the cloud service providers leading to tightly coupled workload provisioning engines. This creates challenges to seamlessly provision and migrate workloads across multiple clouds and embrace diverse cloud providers.
A recommended approach is to isolate the workload provisioning and migrations from the cloud-native workload management interfaces and make it agnostic to the underlying cloud.
- Disjointed view of resources across clouds
The compute and storage resources of a particular cloud is usually viewed and leveraged independently of the resource pools in other clouds. This leads to disconnected scheduling, scaling and monitoring of the workloads across clouds and creates significant operational overhead.
“The whole is greater than sum of its parts” – Aristotle
A recommended approach is to create a ‘Single Pane of Glass’ view into your resource pools to optimally manage the workloads across the basement, public and the private clouds. This approach provides a holistic view of the available and utilized resources and can yield in optimal workload scheduling and scaling.
- Sub-optimal utilization of Cloud resources
Very often, workloads are statically targeted to be deployed on pre-determined hosts and VMs in the cloud which creates static compute and storage partitions. This leads to a sprawl in fine-grained VMs and hosts. Fine-grained static partitions create barriers to run mixed workloads which leads to sub-optimal usage of resources.
A recommended approach is to create a shared resource pool across multiple clouds and dynamically allocate resources to the respective workloads based on policies and changing needs. This will eliminate static partitioning and will enable optimal execution of mixed workloads.
- Inability to automatically failover across Clouds
Cloud ready workloads are usually capable of being resilient (automatic recovery on failures) within a cloud. However, these workloads may not be resilient when a particular cloud is inaccessible due to site disasters, planned and unplanned cloud outages.
A recommended approach is to setup a distributed system that can detect failures / outages in a particular cloud and automatically failover the workload to a different cloud.
- Performance and Latency bottlenecks for distributed workloads
Distributed workloads are usually deployed across wide geographies leveraging multiple clouds. These workloads need to communicate and exchange data between clouds and tend to use public internet leading to unpredictable latency causing slow performance.
A recommended approach is to enable private low-latency connections across clouds by leveraging Platform Equinix capabilities to improve performance.
There are several tools and frameworks in the market focusing on cloud workload management. Below is a suggested approach leveraging open source Apache Mesos in conjunction with Platform Equinix to address the challenges discussed above.
Apache Mesos enables a ‘Single Pane of Glass’ approach to manage workloads across hybrid and multi-clouds. It is a cluster manager that creates a shared resource pool across bare metal, private and public cloud resources with standardized APIs and agent based approach. It also acts as a resource negotiator and meta-scheduler to manage the workloads.
By applying the capabilities of Apache Mesos in conjunction to leveraging Platform Equinix, enterprises can have an optimal and flexible hybrid / multi-cloud architecture effectively managing the cloud-silos.