Cloud Edge Solution Comparison Accelerates Pharma's Path to Business Value
In this ATC Insight
As part of an edge transformation, a particular challenge global enterprises face in developing and deploying new business-value producing capabilities at scale is the matrix of business and technical decisions required to move into a new, cloud-first paradigm. Many things must fall into line, including operating model, budgeting and procurement processes, plus site and line of business constraints and initiatives — all while incorporating security and governance controls.
One of WWT's key differentiators is our Advanced Technology Center (ATC). The ATC supports complex customer decisions with a breadth and depth of technological expertise contained within a single ecosystem. When the ATC Lab Services, Cloud Services and Consulting Services teams partner on use case-driven approaches, we elevate business value as part of the technology testing and evaluation process. We can deliver at unparalleled speed and comprehensiveness that gives customers the confidence that their requirements will be met and hypotheses answered.
Recently, a large pharmaceutical company (Pharma Co.) reached out to WWT for support in their multicloud transformation activities that included conducting a proof of concept (PoC) comparison of edge devices from three cloud service providers (CSPs): Microsoft Azure Stack Edge, AWS Outposts and Google SITE Appliance (not yet in general availability as of October 2021). What we found was a rapidly evolving ecosystem of CSP edge capabilities. While no one-size-fits-all solution ended up meeting all of Pharma Co.'s enterprise use cases, they derived significant value from leveraging best-of-breed edge devices to properly test three value-driven use cases in the ATC.
As organizations and CSPs continue to innovate in the edge space, product roadmaps will rapidly evolve over the next 12 to 18 months, and the suitability of edge devices for supporting additional use cases may mature — so watch this space!
The cloud operating model is an attractive proposition for many organizations due to the ease of deployment and use, uniformity and integration of user experience, and overall agility and connectedness. As more companies operationalize their cloud operating model, they are faced with some common challenges of moving workloads freely across clouds, extending cloud capabilities closer to users on the ground, and optimizing cost as operations scale. With many cloud providers now offering cloud-managed edge hardware, there is an emerging solution for the cloud-to-ground challenge.
Pharma Co. was interested in testing the edge devices (Stack Edge, Outposts, SITE Appliance) for out-of-box capabilities and suitability to support a wide range of business use cases. This required the setup of a complex test environment within WWT's ATC, with key questions including what should be in the environment and what does success look like?
Focus on business value
We worked closely with use case owners, practitioners and technology experts across Pharma Co., WWT and technology vendors to thoroughly identify, define, design and test a limited-but-diverse set of use cases. This allowed us to finalize testing and technology recommendations in a short window. Our approach collected business goals and KPIs, outlined existing processes, and documented all relevant components such as cloud services, applications, infrastructure, data and more.
The three use cases identified for PoC were:
- Manufacturing: A near real-time feedback control system at the edge, on a standardized device with existing cloud services, allowing greater deployment scalability of innovative algorithms and approaches.
- Vision: An innovative approach to video analytics on the edge that allows for quicker and more accurate classification and analysis of images in various scenarios, such as production, warehousing, robotics and more.
- R&D: An edge device that allows for a more scalable and easier-to-manage approach to processing complex molecular structures using established third-party software to analyze terabytes of data produced daily.
For these use cases, we broke down the cloud-to-ground challenges into specific questions that mattered. The uncovered business and technical requirements influenced how we defined success criteria and informed the build and test of the environment. These requirements also provided the confidence in value across many other enterprise use cases. Some key examples of these requirements include (but are not limited to):
- "Manufacturing" requires the ability to operate at different layers of the ISA 95 framework for industrial security.
- "Manufacturing" requires the continuation of critical operations in unstable environments that may experience disconnected states for hours to days at a time.
- "Vision" requires high availability to support potential intermittent cloud connectivity or malfunctioning cameras, which should not halt operations.
- "Vision" requires scalable and cost-reasonable GPUs to support a diverse set of intense analytical models required per vision scenario.
- "R&D" requires the ability to support and run various niche third-party scientific software applications, running on both containers and virtual machines.
- "R&D" requires large, fast and cost-effective data transfer and storage to deal with large scientific datasets.
With requirements uncovered and test environments designed, the team had just four weeks to set up the environment (~one week), execute the tests (~two to three weeks), and summarize results (~one week, including final validation and demos). The cross-functional team — comprised client stakeholders, vendors and WWT subject matter experts — approached the PoC in an agile manner to progress the setup and iteratively test of each device and use case in parallel.
The technology environment we created in the ATC was designed to validate the key value drivers that the business expected from the respective edge solutions. To reflect a remote site, the ATC test environment consisted of three zones: the shop floor (where data was generated), the edge device (where data was ingested and processed by workloads), and the cloud (where data was stored, and further processing was possible later).
Edge devices were deployed leveraging cloud direct connections and Equinix handoff to an associated cloud account. Configurations were applied in partnership with CSP experts to ensure best practices and recommended architectures. Due to differences in the respective CSP device feature sets, we standardized configurations where possible and enabled the most comparable cloud services to minimize differences in testing to perform a fair comparison. Working with Pharma Co. use case practitioners, the use case workloads were then deployed onto the edge devices.
With device and use cases set up, each step of the use case process was executed as functional tests were accompanied by or reviewed by the use-case practitioners and owner:
Test Case 1: Manufacturing use case
- Business Process:
- Industrial Internet of Things (IoT) messages generated as json files and sent to edge storage on the device.
- Machine Learning (ML) model reads and processes the messages from edge storage.
- ML model outputs result and sends a message out to a receiver.
- Results copied into edge storage.
- Components: IoT message generator, dynamically generated json files, LSTM model.
- ATC Implementation: WWT's data science team quickly created an LSTM model, intaking randomly generated inside and outside temperatures (and for more fun, introduced some "noise") to then predict a future temperature message. Alerts were generated if forecasted temperature exceeded an acceptable threshold.
Test Case 2: Vision use case
- Business Process:
- Images copied onto edge storage from source.
- Images sent to cloud data gateway.
- Images copied into cloud storage.
- ML model inference of images on edge storage.
- ML model outputs results, copied into cloud storage.
- Components: Dataset of realistic large images, CNN image classification model.
- ATC Implementation: The client provided the CNN model. WWT Data Science team made some minor code changes for containerization and created a full data pipeline to the image dataset.
Test Case 3: R&D use case
- Business Process:
- Images copied into edge storage from source.
- First vendor software reads images and performs various CPU/GPU/RAM-heavy jobs and outputs results.
- Independently, second vendor software reads images and performs various CPU/GPU/RAM-heavy jobs and outputs results.
- Components: Dataset of realistic large images, two third-party software products for cryogenic electron microscopy.
- ATC Implementation: The vendor provided trial software, and we collaborated with Pharma Co. scientists to install and operate the software as part of executing the test cases as the software required scientific understanding.
Results and a bright future
Pharma Co. validated their hypotheses and developed a more in-depth understanding of each edge device's current functional and non-functional capabilities as it pertained to supporting their unique use cases. Summarized takeaways at time of testing include (but are not limited to):
- Each cloud service provider possesses a broad edge product portfolio to serve diverse use cases and constraints, such as rack/floor space. AWS Outposts is a 42U rack, whereas Azure Stack Edge and Google SITE Appliance are smaller (1U rack-mountable).
- Each edge device can support different combinations of virtual machine instances and container engines. For example, at the time of testing, Google SITE Appliance only supported containers.
- The specific CSP implementation of the container engine and the management plane of the edge deployment directly impacts the CNCF compliance, consistency of container experience from cloud to edge, and the ability to continue working in a disconnected state. For example, at the time of testing, Azure Stack Edge was best at operating in a disconnected state because it deployed management plane components in a default container instance; thus, it relied less on cloud components. The tradeoff was that the container experience on ASE was not 100% consistent with using Azure's container engine, AKS, in Azure directly. On the AWS side, Outposts has a consistent AWS container engine experience, EKS, but it was not designed for limited connectivity operations. In both examples, the CSPs offer other solutions in their edge family to address a broader range of considerations.
- All edge devices have some GPU capabilities, but the number and type of GPU varies by device. The physical GPU mapping to be shared against VMs and container instances also varies.
- Each device has a local storage option and cloud sync that is either native or requires a separate setup. Where a specific storage type was not available on the device, such as object storage on the SITE appliance, third-party apps were connected to provide the capability.
With proof of concept complete, these learnings and vendor discussions enabled the Pharma Co. team to be confident in identifying the best-fit technology for their current use cases. The team also left better prepared to capitalize on the benefits of edge technology by extending cloud capabilities closer to their business across a variety of use cases.
Now moving into the Minimum Viable Product phase, WWT continues to expand the worldview of business use cases by interviewing more stakeholders to capture their business and technical requirements and prioritizing the use cases for pilot alongside Pharma Co. business leadership. This will ensure the pilot use cases on the chosen edge device are a priority for the business and will create meaningful business value.
Maintaining close partnership between business users, technology teams, partners and cloud providers accelerated the overall delivery of business value and the adoption of edge technology at an enterprise level. As technology and application patterns gain approval, future work remains, including aligning Pharma Co.'s operating model from a legacy approach to a cloud consumption model and integrating these new technology patterns to processes deployed at scale: budget planning, procurement request, service management, deployment automation, MLOps, support and more.