To say that public cloud infrastructure as a service is seeing explosive growth is an understatement. Amazon Web Services alone experienced 78 percent year over year growth in revenues from 2014 to 2015. Most of this hyper-growth has come from small and medium businesses that have fewer resources to build out and manage their own data centers. Enterprises have been much slower to adopt public cloud infrastructure, particularly for production workloads.
That said, there are strong indications that 2016 may be a breakout year for enterprise IaaS adoption. In fact, a recent CIO survey conducted by Morgan Stanley in October 2015 places cloud computing as the top spending priority for enterprises for the first time.
Barriers to Public Cloud Adoption
While enterprise IT executives have long understood the potential benefits of public cloud in terms of agility and cost-efficiency, they have also faced major barriers to widespread cloud adoption. These barriers include:
- Complexity of Migration: Public clouds often provide a fundamentally different computing environment than on-premises data centers. IT experts must consider whether their applications will play nicely with different hypervisors, networking, and storage. Migration may require complex changes to existing applications and workloads, especially stateful workloads with persistent data. According to Morgan Stanley’s CIO survey, 91 percent of cloud migrations fell short of expectations, with “complexity” sited as the primary reason.
- Security Concerns: Many organizations must comply with internal or industry regulations governing data privacy, and placing production data in the cloud can be a problem.
- Data Migration Time: With stateful workloads that contain large data sets, the time to migrate can be significant. For instance, migrating 20 terabytes of data over a dedicated 20Mb/sec link would take 100 days.
- Data Migration Cost: Unlike the cost-effective cloud storage for cold data, backup, and archiving, the cost of high-performance block storage is significantly higher. Further, customers pay based on capacity or reserved performance (IOPS), not solely on use. This increases the total cost of ownership for these workloads in the cloud when compared to on-premises shared storage arrays.
- Lock-in: Enterprise IT managers are often concerned that moving to the cloud is a one-way ticket, and they will not be able to easily revert back to on-premises or switch to another cloud.
- Sunk investment: Most of today’s large enterprises have invested for years in building extensive on-premises data centers to support their worldwide operations. Even if their strategies going forward include a ‘cloud first’ directive, it will be years before their existing data center investments can be taken offline and fully depreciated.
So What Has Changed?
These barriers are beginning to erode with the emergence of more mature, secure cloud offerings along with better migration technologies, enabling enterprises to seriously consider a shift to the cloud. But which public cloud offerings are most effective in addressing these barriers?
Broadly speaking, there are two camps of clouds vying for dominance. The first is single source cloud, where new workloads are developed from scratch to run in the cloud, while existing workloads are migrated to the cloud as resources allow. Once migration is over, the data center is eliminated. AWS is the dominant representative of this approach.
Hybrid cloud, where customers split their workloads between the public and private cloud. This approach is led by Microsoft Azure, and further boosted by the recent partnership with Hewlett Packard Enterprise as a major systems integrator.
When it comes to enterprise adoption, each approach has its own characteristics, as well as associated pros and cons.
The single source cloud is characterized by the following:
- The ultimate objective is to eliminate the entire on-premises infrastructure and do an “all-in” migration of existing workloads, i.e., migrate all apps, data, and active workloads into a single cloud. In particular, there is limited interest in hybrid-clouds, except for migration tools that serve as an interim solution until migration is complete.
- Any new applications will be “born in the cloud” with no need for migration.
- In addition to Infrastructure (IaaS) services, extensive PaaS services are provided, as well as developer tools and monitoring and configuration services, which enable developers to deploy solutions based on the advanced services offered by the single source cloud provider.
The hybrid cloud has the following characteristics:
- The objective is to migrate appropriate workloads to the public cloud, and provide a framework that bridges the gap between private and public clouds in terms of management and execution transparency.
- For even greater flexibility, organizations can leverage emerging technologies that let administrators dynamically change the placement of workloads between private and across different public clouds, based on considerations such as cost, capacity, and performance.
There are several workload characteristics that can help IT managers decide where to run workloads in hybrid cloud environments:
- I/O intensive versus CPU/memory intensive: workloads that manipulate large amounts of sensitive data lend themselves to on-premises execution, as moving the data is costly, or you incur a performance penalty of accessing data remotely over a WAN. In contrast, CPU-intensive workloads with minimal state and scale-out architecture may run better in the cloud, where compute instances can be added as needed and paid for only when used.
- Interactive versus batch: Workloads that involve “chatty” and real-time interaction with users or other on-premises components may be negatively impacted by the latency introduced by cloud-based placement. In contrast, batch workloads that need throughput but are not latency-sensitive often benefit from the scalability and cost-effectiveness of the public cloud.
- Existing legacy versus newly developed: Many data center workloads were designed for on-premises execution and cannot be easily migrated to a cloud environment. For instance, the assumption of low-latency, high throughput mass block storage arrays is unlikely to be feasible in a cloud environment. Similarly, assumptions about network configuration and management are not directly applicable in a cloud environment, which uses different methods and policies and may require extensive changes.
- Compliance and other concerns related to data exposure: Customers may not want or be able to move storage permanently to the cloud due to regulatory compliance or other concerns.
In considering whether to use a single-source or hybrid cloud strategy, evaluate the following pros and cons. Here are the advantages of single source public cloud strategy:
- Ease of management – Customers operate and manage one virtual infrastructure from a centralized interface, dramatically simplifying infrastructure management.
- Consistent performance, functionality, scalability.
- “One stop shop,” a single vendor to deal with for all infrastructure needs.
- One integration – Application developers can leverage the growing number of platform services such as analytics, business intelligence, monitoring and security, to improve software quality as well as accelerate development time.
Disadvantages of a single source public cloud strategy:
- Fault tolerance: With a one-cloud strategy, customers place “all their eggs in one basket.” What happens when there is a major cloud outage? At the very least, customers should consider a cross-cloud disaster recovery strategy to help mitigate this risk.
- Dealing with workloads that fail to migrate: Several surveys show that close to 50 percent of cloud migration projects fail (see, for instance, this link) for reasons including security, compliance, capacity planning, provisioning and billing issues. With a single source cloud strategy, there is no alternative. Research from Gartner suggests that “83 percent of data migration projects either fail outright, or suffer significant cost overruns or delays.”
- Lock-in: Many customers are concerned about being locked into a single cloud vendor for various reasons. Using a single vendor affects the ability to negotiate pricing and there is no safety net if performance for certain workloads does not meet expectations.
- Compliance: Customers may not be allowed to run certain workloads in the public cloud due to regulatory constraints.
- Best of breed compute environment: Certain workloads demand strict performance and high availability SLAs which might not be available or are prohibitively expensive in the public cloud. Other workloads might be able to take advantage of some platform facilities such as security mechanisms, or platform services that are only available on one cloud and not the other. In such cases, the dependency on a single IaaS cloud vendor might be too limiting.
So What Should Enterprises Do?
So what’s the best cloud strategy? The answer is – it depends. IT managers should ask the following questions: Do we have workloads that can only run on-premises or in a specific cloud? Do my workloads require massive changes to adapt them for cloud, or are they third-party apps that I can’t modify? Do I have to write off existing infrastructure investments in my data center? Will I have to create a new backup plan that can complicate migration?
If the answer to these questions is no, then a single source cloud strategy may be the right choice, since it does bring the advantages of simplicity, consistency and the ability to take advantage of cloud-specific platform services to enhance the development of cloud services. For all others, a hybrid cloud strategy is the logical choice.
Issy Ben-Shaul is the CEO and co-founder of Velostrata and a serial entrepreneur. Prior to co-founding Velostrata, Issy was the CTO for VMware’s Mirage desktop virtualization product and joined VMware through the acquisition of Wanova, a desktop virtualization company founded by himself and Ilan Kessler, in 2012. Prior to Wanova, Ben-Shaul was CTO for the Application Delivery Business Unit at Cisco Systems, which he joined in 2004 via the acquisition of Actona Technologies, where he was co-founder, and CTO. Before founding Actona, Issy was a tenured faculty member at the Technion, Israel Institute of Technology, where he worked on design and implementation of wide-area distributed systems, and prior to the Technion, he was a staff member at the IBM Haifa Research Laboratory. Ben-Shaul has published over 40 papers and holds 14 patents in the areas of WAN optimization and distributed systems. He holds a PhD and master’s degree in computer science from Columbia University and holds a bachelor’s degree in mathematics and computer science from Tel Aviv University.