The public cloud made a big mark in IT, enabling companies to more flexibly manage server capacity. Instead of having large, upfront capital expenses for IT systems, organizations that adopt infrastructure as a service can rent compute capacity, along with millions of other “tenants,” and only pay for what they use. Now, it’s also possible to add or subtract virtual servers from a public cloud cluster in minute increments.
Public clouds became economically feasible through huge buying power. It’s no coincidence that Google, Amazon Web Services (AWS) and Microsoft, the “big three” of the public cloud IaaS market, have acquired huge volumes of computer gear. Moreover, they have operating models that reduce costs, such as minimized hardware maintenance and the use of containers.
The economic scale of these cloud service providers (CSPs) has, in general, created IaaS benefits for cloud users. Public cloud providers have raged price wars for years, as emerging, commercial off-the-shelf-based technologies enable new ways to build tools for their tenants and performance gains lead to reductions in the cost-per-compute cycle.
So, what does this all mean to IaaS tenants? While admins might be reluctant to give up control to a third party, these CSPs have made the entry cost for the cloud so low that, for many companies, it’s become a no-brainer. For some segments of the market, CSPs have created turnkey environments to make it even simpler. AWS, for example, offers everything from webpage design to services that handle online transactions. Google has begun to specialize in artificial intelligence and machine learning, while Microsoft’s Azure aims for seamless integration with in-house Windows operations.
In general, the definition and value of IaaS will continue to evolve as users tap into new services around serverless computing, big data and more.
An overview of IaaS benefits
Public IaaS provides a venue where tenants can scale compute and storage services on demand to match current needs. IaaS is a good match for many different workloads, from sandboxed prototyping and testing to the deployment and upgrading of complex apps.
Today’s large IaaS offerings are global in reach. Providers have data centers worldwide, which enables admins to spread the compute cluster over multiple geographic locations for increased resiliency. For example, if a hurricane knocks out a local AWS facility, admins can fire up a workload in another state or country. This is high-quality disaster recovery on a low-cost budget.
Other IaaS benefits include management, provisioning and configuration features, such as CSP-provided templates, which can save admins time. One caveat, however, is that the cloud rapidly evolves and requires ongoing performance, usage and cost monitoring. If admins fail to do this, it could result in less satisfied users and higher costs. For example, if they don’t track and prevent the overprovisioning of cloud resources, organizations’ monthly charges will quickly climb.
IaaS compute options
When organizations move to IaaS, they have a choice of service level. Basic IaaS delivers a virtual instance that acts like a real server. Virtual switching couples these servers together, and admins can connect them to either networked object or block storage.
Tools orchestrate requests for instances, and IaaS providers price out those instances by memory size and compute power. Each CSP has a range of instance types, with some “bigger” instances having as many as four CPUs or terabytes of dynamic RAM, as well as added performance boosters, such as local solid-state drive (SSD) storage or GPU accelerators. With these new offerings, it’s fair to say that there aren’t too many in-house configurations that the cloud can’t support.
While this range of instance options is one of the top IaaS benefits, organizations must carefully choose how they purchase compute power. They can buy compute time by the minute, although there’s a downside to this. Providers generally charge more for short-term instances than instances rented on a monthly basis, and some CSPs even offer annual contracts at lower prices. The best way to handle this is to build a baseline usage model and fulfill that with longer contracts at low prices, with monthly or minute instances to cover extra spikes.
Admins can use automation tools to help shape their instance decisions and don’t necessarily have to ask a CSP.
IaaS performance, security and storage
With public cloud, users share a provider’s huge compute cluster with up to millions of other users. But with hardware advances from Intel and Advanced Micro Devices and great diligence in software design, public cloud has reached the point that, most of the time, admins won’t know they share resources with other tenants at all. The “noisy neighbor problem” — where one tenant’s instances use a significant number of resources, to the point where it can negatively affect other users’ workloads — continues to diminish.
Better preventative measures, coupled with the high SSD performance of newer and faster networks, make this issue rarer than in the past. With the noisy neighbor problem becoming less common, multi-tenancy is generally as efficient as completely separate servers.
In addition, because of the high level of automation in IaaS environments, as well as providers’ security investments, public cloud security likely exceeds what is available in most on-premises IT operations, even at the enterprise level. Still, security is not perfect. If a cloud tenant creates a virtual data center without proper authentication and firewalls, bad actors can still gain access. The good news is that all the major CSPs — not just the big three — have security support staff and educational materials available to provide best practices. Still, security is a responsibility that both the providers and tenants need to share and take seriously.
Finally, IaaS benefits also include access to a range of cloud storage technologies and features. Block and object stores are standard today in the public cloud and can be petabytes in size, with data integrity features built in, including RAID or replication. Storage is tiered, and instance stores can be either fast or ultrafast, with IOPS guarantees determining the price.
Providers price networked storage per GB per month, but there are charges for extracting data based on GB reads. These are small fractions of a penny per GB, but they can add up. Local instance storage differs from network storage in that it only survives as long as the instance.