While it’s too early to judge whether migrating workloads will help Google and Microsoft catch Amazon Web Services, their customers stand to benefit from lower costs for cloud computing.
That’s because their new offerings are aimed at extracting greater valueOpens a new window from data center hardware. Using a virtual machine – an industry term for software that mimics the actions of a central processing unit – Amazon’s competitors can create, execute and shift processing orders to available resources without significant gaps in service.
Both are chasing AWS, the Amazon subsidiary that pioneered so-called elastic computing in its market-leading cloud. It commands a 47% share of global spending for cloud computing and maintains a fluid pricing model for spare capacity.
Of course, virtualization isn’t newOpens a new window . Large corporations use the software to replicate processes that incorporate both standard and specialist hardware across countless landscapes and applications. In an effort to do more with less, Google and Microsoft are cutting features as they cut costs. For less time-sensitive tasks, they say, moving workloads to regions and racksOpens a new window and executing them during off-peak periods can make sense.
Slim down to scale up
Google’s claims that renting its newly released E2 family of virtual machinesOpens a new window  can trim more than 30% from the ownership cost of the processing capacity. Google’s slimmed-down offering doesn’t support the graphics processors or local solid-state drives that demand higher levels of backup.
Instead, Google is touting its ability to manage compute and storage resources in ways that allow customers to process workloads without disruptions. The trick is moving the workloads seamlessly around the data center’s hardware.A technician uses a hypervisor scheduler to map the virtual machine’s CPUs onto their physical counterparts, automatically registering as lines of code in execution strings. Virtual RAM works the same way, with host pages that are activated when a guest page is accessed.
In both cases, positions remained fixed until the virtual machine signals there’s no more need for them.
Azure, on the spot
Microsoft Azure’s virtual machines can be jettisonedOpens a new window when demand for processing exceeds supply. The VMs are best for undertakings less sensitive to time and region requirements. Such on-demand instances can cut the cost of software development and testing.
With Azure Spot, prices and capacity dictate whether cut-rate processing is evicted from servers when demand rises. A dashboard shows both figures across regions and customers can choose what to pay and where to run those processes.
Microsoft lets users create VMs in a variety of APIs and users get prompts when their VMs are about to be terminated. They then can choose whether to pay up to keep the machines online.
AWS improves EC2
To be sure, VM rollouts from Google and Microsoft are timed to steal the thunder from AWS. At its recent “re-Invent†conference, AWS announced a host of upgradesOpens a new window to its EC2 service that also rely on software to manage volumes of workloads.
These include a souped-up combination of dedicated software and bare-metal servers called Nitro that can cut the overhead for VMs. Users also can run EC2 instances with Graviton2 processors, Amazon’s 64-bit architecture that uses ARM cores.
Company execs say customers now can call on Image Builder for more process-heavy projects. A new EC2 element lets them conduct machine-learning inferences using spot VMs.
Emergent tech considerations
Given that AWS began marketing EC2 a decade agoOpens a new window , it’s small wonder that on-demand has reached emergent technologies such as artificial intelligence. But that doesn’t mean that cut-rate computing is right for every application.
Providers stress that fault tolerance is the best indicator of business benefit. Along with batch processing and back-office workloads, virtual desktops and app servers are primary targets for spot-instance VMs.
Stateless applications that shed data after a single process run also may be more economically executed on-demand and off-peak.
Performance penalties will work to limit use cases. For workloads that can withstand eviction when demand spikes, companies can see bottom-line improvement if someone else foots the bill for infrastructure upkeep.