AWS launches new instance series for data-hungry workloads


Running analytics software and other storage-intensive workloads on Amazon Web Services is set to become much easier thanks to its latest infrastructure update. Inc.’s cloud service on Thursday launched a new family of instances branded as I3 that offers a generous amount of flash memory for performing high-speed data processing. AWS chief evangelist Jeff Barr detailed in a blog post that the underlying drives use NVMe, a communications protocol capable of shuffle information to and from storage much faster than older interfaces. The past few quarters have seen numerous array makers adopt the technology in a bid to lure chief information officers away from the public cloud.

Amazon’s new I3 family ups the ante by offering NVMe-powered virtual machines for as little as 15 cents an hour. That’s the base price for the i3.large, the smallest of the six instances in the series, which features a pair of virtual CPUs, 16 gigabytes of memory and over half a terabyte of NVMe flash.

On the other end of the price scale is the i3.16xlarge, which in turn provides 15.2 terabytes of storage capacity along with 64 processors to help users make the most out of their storage capacity. A single instance can handle 3.3 million I/O operations when processing 4 KB data blocks and provide up to 16 GBps of sequential throughput, considerably more than the previous-generation I2 series. It’s available from $4.99 per hour.

AWS’ Barr noted in his post that the base rates for the I3 series apply to instances bought under the On-Demand pricing model, the most popular but also most expensive of the plans on offer. The provider offers significant discounts for customers if they rent instances for an extended period of time or run them on unused infrastructure in its data centers when demand is low. According to AWS, the latter option makes it possible to shave up to 90 percent of the bill in some cases.

The I2 series is available immediately in 15 of the regions where the provider maintains cloud infrastructure. It was first previewed at the annual AWS re:Invent conference last November, which also saw the introduction of about half a dozen other new instance families. Among them was the F1 series, a lineup based on field-programmable gate arrays that can be customized for specific workloads to increase performance.

Photo: Robert Hof