AWS loves itself some compute

While AWS made many, many announcements at its re:Invent event, the reworking of its family of instances, and availability of several interesting use case capabilities are likely to turn heads.


Andy Jassy onstage during his keynote

At its re:Invent show last week, Amazon Web Services [AWS] significantly expanded the number of instance options in its inventory. The company unveiled four new instance families coming your way. They also announced the ability to attach GPU to any instance, new easy-to generate Virtual Private Servers for the low end of the market, and the ability to easily attach custom-written FPGA accelerations at the high end.

The new instances were announced by AWS CEO Andy Jassy, who spoke at the company’s fifth annual re:invent conference to 32,000 attendees – eight times the audience at the first show.

Jassy strongly emphasized that AWS simply offers much more choice than its cloud competition – on several levels.

“This year we believe we will have close to 1000 new features,” Jassy said in his keynote. “Every day, you wake up with a choice of three new capabilities you can choose to use or not.”

Jassy noted that while lots of companies have a relational database service, the AWS one has six different engines, including one of their own design.

“You won’t find more than a third of those anywhere else,” he stressed. “We have a huge differentiation in capability and control, which is very important to customers.”

Part of this differentiation is also the nine separate instance families AWS has available, from the Burstable T2 at the low end to the G2 and P2 at the high end, with others optimized for different use cases in between.

“Nobody else even has half of these,” Jassy said. “This type of breadth and capability is incredibly important. You know your workloads aren’t vanilla. They are constrained by different pieces. It is so much easier to build what you want to build if you have the right tools at your disposal.”

To this end, Jassy announced a plethora of announcements around instances, to expand options and use case functionality.

“We are constantly iterating and reinventing in our instance families,” he said. “So we have four new instance families coming your way. We are also announcing the ability to attach GPU to any of our instance families, Virtual Private Servers made simple and FPGA instances at your disposal. I think its fair to say we love ourselves some compute.”

The T2 family, which has been the low end of AWS offerings, is being expanded at its higher end.

“The T2 family is good for simple websites and blogs and test/dev environments. Jassy said. “However, we started getting feedback from some customers that they wanted to take advantage of these characteristics of the T2 family, but they needed larger memory footprints. So today we have launched two new instance sizes in our T2 family.” The high end used to be 8 GiB. That has now been expanded with new 16 GiB and 32 GiB models.

AWS is also building on its R3 family, for memory intensive applications like high performance databases, distributed memory caches, in-memory analytics, genomic analysis, and enterprise applications like Sharepoint.

“Available today is a new memory optimized family, the R4,” Jassy said. “It has twice as much memory as the R3, twice as fast memory and faster throughput with DDR4s.”

An upgrade on the I2 family for IO intensive applications like NoSQL databases, scale out transactional databases, data warehousing and Hadoop, will be available in Q1. The new I3 family will have over 9 times the IOPS as the I2, over twice as much storage, over two times the memory and twice the number of vCPUs.

The C4 family, for compute intensive applications like web servers, batch processing, distributed analytics, and ad serving is also being pgraded with a new C5 family, also scheduled for Q1 availability. The C5 will feature the newest Intel processors, Skylake. It will have twice the performance for floating point, transcoding and machine learning applications, twice as many vCPUs, 3 times the throughput and 2.4 times the memory of the C4.

Jassy also announced the preview of a new service – Elastic GPUs for EC2.

“Customers have told us there are times when they want a little GPU, but don’t want a GPU instance, which is more expensive,” he said, noting that this is common with gaming companies. Elastic GPUs for EC2 lets Graphics GPUs be used as if they were EBS volumes.

“It will allow you to attach GPU to any of our nine instances,” Jassy added. “It’s super useful if you need some amount of GPU but don’t need the full GPU instances.”

Jassy also announced Amazon Lightsail, a virtual private server for the low end of the market, that doesn’t want to deal even with the limited management responsibilities of the EC2.

“Some customers say ‘all I want to do is spin up a virtual private server,’ so for them we are launching a new service, Amazon Lightsail. This is VPS made easy. It is incredibly easy to launch. You choose whatever image you want. You then choose from one of five bundles, give it a name, hit create and you have a Virtual Private Server.” AWS does everything else behind the scenes, although customers do have the option to open the hood themselves if they want.

“This starts at $5 dollars a month, so is very cost effective,” Jassy said.

On the other side of the spectrum, for the high end, are new F1 instances, a new family with customizable Field Programmable Gate Arrays, which are available now in preview. Jassy said that these are for users who, like AWS, are focused on hardware acceleration.

“These let you build and write your own accelerations and attach them to F1 instances,” he said, noting that a hardware development kit is available.

“We will make your FPGA accelerations available in the AWS marketplace to offer to consumers as well,” Jassy added.