Covering Disruptive Technology Powering Business in The Digital Age

Written by: Izzat Najmi, Journalist, AOPG.

Peter DeSantis, Senior Vice President of AWS Utility Computing, announced some big updates in his “Monday Night Live with Peter DeSantis” keynote that took place in the Venetian Ballroom, Las Vegas.

At the evening keynote of AWS re:Invent 2022, AWS revealed a number of updates to its cloud infrastructure, including a new version of the Nitro hypervisor, additional instance types, and a new version of the Arm-based Graviton chips built exclusively for high-performance computing. For workloads that rely largely on vector instructions, the new Graviton3E chip claim a 35% boost in performance.

We should expect new AWS EC2 instance types, the first of which is conceptually named “Hpc7G,” to be powered by these new chips. Depending on the size you choose, this new instance type can support up to 64 vCPUs and 128 GB of RAM. However, access to these instances will not be possible until early 2023. Moreover, AWS is releasing the Graviton3E instance type, which is designed to handle more network-intensive applications (c7gn).

Peter DeSantis, Senior Vice President of AWS Utility Computing

There are also new Xeon-based devices based on Intel’s Ice Lake architecture.

The new instances will run on AWS’s recently released Nitro 5 hardware hypervisor. The new Nitro v5 version boasts a 60% increase in Frames Per Second (FPS), a 40% increase in performance per watt, and a much lower latency. The AWS team achieved this by significantly increasing the transistor count of the Nitro chips they had developed for this purpose.

AWS also announced at the same time, another update to Lambda, AWS’s serverless platform, and it addresses a major complaint that users have with the service. Even with enhancements like the quicker Firecracker microVMs, Lambda still takes a bit to shut down the virtual machine when a function has not been utilised for a long time. SnapStart is AWS’s response to this problem; it allows users to save snapshots of their Lambda functions and immediately deploy them without going through the setup procedure again.

However, as Peter DeSantis pointed out in his speech today, Lambda (and all other serverless platforms) were pretty much created for spiky workloads, and this has long been one of the greatest complaints about Lambda. By utilising its Firecracker microVMs, AWS has reduced cold start times from several seconds to well under a second. Firecracker’s Snapshotting function has allowed the company to make the hitherto unachievable claim that cold start times can be reduced by 90%.

All current Lambda users have access to this new capability, albeit it must be activated for pre-existing Lambda functions and currently only supports Java functions using the Corretto runtime.

Once activated, the function will go through its usual start-up procedure upon being called for the first time. After that, an encrypted copy of the current state of memory and disk is saved in a cache for later use. The next time the function is called, Lambda will use the cache to kick off the execution. A 14-day inactivity period triggers the purge of cached snapshots. Thankfully, with the release of SnapStart, AWS has eliminated the previously annoying Lambda cold start latency.

DeSantis echoed this sentiment, noting that such enhancements will encourage more users to migrate their workloads to a service like Lambda. It was evident to the corporation even before the release of Firecracker on Lambda, he said.

(0)(0)