AWS Blog

Natural Language Processing at Clemson University – 1.1 Million vCPUs & EC2 Spot Instances

My colleague Sanjay Padhi shared the guest post below in order to recognize an important milestone in the use of EC2 Spot Instances.

Jeff;


A group of researchers from Clemson University achieved a remarkable milestone while studying topic modeling, an important component of machine learning associated with natural language processing, breaking the record for creating the largest high-performance cluster in the cloud by using more than 1,100,000 vCPUs on Amazon EC2 Spot Instances running in a single AWS region. The researchers conducted nearly half a million topic modeling experiments to study how human language is processed by computers. Topic modeling helps in discovering the underlying themes that are present across a collection of documents. Topic models are important because they are used to forecast business trends and help in making policy or funding decisions. These topic models can be run with many different parameters and the goal of the experiments is to explore how these parameters affect the model outputs.

The Experiment
Professor Amy Apon, Co-Director of the Complex Systems, Analytics and Visualization Institute at Clemson University with Professor Alexander Herzog and graduate students Brandon Posey and Christopher Gropp in collaboration with members of the AWS team as well as AWS Partner Omnibond performed the experiments.  They used software infrastructure based on CloudyCluster that provisions high performance computing clusters on dynamically allocated AWS resources using Amazon EC2 Spot Fleet. Spot Fleet is a collection of biddable spot instances in EC2 responsible for maintaining a target capacity specified during the request. The SLURM scheduler was used as an overlay virtual workload manager for the data analytics workflows. The team developed additional provisioning and workflow automation software as shown below for the design and orchestration of the experiments. This setup allowed them to evaluate various topic models on different data sets with massively parallel parameter sweeps on dynamically allocated AWS resources. This framework can easily be used beyond the current study for other scientific applications that use parallel computing.

Ramping to 1.1 Million vCPUs
The figure below shows elastic, automatic expansion of resources as a function of time, in the US East (Northern Virginia) Region. At just after 21:40 (GMT-1) on Aug. 26, 2017, the number of vCPUs utilized was 1,119,196. Clemson researchers also took advantage of the new per-second billing for the EC2 instances that they launched. The vCPU count usage is comparable to the core count on the largest supercomputers in the world.

Here’s the breakdown of the EC2 instance types that they used:

Campus resources at Clemson funded by the National Science Foundation were used to determine an effective configuration for the AWS experiments as compared to campus resources, and the AWS cloud resources complement the campus resources for large-scale experiments.

Meet the Team
Here’s the team that ran the experiment (Professor Alexander Herzog, graduate students Christopher Gropp and Brandon Posey, and Professor Amy Apon):

Professor Apon said about the experiment:

I am absolutely thrilled with the outcome of this experiment. The graduate students on the project are amazing. They used resources from AWS and Omnibond and developed a new software infrastructure to perform research at a scale and time-to-completion not possible with only campus resources. Per-second billing was a key enabler of these experiments.

Boyd Wilson (CEO, Omnibond, member of the AWS Partner Network) told me:

Participating in this project was exciting, seeing how the Clemson team developed a provisioning and workflow automation tool that tied into CloudyCluster to build a huge Spot Fleet supercomputer in a single region in AWS was outstanding.

About the Experiment
The experiments test parameter combinations on a range of topics and other parameters used in the topic model. The topic model outputs are stored in Amazon S3 and are currently being analyzed. The models have been applied to 17 years of computer science journal abstracts (533,560 documents and 32,551,540 words) and full text papers from the NIPS (Neural Information Processing Systems) Conference (2,484 documents and 3,280,697 words). This study allows the research team to systematically measure and analyze the impact of parameters and model selection on model convergence, topic composition and quality.

Looking Forward
This study constitutes an interaction between computer science, artificial intelligence, and high performance computing. Papers describing the full study are being submitted for peer-reviewed publication. I hope that you enjoyed this brief insight into the ways in which AWS is helping to break the boundaries in the frontiers of natural language processing!

Sanjay Padhi, Ph.D, AWS Research and Technical Computing

 

AWS Pinpoint Launches Two-Way Text Messaging

Last week Amazon Pinpoint launched AWS Global SMS two-way text messaging and we didn’t get an opportunity to cover the launch. AWS Pinpoint users can now programmaticaly respond to their end-users’ text messages. Users can provision both short codes and long codes (10-digit phone numbers) which send inbound messages to an SNS topic. Let’s take a look.

First I’ll navigate to the Pinpoint console where I’ll use the “Create a project in Mobile Hub” button in the top right corner. I’ll follow the steps in the wizard until the project is created.

Next, in the Pinpoint console I’ll click “Account Settings” in the top-right of the console window.

At the bottom of the Account Settings page there is a “Number Settings” section. If you don’t already have any short codes or long codes provisioned you’ll have to open a support ticket to request one. It can take multiple weeks for a short code to be approved by all carriers. Long codes are typically easier to provision.

Since I already have a few numbers provisioned I’ll use one of them now by clicking on it which brings me to that number’s configuration page.

Here I’ll enable 2-way SMS and create an SNS topic for messages.

I could create a quick lambda function to trigger on the SNS topic messages and then respond again with pinpoint.


import boto3
pinpoint = boto3.client('pinpoint')


def lambda_handler(event, context):
    pinpoint.send_messages(
        ApplicationId='557d87b57bdb499f8b5eef575435d3b8',
        MessageRequest={
            'Addresses': {
                event['Records'][0]['Sns']['originationNumber']: {'ChannelType': 'SMS'}
            },
            'MessageConfiguration': {
                'SMSMessage': {
                    'Body': 'Vim is the best!',
                    'MessageType': 'TRANSACTIONAL'
                }
            }
        }
    )

But pinpoint is extremely full-featured and you’re not limited to this simple message type. You can define rich messaging campaigns with various substitutions based on stored user data.

SMS is an area of continued investment for AWS so you can expect to see more advances and improvements as customers give us feedback on these new features. Let us know if you build anything cool with this!

Useful Links

In the Works – AWS Region in the Middle East

Last year we launched new AWS Regions in Canada, India, Korea, the UK, and the United States, and announced that new regions are coming to China, France, Hong Kong, Sweden, and a second GovCloud Region in the US throughout 2017 and 2018.

Middle East Region by Early 2019
Today, I am happy to announce that we will be opening an AWS Region in the Middle East by early 2019. The new Region will be based in Bahrain, will be comprised of three Availability Zones at launch, and will give AWS customers and partners the ability to run their workloads and store their data in the Middle East.

AWS customers are already making use of 44 Availability Zones across 16 geographic regions. Today’s announcement brings the total number of global regions (operational and in the works) up to 22.

UAE Edge Location in 2018
We also plan to open an edge location in the UAE in the first quarter of 2018. This will bring Amazon CloudFront, Amazon Route 53, AWS Shield, and AWS WAF to the region, adding to our existing set of 84 points of presence world-wide.

These announcements add to our continued investment in the Middle East. Earlier this year we announced the opening of AWS offices in Dubai, UAE and Manama, Bahrain. Prior to this we have supported the growth of technology education in the region with AWS Educate and have supported the growth of new businesses through AWS Activate for many years.

The addition of AWS infrastructure in the Middle East will help countries across the region to innovate, grow their economies, and pursue their vision plans (Saudi Vision 2030, UAE Vision 2021, Bahrain Vision 2030, and so forth).

Talk to Us
As always, we are looking forward to serving new and existing customers in the Middle East and working with partners across the region. Of course, the new Region will also be open to existing AWS customers who would like to serve users in the Middle East.

To learn more about the AWS Middle East Region feel free to contact our team at aws-me@amazon.com .

If you are interested in joining the team and would like to learn more about AWS positions in the Middle East, take a look at the Amazon Jobs site.

Jeff;

Catching Up on Some Recent AWS Launches and Publications

As I have noted in the past, the AWS Blog Team is working hard to make sure that you know about as many AWS launches and publications as possible, without totally burying you in content! As part of our balancing act, we will occasionally publish catch-up posts to clear our queues and to bring more information to your attention. Here’s what I have in store for you today:

  • Monitoring for Cross-Region Replication of S3 Objects
  • Tags for Spot Fleet Instances
  • PCI DSS Compliance for 12 More Services
  • HIPAA Eligibility for WorkDocs
  • VPC Resizing
  • AppStream 2.0 Graphics Design Instances
  • AMS Connector App for ServiceNow
  • Regtech in the Cloud
  • New & Revised Quick Starts

Let’s jump right in!

Monitoring for Cross-Region Replication of S3 Objects
I told you about cross-region replication for S3 a couple of years ago. As I showed you at the time, you simply enable versioning for the source bucket and then choose a destination region and bucket. You can check the replication status manually, or you can create an inventory (daily or weekly) of the source and destination buckets.

The Cross-Region Replication Monitor (CRR Monitor for short) solution checks the replication status of objects across regions and gives you metrics and failure notifications in near real-time.

To learn more, read the CRR Monitor Implementation Guide and then use the AWS CloudFormation template to Deploy the CRR Monitor.

Tags for Spot Instances
Spot Instances and Spot Fleets (collections of Spot Instances) give you access to spare compute capacity. We recently gave you the ability to enter tags (key/value pairs) as part of your spot requests and to have those tags applied to the EC2 instances launched to fulfill the request:

To learn more, read Tag Your Spot Fleet EC2 Instances.

PCI DSS Compliance for 12 More Services
As first announced on the AWS Security Blog, we recently added 12 more services to our PCI DSS compliance program, raising the total number of in-scope services to 42. To learn more, check out our Compliance Resources.

HIPAA Eligibility for WorkDocs
In other compliance news, we announced that Amazon WorkDocs has achieved HIPAA eligibility and PCI DSS compliance in all AWS Regions where WorkDocs is available.

VPC Resizing
This feature allows you to extend an existing Virtual Private Cloud (VPC) by adding additional blocks of addresses. This gives you more flexibility and should help you to deal with growth. You can add up to four secondary /16 CIDRs per VPC. You can also edit the secondary CIDRs by deleting them and adding new ones. Simply select the VPC and choose Edit CIDRs from the menu:

Then add or remove CIDR blocks as desired:

To learn more, read about VPCs and Subnets.

AppStream 2.0 Graphics Design Instances
Powered by AMD FirePro S7150x2 Server GPUs and equipped with AMD Multiuser GPU technology, the new Graphics Design instances for Amazon AppStream 2.0 will let you run and stream graphics applications more cost-effectively than ever. The instances are available in four sizes, with 2-16 vCPUs and 7.5 GB to 61 GB of memory.

To learn more, read Introducing Amazon AppStream 2.0 Graphics Design, a New Lower Costs Instance Type for Streaming Graphics Applications.

AMS Connector App for ServiceNow
AWS Managed Services (AMS) provides Infrastructure Operations Management for the Enterprise. Designed to accelerate cloud adoption, it automates common operations such as change requests, patch management, security and backup.

The new AMS integration App for ServiceNow lets you interact with AMS from within ServiceNow, with no need for any custom development or API integration.

To learn more, read Cloud Management Made Easier: AWS Managed Services Now Integrates with ServiceNow.

Regtech in the Cloud
Regtech (as I learned while writing this), is short for regulatory technology, and is all about using innovative technology such as cloud computing, analytics, and machine learning to address regulatory challenges.

Working together with APN Consulting Partner Cognizant, TABB Group recently published a thought leadership paper that explains why regulations and compliance pose huge challenges for our customers in the financial services, and shows how AWS can help!

New & Revised Quick Starts
Our Quick Starts team has been cranking out new solutions and making significant updates to the existing ones. Here’s a roster:

Alfresco Content Services (v2) Atlassian Confluence Confluent Platform Data Lake
Datastax Enterprise GitHub Enterprise Hashicorp Nomad HIPAA
Hybrid Data Lake with Wandisco Fusion IBM MQ IBM Spectrum Scale Informatica EIC
Magento (v2) Linux Bastion (v2) Modern Data Warehouse with Tableau MongoDB (v2)
NetApp ONTAP NGINX (v2) RD Gateway Red Hat Openshift
SAS Grid SIOS Datakeeper StorReduce SQL Server (v2)

And that’s all I have for today!

Jeff;

New – Stop & Resume Workloads on EC2 Spot Instances

EC2 Spot Instances give you access to spare EC2 compute capacity at up to 90% off of the On-Demand rates. Starting with the ability to request a specific number of instances of a particular size, we made Spot Instances even more useful and flexible with support for Spot Fleets and Auto Scaling Spot Fleets, allowing you to maintain any desired level of compute capacity.

EC2 users have long had the ability to stop running instances while leaving EBS volumes attached, opening the door to applications that automatically pick up where they left off when the instance starts running again.

Stop and Resume Spot Workloads
Today we are blending these two important features, allowing you to set up Spot bids and Spot Fleets that respond by stopping (rather than terminating) instances when capacity is no longer available at or below your bid price. EBS volumes attached to stopped instances remain intact, as does the EBS-backed root volume. When capacity becomes available, the instances are started and can keep on going without having to spend time provisioning applications, setting up EBS volumes, downloading data, joining network domains, and so forth.

Many AWS customers have enhanced their applications to create and make use of checkpoints, adding some resilience and gaining the ability to take advantage of EC2’s start/stop feature in the process. These customers will now be able to run these applications on Spot Instances, with savings that average 70-90%.

While the instances are stopped, you can modify the EBS Optimization, User data, Ramdisk ID, and Delete on Termination attributes. Stopped Spot Instances do not incur any charges for compute time; space for attached EBS volumes is charged at the usual rates.

Here’s how you create a Spot bid or Spot Fleet and specify the use of stop/start:

Things to Know
This feature is available now and you can start using it today in all AWS Regions where Spot Instances are available. It is designed to work well in conjunction with the new per-second billing for EC2 instances and EBS volumes, with the potential for another dimension of cost savings over and above that provided by Spot Instances.

EBS volumes always exist within a particular Availability Zone (AZ). As a result, Spot and Spot Fleet requests that specify a particular AZ will always restart in that AZ.

Take care when using this feature in conjunction with Spot Fleets that have the potential to span a wide variety of instance types. Because the composition of the fleet can change over time, you need to pay attention to your account’s limits for IP addresses and EBS volumes.

I’m looking forward to hearing about the new and creative uses that you’ll come up with for this feature. If you thought that your application was not a good fit for Spot Instances, or if the overhead needed to handle interruptions was too high, it is time to take another look!

Jeff;

 

New – Per-Second Billing for EC2 Instances and EBS Volumes

Back in the old days, you needed to buy or lease a server if you needed access to compute power. When we launched EC2 back in 2006, the ability to use an instance for an hour, and to pay only for that hour, was big news. The pay-as-you-go model inspired our customers to think about new ways to develop, test, and run applications of all types.

Today, services like AWS Lambda prove that we can do a lot of useful work in a short time. Many of our customers are dreaming up applications for EC2 that can make good use of a large number of instances for shorter amounts of time, sometimes just a few minutes.

Per-Second Billing for EC2 and EBS
Effective October 2nd, usage of Linux instances that are launched in On-Demand, Reserved, and Spot form will be billed in one-second increments. Similarly, provisioned storage for EBS volumes will be billed in one-second increments.

Per-second billing also applies to several other AWS services:

Amazon EMR – Our customers add capacity to their EMR clusters in order to get their results more quickly. With per-second billing for the EC2 instances in the clusters, adding nodes is more cost-effective than ever. To learn more, read Amazon EMR Now Supports Per-Second Billing.

AWS Batch – Many of the batch jobs that our customers run complete in less than an hour. AWS Batch already launches and terminates Spot Instances; with per-second billing for the EC2 instances, batch processing will become even more economical.

Elastic GPUs – Usage of Elastic GPUs is billed by the second, with a 1 minute minimum.

Provisioned IOPSProvisioned IOPS for io1 EBS volumes is billed by the second.

Some of our more sophisticated customers have built systems to get the most value from EC2 by strategically choosing the most advantageous target instances when managing their gaming, ad tech, or 3D rendering fleets. Per-second billing obviates the need for this extra layer of instance management, and brings the costs savings to all customers and all workloads.

While this will result in a price reduction for many workloads (and you know we love price reductions), I don’t think that’s the most important aspect of this change. I believe that this change will inspire you to innovate and to think about your compute-bound problems in new ways. How can you use it to improve your support for continuous integration? Can it change the way that you provision transient environments for your dev and test workloads? What about your analytics, batch processing, and 3D rendering?

One of the many advantages of cloud computing is the elastic nature of provisioning or deprovisioning resources as you need them. By billing usage down to the second we will enable customers to level up their elasticity, save money, and customers will be positioned to take advantage of continuing advances in computing.

Things to Know
This change is effective in all AWS Regions and will be effective October 2, for all Linux instances that are newly launched or already running. There is a 1 minute minimum charge per-instance.

Per-second billing is not currently applicable to instances running Microsoft Windows or Linux distributions that have a separate hourly charge. Marketplace AMIs that do not have a separate hourly charge are eligible for per-second billing.

List prices and Spot Market prices are still listed on a per-hour basis, but bills are calculated down to the second, as is Reserved Instance usage (you can launch, use, and terminate multiple instances within an hour and get the Reserved Instance Benefit for all of the instances). Also, bills will show times in decimal form, like this:

The Dedicated Per Region Fee, EBS Snapshots, and products in AWS Marketplace are still billed on an hourly basis.

Jeff;

 

AWS Partner Webinar Series – September & October 2017

The wait is over. September and October’s Partner Webinars have officially arrived! In case you missed the intro last month, the AWS Partner Webinar Series is a selection of live and recorded presentations covering a broad range of topics at varying technical levels and scale. A little different from our AWS Online TechTalks, each AWS Partner Webinar is hosted by an AWS solutions architect and an AWS Competency Partner who has successfully helped customers evaluate and implement the tools, techniques, and technologies of AWS.

 

 

September & October Partner Webinars:

 

SAP Migration
Velocity: How EIS Reduced Costs by 20% and Optimized SAP by Leveraging the Cloud
September 19, 2017 | 10:00 AM PDT

 

Mactores: SAP on AWS: How UCT is Experiencing Better Performance on AWS While Saving 60% in Infrastructure Costs with Mactores
September 19, 2017 | 1:00 PM PDT

 

Accenture: Reduce Operating Costs and Accelerate Efficiency by Migrating Your SAP Applications to AWS with Accenture
September 20, 2017 | 10:00 AM PDT

 

Capgemini: Accelerate your SAP HANA Migration with Capgemini & AWS FAST
September 21, 2017 | 10:00 AM PDT

 

Salesforce
Salesforce IoT: Monetize your IOT Investment with Salesforce and AWS
September 27, 2017 | 10:00 am PDT

 

Salesforce Heroku: Build Engaging Applications with Salesforce Heroku and AWS
October 10, 2017 | 10:00 AM PDT

 

Windows Migration
Cascadeo: How a National Transportation Software Provider Migrated a Mission-Critical Test Infrastructure to AWS with Cascadeo
September 26, 2017 | 10:00 AM PDT

 

Datapipe: Optimize App Performance and Security by Managing Microsoft Workloads on AWS with Datapipe
September 27, 2017 | 10:00 AM PDT

 

Datavail: Datavail Accelerates AWS Adoption for Sony DADC New Media Solutions
September 28, 2017 | 10:00 AM PDT

 

Life Sciences

SAP, Deloitte & Turbot: Life Sciences Compliance on AWS
October 4, 2017 | 10:00 AM PDT

 

Healthcare

AWS, ClearData & Cloudticity: Healthcare Compliance on AWS 
October 5, 2017 | 10:00 AM PDT

 

Storage

NetApp: Give your SaaS Data Security with NetApp and AWS

September 20, 2017 | 9:00 am PDT

N2WS: Learn How Goodwill Industries Ensures 24/7 Data Availability on AWS
October 10, 2017 | 8:00 AM PDT

 

Big Data

Zoomdata: Taking Complexity Out of Data Science with AWS and Zoomdata
October 10, 2017 | 10:00 AM PDT

 

Attunity: Cardinal Health: Moving Data to AWS in Real-Time with Attunity 
October 11, 2017 | 11:00 AM PDT

 

Splunk: How TrueCar Gains Actionable Insights with Splunk Cloud
October 18, 2017 | 9:00 AM PDT

Prime Day 2017 – Powered by AWS

The third annual Prime Day set another round of records for global orders, topping Black Friday and Cyber Monday, making it the biggest day in Amazon retail history. Over the course of the 30 hour event, tens of millions of Prime members purchased things like Echo Dots, Fire tablets, programmable pressure cookers, espresso machines, rechargeable batteries, and much more! July 11th also set a record for the number of new Prime memberships, as people signed up in order to take advantage of hundreds of thousands of deals. Amazon customers shopped online and made heavy use of the Amazon App, with mobile orders more than doubling from last Prime Day.

Powered by AWS
Last year I told you about How AWS Powered Amazon’s Biggest Day Ever, and shared what the team had learned with regard to preparation, automation, monitoring, and thinking big. All of those lessons still apply and you can read that post to learn more. Preparation for this year’s Prime Day (which started just days after Prime Day 2016 wrapped up) started by collecting and sharing best practices and identifying areas for improvement, proceeding to implementation and stress testing as the big day approached. Two of the best practices involve auditing and GameDay:

Auditing – This is a formal way for us to track preparations, identify risks, and to track progress against our objectives. Each team must respond to a series of detailed technical and operational questions that are designed to help them determine their readiness. On the technical side, questions could revolve around time to recovery after a database failure, including the all-important check of the TTL (time to live) for the CNAME. Operational questions address schedules for on-call personnel, points of contact, and ownership of services & instances.

GameDay – This practice (which I believe originated with former Amazonian Jesse Robbins), is intended to validate all of the capacity planning & preparation and to verify that all of the necessary operational practices are in place and work as expected. It introduces simulated failures and helps to train the team to identify and quickly resolve issues, building muscle memory in the process. It also tests failover and recovery capabilities, and can expose latent defects that are lurking under the covers. GameDays help teams to understand scaling drivers (page views, orders, and so forth) and gives them an opportunity to test their scaling practices. To learn more, read Resilience Engineering: Learning to Embrace Failure or watch the video: GameDay: Creating Resiliency Through Destruction.

Prime Day 2017 Metrics
So, how did we do this year?

The AWS teams checked their dashboards and log files, and were happy to share their metrics with me. Here are a few of the most interesting ones:

Block Storage – Use of Amazon Elastic Block Store (EBS) grew by 40% year-over-year, with aggregate data transfer jumping to 52 petabytes (a 50% increase) for the day and total I/O requests rising to 835 million (a 30% increase). The team told me that they loved the elasticity of EBS, and that they were able to ramp down on capacity after Prime Day concluded instead of being stuck with it.

NoSQL Database – Amazon DynamoDB requests from Alexa, the Amazon.com sites, and the Amazon fulfillment centers totaled 3.34 trillion, peaking at 12.9 million per second. According to the team, the extreme scale, consistent performance, and high availability of DynamoDB let them meet needs of Prime Day without breaking a sweat.

Stack Creation – Nearly 31,000 AWS CloudFormation stacks were created for Prime Day in order to bring additional AWS resources on line.

API Usage – AWS CloudTrail processed over 50 billion events and tracked more than 419 billion calls to various AWS APIs, all in support of Prime Day.

Configuration TrackingAWS Config generated over 14 million Configuration items for AWS resources.

You Can Do It
Running an event that is as large, complex, and mission-critical as Prime Day takes a lot of planning. If you have an event of this type in mind, please take a look at our new Infrastructure Event Readiness white paper. Inside, you will learn how to design and provision your applications to smoothly handle planned scaling events such as product launches or seasonal traffic spikes, with sections on automation, resiliency, cost optimization, event management, and more.

Jeff;

 

Now Available – EC2 Instances with 4 TB of Memory

Earlier this year I told you about our plan to launch EC2 instances with up to 16 TB of memory. Today I am happy to announce that the new x1e.32xlarge instances with 4 TB of DDR4 memory are available in four AWS Regions. As I wrote in my earlier post, these instances are designed to run SAP HANA and other memory intensive, in-memory applications. Many of our customers are already running production SAP applications on the existing x1.32xlarge instances. With today’s launch, these customers can now store and process far larger data sets, making them a great fit for larger production deployments.

Like the x1.32xlarge, the x1e.32xlarge is powered by quad socket Intel Xeon E7 8880 v3 Haswell processors running at 2.3GHz (128 vCPUs), with large L3 caches, plenty of memory bandwidth, and support for C-state and P-state management.

On the network side, the instances offer up to 25 Gbps of network bandwidth when launched within an EC2 placement group, powered by the Elastic Network Adapter (ENA), with support for up to 8 Elastic Network Interfaces (ENIs) per instance. The instances are EBS-optimized by default, with an additional 14 Gbps of dedicated bandwidth to your EBS volumes, and support for up to 80,000 IOPS per instance. Each instance also includes a pair of 1,920 GB SSD volumes.

A Few Notes
Here are a couple of things to keep in mind regarding the x1e.32xlarge:

SAP Certification – The x1e.32xlarge instances are our largest cloud-native instances certified and supported by SAP for production HANA deployments of SAP Business Suite on HANA (SoH), SAP Business Warehouse on HANA (BWoH), and the next-generation SAP S/4HANA ERP and SAP BW/4HANA data warehouse solution. If you are already running SAP HANA workloads on smaller X1 instances, scaling up will be quick and easy. The SAP HANA on the AWS Cloud Quick Start Reference Deployment has been updated and will help you to set up a deployment that follows SAP and AWS standards for high performance and reliability. The SAP HANA Hardware Directory and the SAP HANA Sizing Guidelines are also relevant.

Reserved Instances – The regional size flexibility for Reserved Instances does not apply across x1 and x1e.

Now Available
The x1e.32xlarge instances can be launched in On-Demand and Reserved Instance form via the AWS Management Console, AWS Command Line Interface (CLI), AWS SDKs, and AWS Marketplace in the US East (Northern Virginia), US West (Oregon), EU (Ireland), and Asia Pacific (Tokyo) Regions.

I would also like to make you aware of a couple of other upgrades to the X1 instances:

EBS – As part of today’s launch, existing X1 instances also support up to 14 Gbps of dedicated bandwidth to EBS, along with 80,000 IOPS per instance.

Network – Earlier this week, we announced that existing x1.32xlarge instances also support up to 25 Gbps of network bandwidth within placement groups.

Jeff;

New Network Load Balancer – Effortless Scaling to Millions of Requests per Second

Elastic Load Balancing (ELB) has been an important part of AWS since 2009, when it was launched as part of a three-pack that also included Auto Scaling and Amazon CloudWatch. Since that time we have added many features, and also introduced the Application Load Balancer. Designed to support application-level, content-based routing to applications that run in containers, Application Load Balancers pair well with microservices, streaming, and real-time workloads.

Over the years, our customers have used ELB to support web sites and applications that run at almost any scale — from simple sites running on a T2 instance or two, all the way up to complex applications that run on large fleets of higher-end instances and handle massive amounts of traffic. Behind the scenes, ELB monitors traffic and automatically scales to meet demand. This process, which includes a generous buffer of headroom, has become quicker and more responsive over the years and works well even for our customers who use ELB to support live broadcasts, “flash” sales, and holidays. However, in some situations such as instantaneous fail-over between regions, or extremely spiky workloads, we have worked with our customers to pre-provision ELBs in anticipation of a traffic surge.

New Network Load Balancer
Today we are introducing the new Network Load Balancer (NLB). It is designed to handle tens of millions of requests per second while maintaining high throughput at ultra low latency, with no effort on your part. The Network Load Balancer is API-compatible with the Application Load Balancer, including full programmatic control of Target Groups and Targets. Here are some of the most important features:

Static IP Addresses – Each Network Load Balancer provides a single IP address for each Availability Zone in its purview. If you have targets in us-west-2a and other targets in us-west-2c, NLB will create and manage two IP addresses (one per AZ); connections to that IP address will spread traffic across the instances in all the VPC subnets in the AZ. You can also specify an existing Elastic IP for each AZ for even greater control. With full control over your IP addresses, Network Load Balancer can be used in situations where IP addresses need to be hard-coded into DNS records, customer firewall rules, and so forth.

Zonality – The IP-per-AZ feature reduces latency with improved performance, improves availability through isolation and fault tolerance and makes the use of Network Load Balancers transparent to your client applications. Network Load Balancers also attempt to route a series of requests from a particular source to targets in a single AZ while still providing automatic failover should those targets become unavailable.

Source Address Preservation – With Network Load Balancer, the original source IP address and source ports for the incoming connections remain unmodified, so application software need not support X-Forwarded-For, proxy protocol, or other workarounds. This also means that normal firewall rules, including VPC Security Groups, can be used on targets.

Long-running Connections – NLB handles connections with built-in fault tolerance, and can handle connections that are open for months or years, making them a great fit for IoT, gaming, and messaging applications.

Failover – Powered by Route 53 health checks, NLB supports failover between IP addresses within and across regions.

Creating a Network Load Balancer
I can create a Network Load Balancer opening up the EC2 Console, selecting Load Balancers, and clicking on Create Load Balancer:

I choose Network Load Balancer and click on Create, then enter the details. I can choose an Elastic IP address for each subnet in the target VPC and I can tag the Network Load Balancer:

Then I click on Configure Routing and create a new target group. I enter a name, and then choose the protocol and port. I can also set up health checks that go to the traffic port or to the alternate of my choice:

Then I click on Register Targets and the EC2 instances that will receive traffic, and click on Add to registered:

I make sure that everything looks good and then click on Create:

The state of my new Load Balancer is provisioning, switching to active within a minute or so:

For testing purposes, I simply grab the DNS name of the Load Balancer from the console (in practice I would use Amazon Route 53 and a more friendly name):

Then I sent it a ton of traffic (I intended to let it run for just a second or two but got distracted and it created a huge number of processes, so this was a happy accident):

$ while true;
> do
>   wget http://nlb-1-6386cc6bf24701af.elb.us-west-2.amazonaws.com/phpinfo2.php &
> done

A more disciplined test would use a tool like Bees with Machine Guns, of course!

I took a quick break to let some traffic flow and then checked the CloudWatch metrics for my Load Balancer, finding that it was able to handle the sudden onslaught of traffic with ease:

I also looked at my EC2 instances to see how they were faring under the load (really well, it turns out):

It turns out that my colleagues did run a more disciplined test than I did. They set up a Network Load Balancer and backed it with an Auto Scaled fleet of EC2 instances. They set up a second fleet composed of hundreds of EC2 instances, each running Bees with Machine Guns and configured to generate traffic with highly variable request and response sizes. Beginning at 1.5 million requests per second, they quickly turned the dial all the way up, reaching over 3 million requests per second and 30 Gbps of aggregate bandwidth before maxing out their test resources.

Choosing a Load Balancer
As always, you should consider the needs of your application when you choose a load balancer. Here are some guidelines:

Network Load Balancer (NLB) – Ideal for load balancing of TCP traffic, NLB is capable of handling millions of requests per second while maintaining ultra-low latencies. NLB is optimized to handle sudden and volatile traffic patterns while using a single static IP address per Availability Zone.

Application Load Balancer (ALB) – Ideal for advanced load balancing of HTTP and HTTPS traffic, ALB provides advanced request routing that supports modern application architectures, including microservices and container-based applications.

Classic Load Balancer (CLB) – Ideal for applications that were built within the EC2-Classic network.

For a side-by-side feature comparison, see the Elastic Load Balancer Details table.

If you are currently using a Classic Load Balancer and would like to migrate to a Network Load Balancer, take a look at our new Load Balancer Copy Utility. This Python tool will help you to create a Network Load Balancer with the same configuration as an existing Classic Load Balancer. It can also register your existing EC2 instances with the new load balancer.

Pricing & Availability
Like the Application Load Balancer, pricing is based on Load Balancer Capacity Units, or LCUs. Billing is $0.006 per LCU, based on the highest value seen across the following dimensions:

  • Bandwidth – 1 GB per LCU.
  • New Connections – 800 per LCU.
  • Active Connections – 100,000 per LCU.

Most applications are bandwidth-bound and should see a cost reduction (for load balancing) of about 25% when compared to Application or Classic Load Balancers.

Network Load Balancers are available today in all AWS commercial regions except China (Beijing), supported by AWS CloudFormation, Auto Scaling, and Amazon ECS.

Jeff;