Soldiers can now use their CAC cards to log into the service’s newly established Google Workspace account. With it, soldiers who don’t need the full spectrum of information-sharing provided by Army365 can nevertheless find comrades, schedule meetings and coordinate other activities.

“All new entrants to the Army will be provisioned automatically with a Google Workspace account with a new email address … when they receive their CAC card,” Raj Iyer, the service’s chief information officer, said in a social-media post.

“Soldiers will retain their Google account till they fully transition to their first unit after completion of basic training and AIT [advanced infantry training], at which point their commanders will determine if they need an Army365 account.”

The directive applies to all active-duty, reserve, and National Guard soldiers.

Communication services providers (CSPs) are experiencing a period of disruption. Overall revenue growth is decelerating and is projected to remain below 1 percent per year, following a trend that started even before the pandemic.¹ At the same time, driven by the pandemic, data consumption in 2020 increased by 30 percent relative to 2019, with some operators even reporting increases of 60 percent.² 

The combination of pressure on revenues with rising data traffic costs is forcing operators to innovate in three fundamental ways. First, operators are looking to establish new sources of revenue. Second, increased network utilization must be met with a reduction in network cost. And third, there is an opportunity to gain new customers by improving the customer experience.

Fortunately, 5G offers a path forward across each of these three areas. Concepts such as network slicing and private networks allow CSPs to offer differentiated network services to public sector and enterprise customers. The disaggregation of hardware and software allows new vendors with unique strengths to enter the market and to enable CSPs to build, deploy, and operate networks in fundamentally new ways. And the ability to place workloads at the edge permits CSPs to offer compelling experiences to consumers and businesses alike. In this blog, we will discuss how CSPs can create a solid foundation for their cloud networks. 

Understanding telecommunications networks

First, it is useful to consider the way telecommunications networks were traditionally built. Initially, networks were built using physical network functions (PNFs) — appliances that used a tight combination of hardware and software to perform a specific function. PNFs offered the benefit of being purpose-built for a specific application, but they were inflexible and difficult to upgrade. As an example, deploying new features frequently required replacing the entire PNF, i.e., deploying a new hardware appliance.

The first step in improving deployment agility came with the concept of virtualized network functions (VNFs), software workloads designed to operate on commercial off-the-shelf (COTS) hardware. Rather than utilizing an integrated hardware and software appliance, VNFs disaggregated the hardware from the software. As such, it became possible to procure the hardware from one vendor and the software from another. It also became possible to separate the hardware and software upgrade cycles. 

However, while VNFs offered advantages over PNFs, VNFs were still an intermediate step. First, they typically needed to be run within a virtual machine (VM), and as such required a hypervisor to interface between the host operating system (OS) and the guest OS inside the VM. The hypervisor consumed CPU cycles and added inefficiency. Second, the VNF itself was frequently designed as a monolithic function. This meant that while it was possible to upgrade the VNF separately from the hardware, such an upgrade, even for a feature that affected only a portion of the VNF, required deployment of the entire large VNF. This created risk and operational complexity, which in turn meant that upgrades were delayed just as they were with PNFs.

Creating the foundation for cloud networks

The trick to establishing your cloud based network resides in the challenge of moving from VNFs to containerized network functions (CNFs) — network functions organized as containers as a collection of small programs, each of which can be independently operated. 

The concept of containers is not new. In fact, Google has been using containerized workloads for over 15 years. Kubernetes, which Google developed and open-sourced, is the world’s most popular container orchestration system, and is based on Borg, Google’s internal container management system.³ There are lots of benefits to using containers, but fundamentally, it frees developers from worrying about resource scheduling, interprocess communication, security, self-healing, load balancing, and many other tedious (but important!) tasks. 

Consider just a couple examples of benefits that containerization brings to network functions. First, when upgrading the network function to implement new features, you no longer need to re-deploy the entire network function. Instead, you only need to re-deploy the containers that are affected by the upgrade. This improves developer velocity and reduces the risk of the upgrade because, rather than infrequent upgrades that each introduce substantial changes, you can now have frequent upgrades that each deploy small changes. Small changes are less risky because they are easier to understand and to roll back in case of anomaly. Incidentally, this also improves your security posture because it reduces the time between when a security vulnerability is discovered and when a patch is deployed.

Speaking of security, another example of the benefits that containerization brings to network functions is an automatic zero-trust security posture. In Kubernetes, the communication among microservices can be handled by a service mesh, which manages mundane aspects of inter-services communication such as retries in case of failure and providing observability into communication. It can also manage other essential aspects such as security. For example, Anthos Service Mesh, which is a fully-managed implementation of the open-source Istio service mesh (also co-developed by Google), includes the ability to authenticate and encrypt all communications using mutual TLS (mTLS) and to deploy fine-grained access control for each individual microservice.

Automation and orchestration for cloud networks

CNFs bring tremendous benefits, but they also bring challenges. In place of a relatively small number of network appliances, we now have a large number of containers, each of which requires configuration, management, and maintenance. In the past, many of these processes were accomplished using manual techniques, but this is impossible to accomplish economically and reliably at the scale required by CNFs.

Fortunately, there are cloud-native approaches to solving these challenges. First, consider the problem of autonomously deploying and maintaining CNFs. The ideal way is to use the concept of Configuration as Data. Unlike imperative techniques such as Infrastructure as Code, which provide a detailed description of a sequence of steps that need to be executed to achieve an objective, Configuration as Data is a declarative method whereby the user specifies the desired end state (i.e., the actual desired configuration) and relies on automated controllers to continuously drive the infrastructure to achieve that state. Kubernetes includes such automated controllers, and the great news is that this method can be used not just for infrastructure but also for the applications residing on top of it, including CNFs. This cloud-native technique frees you from the toil and associated risk of writing detailed configuration procedures, so you can focus on the business logic of your applications.

As another example, consider the problem of understanding your network performance, including anomaly detection, root cause analysis, and resolution. The cloud-native approach starts with creating a data platform where both infrastructure and CNF monitoring data can be ingested, regularized, processed, and stored. You can then correlate data sets against each other to detect anomalies, and with AI/ML techniques, you can even anticipate anomalies before they happen. AI/ML is likewise indispensable in gaining an understanding of why the anomaly is happening, i.e. performing root cause analysis, and automated closed-loop controllers can be developed to correct the problem, ideally before it even happens.

Architecting for the edge

The transition from VNFs to CNFs is a critical piece in addressing the challenge that CSPs face today, but it alone is not enough. CNFs need infrastructure to run on, and not all infrastructure is created equal. 

Consider a typical 5G network. There are some functions, such as those associated with an access network, that need to be deployed at the edge. These functions require low latency, high throughput, or even a combination of the two. In 5G networks, examples of such functions include the radio unit (RU), distributed unit (DU), centralized unit (CU), and the user plane function (UPF). The first three are components of the radio access network (RAN), while the last is a component of the 5G core. At the same time, there are some other control plane functions such as the session management function (SMF) or the authentication and mobility management function (AMF) that do not have such tight latency and high throughput requirements and can thus be placed in a more centralized data center. Furthermore, consider an AI/ML use case where a particular model (perhaps for radio traffic steering) needs to run at the network edge because of its latency requirements. While the model itself needs to run at the edge, model training (i.e., generating the model coefficients) is frequently a compute-intensive exercise that is latency-insensitive and is thus more optimal to run in a public cloud region.

All of these use cases have one thing in common: they call for a hybrid deployment environment. Some applications must be deployed at the edge as close to the user as possible. Others can be deployed in a more centralized environment. Still others can be deployed in a public cloud region to take advantage of the large amount of compute and economies of scale available therein. Wouldn’t it be convenient — if not transformational — if you could use a single environment for deploying at the edge, in a private datacenter, and in public cloud, with a consistent set of security, lifecycle management, policy, and orchestration resources across all such locations? This is indeed what Google Distributed Cloud, enabled by Anthos, brings to the table.

With Google Distributed Cloud, you can architect a 5G network deployment such as the one shown below.

Business benefits of cloud networks

Beyond the technical benefits, consider the business benefits of such an architecture. First, by following the best practices of hardware and software disaggregation, it permits the CSP to procure the infrastructure and the network functions from different vendors, spurring competition among vendors. Second, each workload is placed in precisely the right location, enabling efficient utilization of hardware resources and offering compelling low-latency, high-throughput services to users. Third, because the architecture utilizes a common hybrid platform (Anthos), it makes it easy to move workloads across infrastructure locations. Fourth, the separation of workloads into microservices accelerates time-to-market when developing new features or applications, such as those enabling enterprise use cases. And finally, the container management platform supports the simultaneous deployment of both network functions and edge applications on the same infrastructure, allowing the operator to deploy new experiences such as AR/VR directly on bare metal as close to the user as possible.

The next generation cloud network is now

There is a lot more we could say, but perhaps the most important takeaway is that this architecture is not a future dream. It exists today, and Google is working with leading CSPs and network vendor partners to deploy it, helping them realize the promise of 5G to deliver new revenues, reduce operating costs, and enable new customer experiences.

To learn more, watch the video series on the cloudification of CSP networks.

Discover what’s happening at the edge: How CSPs Can Innovate at the Edge.

^{1.Statista, Forecast growth worldwide telecom services spending from 2019 to 2024
2 PricewaterhouseCoopers, Global entertainment and media outlook 2021-2025
3. Borg: The Predecessor to Kubernetes}

Records are made to be broken. In 2019, we calculated 31.4 trillion digits of π — a world record at the time. Then, in 2021, scientists at the University of Applied Sciences of the Grisons calculated another 31.4 trillion digits of the constant, bringing the total up to 62.8 trillion decimal places. Today we’re announcing yet another record: 100 trillion digits of π.

This is the second time we’ve used Google Cloud to calculate a record number1 of digits for the mathematical constant, tripling the number of digits in just three years.

This achievement is a testament to how much faster Google Cloud infrastructure gets, year in, year out. The underlying technology that made this possible is Compute Engine, Google Cloud’s secure and customizable compute service, and its several recent additions and improvements: the Compute Engine N2 machine family, 100 Gbps egress bandwidth, Google Virtual NIC, and balanced Persistent Disks. It’s a long list, but we’ll explain each feature one by one.

Before we dive into the tech, here’s an overview of the job we ran to calculate our 100 trillion digits of π.

• Program: y-cruncher v0.7.8, by Alexander J. Yee
• Algorithm: Chudnovsky algorithm
• Compute node: n2-highmem-128 with 128 vCPUs and 864 GB RAM
• Start time: Thu Oct 14 04:45:44 2021 UTC
• End time: Mon Mar 21 04:16:52 2022 UTC
• Total elapsed time: 157 days, 23 hours, 31 minutes and 7.651 seconds
• Total storage size: 663 TB available, 515 TB used
• Total I/O: 43.5 PB read, 38.5 PB written, 82 PB total

Architecture overview

Calculating π is compute-, storage-, and network-intensive. Here’s how we configured our Compute Engine environment for the challenge.

For storage, we estimated the size of the temporary storage required for the calculation to be around 554 TB. The maximum persistent disk capacity that you can attach to a single virtual machine is 257 TB, which is often enough for traditional single node applications, but not in this case. We designed a cluster of one computational node and 32 storage nodes, for a total of 64 iSCSI block storage targets.

The main compute node is a n2-highmem-128 machine running Debian Linux 11, with 128 vCPUs and 864 GB of memory, and 100 Gbps egress bandwidth support. The higher bandwidth support is a critical requirement for the system as we adopted a network-based shared storage architecture.

Each storage server is a n2-highcpu-16 machine configured with two 10,359 GB zonal balanced persistent disks. The N2 machine series provides balanced price/performance, and when configured with 16 vCPUs it provides a network bandwidth of 32 Gbps, with an option to use the latest Intel Ice Lake CPU platform, which makes it a good choice for high-performance storage servers.

Automating the solution

We used Terraform to set up and manage the cluster. We also wrote a couple of shell scripts to automate critical tasks such as deleting old snapshots, and restarting from snapshots (we didn’t need to use this though). The Terraform scripts created OS guest policies to help ensure that the required software packages were automatically installed. Part of the guest OS setup process was handled by startup scripts. In this way, we were able to recreate the entire cluster with just a few commands.

We knew the calculation would run for several months and even a small performance difference could change the runtime by days or possibly weeks. There are also a number of combinations of parameters in the operating system, infrastructure, and application itself. Terraform helped us test dozens of different infrastructure options in a short time. We also developed a small program that runs y-cruncher with different parameters and automated a significant portion of the measurement. Overall, the final design for this calculation was about twice as fast as our first design. In other words, the calculation could’ve taken 300 days instead of 157 days!

The scripts we used are available on GitHub if you want to look at the actual code that we used to calculate the 100 trillion digits.

Choosing the right machine type for the job

Compute Engine offers machine types that support compute- and I/O-intensive workloads. The amount of available memory and network bandwidth were the two most important factors, so we selected n2-highmem-128 (Intel Xeon, 128 vCPUs and 864 GB RAM). It satisfied our requirements: high-performance CPU, large memory, and 100 Gbps egress bandwidth. This VM shape is part of the most popular general purpose VM family in Google Cloud.

100 Gbps networking

The n2-highmem-128 machine type’s support for up to 100 Gbps of egress throughput was also critical. Back in 2019 when we did our 31.4-trillion digit calculation, egress throughput was only 16 Gbps, meaning that bandwidth has increased by 600% in just three years. This increase was a big factor that made this 100-trillion experiment possible, allowing us to move 82.0 PB of data for the calculation, up from 19.1 PB in 2019.

We also changed the network driver from virtio to the new Google Virtual NIC (gVNIC). gVNIC is a new device driver and tightly integrates with Google’s Andromeda virtual network stack to help achieve higher throughput and lower latency. It is also a requirement for 100 Gbps egress bandwidth.

Storage design

Our choice of storage was crucial to the success of this cluster – in terms of capacity, performance, reliability, cost and more. Because the dataset doesn’t fit into main memory, the speed of the storage system was the bottleneck of the calculation. We needed a robust, durable storage system that could handle petabytes of data without any loss or corruption, while fully utilizing the 100 Gbps bandwidth.

Persistent Disk (PD) is a durable high-performance storage option for Compute Engine virtual machines. For this job we decided to use balanced PD, a new type of persistent disk that offers up to 1,200 MB/s read and write throughput and 15-80k IOPS, for about 60% of the cost of SSD PDs. This storage profile is a sweet spot for y-cruncher, which needs high throughput and medium IOPS.

Using Terraform, we tested different combinations of storage node counts, iSCSI targets per node, machine types, and disk size. From those tests, we determined that 32 nodes and 64 disks would likely achieve the best performance for this particular workload.

We scheduled backups automatically every two days using a shell script that checks the time since the last snapshots, runs the fstrim command to discard all unused blocks, and runs the gcloud compute disks snapshot command to create PD snapshots. The gcloud command returns and y-cruncher resumes calculations after a few seconds while the Compute Engine infrastructure copies the data blocks asynchronously in the background, minimizing downtime for the backups.

To store the final results, we attached two 50 TB disks directly to the compute node. Those disks weren’t used until the very last moment, so we didn’t allocate the full capacity until y-cruncher reached the final steps of the calculation, saving four months worth of storage costs for 100 TB.

Results

All this fine tuning and benchmarking got us to the one-hundred trillionth digit of π — 0. We verified the final numbers with another algorithm (Bailey–Borwein–Plouffe formula) when the calculation was completed. This verification was the scariest moment of the entire process because there is no sure way of knowing whether or not the calculation was successful until it finished, five months after it began. Happily, the Bailey-Borwein-Plouffe formula found that our results were valid. Woo-hoo! Here are the last 100 digits of the result:

4658718895 1242883556 4671544483 9873493812 1206904813
2656719174 5255431487 2142102057 7077336434 3095295560

You can also access the entire sequence of numbers on our demo site.

So what?

You may not need to calculate trillions of decimals of π, but this massive calculation demonstrates how Google Cloud’s flexible infrastructure lets teams around the world push the boundaries of scientific experimentation. It’s also an example of the reliability of our products – the program ran for more than five months without node failures, and handled every bit in the 82 PB of disk I/O correctly. The improvements to our infrastructure and products over the last three years made this calculation possible.

Running this calculation was great fun, and we hope that this blog post has given you some ideas about how to use Google Cloud’s scalable compute, networking, and storage infrastructure for your own high performance computing workloads. To get started, we’ve created a codelab where you can create and calculate pi on a Compute Engine virtual machine with step-by-step instructions. And for more on the history of calculating pi, check out this post on The Keyword. Here’s to breaking the next record!

1. We are actively working with Guinness World Records to secure their official validation of this feat as a “World Record”, but we couldn’t wait to share it with the world. This record has been reviewed and validated by Alexander J. Yee, the author of y-cruncher.

Casper Labs announced a collaboration with Google Cloud that will allow developers to launch public and/or private Casper nodes directly from Google Cloud. This enables a much more seamless and highly secure process for the millions of developers who want to build in blockchain environments without having to learn new, highly specialized programming languages. Additionally, Google Cloud will provide its scalable and reliable infrastructure to developers building on the Casper Protocol. 

Blockchain technology is maturing 

As blockchain technology matures, a growing number of businesses are embracing it as a key way to drive new efficiencies and realize cost savings. 

According to a recent Casper Labs study, 87% of executives polled in the United States, United Kingdom and China reported plans to invest in a blockchain solution in 2023. This is due in no small part due to recent innovations that help organizations overcome the so-called Blockchain Adoption Trilemma, which previously held that it was impossible for any blockchain to be simultaneously a) decentralized, b) scalable, and c) secure.

Thanks to the rise of proof-of-stake blockchains like Casper, new models have emerged that enable a more scalable and secure architecture that no longer forces a compromise on decentralization. 

Another trend facilitating these growing adoption rates is the rise of WebAssembly (WASM) as a baseline technology for newer blockchains, including Casper. WASM (created by W3C) makes application development in blockchain environments far more accessible and interoperable to the millions of developers worldwide who specialize in languages like Java, Javascript, C++ and Rust. Previously, any blockchain-based build required a high degree of specialized developer knowledge, which made it a much more challenging option for most organizations. 

Meet Casper

Casper is a permissionless, decentralized public blockchain based on WASM that was built explicitly to foster enterprise adoption of blockchain technology. Beyond its more accessible model, Casper is the first and only blockchain to offer native upgradable smart contracts. This means that organizations can have the option to securely and consistently update software code even after it is running on Casper. This gives organizations the control and flexibility to use industry best practices, such as continuous deployment and continuous integration, which are already in use in their IT departments. Casper is also highly configurable and allows organizations to support public, private, and/or hybrid deployments. 

Casper is also noteworthy for the presence of Casper Labs, a software development and professional services firm that supports organizations building on the Casper network. Unlike most blockchains that follow a more traditional open-source project, Casper Labs provides around-the-clock support and bespoke software development for enterprise organizations. Recently, Casper Labs helped patent management company IPwe execute the largest-ever blockchain deployment, featuring more than 25 million patents being added as custom NFTs to the Casper Blockchain. 

How to get started with Casper on Google Cloud

Developers who want to start building on Casper can find a comprehensive series of tutorials here.

The Casper Association also recently announced a $25 million grant program to support projects and developers building on Casper. Interested participants can apply here.