Google Cloud Results

• Improves performance and reliability, enabling Pantheon to serve larger customers
• Supports 99.95% uptime across 200,000+ websites
• Reduces cloud infrastructure costs by 40%
• Enables future analytics offerings based on machine learning and big data analytics

Nearly every business needs a web presence—but the vast majority of companies don’t want to be involved in the technical aspects of coding, deployment, hosting, security, and scaling websites. To stay focused on the business value and creative aspects of their websites and avoid managing infrastructure, thousands of companies turn to Pantheon, a website operations and hosting platform that powers over 200,000 websites.

Pantheon promises its customers speed, reliability, and scalability plus world-class collaboration and workflow tools. For five years, the company was able to deliver high service levels in all three areas running its platform on bare-metal virtual cloud servers. However, as its business grew, network links began to saturate under heavy load, risking instability. As Pantheon’s business evolved to focus on servicing some of the largest websites in the world, the company wanted to partner with a more innovative cloud services provider.

“We wanted a partner that could give us what we offer to our own customers: the flexibility to scale smoothly and consume services without building them from scratch,” says David Strauss, CTO, Pantheon. “It was time to move beyond custom containers on managed VMs and extend our cloud strategy to include next-generation technologies for container management and analytics.”

Pantheon evaluated several leading cloud providers and determined that Google Cloud Platform would be the best fit for its business and customers. Engineering had the final say, running a battery of functionality and performance tests at the storage, database, and web server layers.

“In every test our engineers did, Google Cloud Platform came in as better, faster, and more cost effective than the competition,” says Niall Hayes, COO, Pantheon. “We compared MariaDB to Google Cloud SQL and Cassandra to Google Bigtable, and container density improved from 250 to 400 containers per server.”

Migrating 200,000+ sites in 2 weeks

Pantheon wanted to make the transition transparent to its customers, so a fast and smooth migration to Google Cloud Platform was essential. With help from Google, Pantheon completed the migration quickly and moved 500TB of databases, code, and files with zero customer impact.

“We migrated over 200,000 websites to Google Cloud Platform in 2 weeks, including 50,000 that are heavily trafficked and actively developed, and nobody noticed,” says Josh Koenig, Co-founder and Head of Products at Pantheon. “The speed was incredible. There was no downtime, and we filed no additional support tickets with Google during the entire process.”

The platform for platforms

For its content management system runtime environment, Pantheon runs its own homegrown container management technology on Google Compute Engine. To automate scaling for other core services such as its routing layer and distributed file system, it uses Google Kubernetes Engine for cluster management and orchestration.

“Google is the clear leader in Kubernetes and container management, which aligns very well with our open source values and our vision for the future,” says Niall. “With Google Kubernetes Engine we get better resource efficiency, and automated operations and autoscaling take a lot of administration off our plate.”

In addition to smooth scaling, Pantheon and its customers benefit from improved performance and reliability thanks to the high-quality private network offered by Google.

Further, the Google partnership with Fastly enables direct connectivity to Google Cloud Platform to improve performance for edge caching. As a result of these improvements, Pantheon raised its availability service level agreement (SLA) from 99.9% to 99.95% and can now take on even larger customers.

“Google’s network topology, both locally and globally, performs better and more reliably than competing solutions, making Google Cloud Platform the best choice for us and for our customers,” says David. “Google beats any other cloud provider as the best platform-for-platforms.”

Adds Josh: “Google has unbelievable technology around persistence and replication between zones and regions, and that is not something we could find anywhere else. This allows us to offer advanced disaster recovery and failover services to our customers.”

Strengthening customer relationships

Pantheon uses Google BigQuery, a fully managed, cloud-based data warehouse, to integrate with Fastly and analyze website traffic on behalf of its customers. Previously, Pantheon was unable to ingest edge data quickly enough from Fastly, limiting its ability to identify issues and provide the best customer service. Today, Fastly streams logs in real time into Google BigQuery for analysis, giving Pantheon a wealth of insights.

“We use Google BigQuery to identify customers that have outgrown their infrastructure or need to right-size for business growth,” says David. “We can have proactive conversations and add a lot of value to the relationships. Soon, we plan to make Google BigQuery available to our customers so they can better understand their own traffic.”

Adds Niall: “Google Cloud Platform is more data-oriented than other cloud providers, making it a better match for our needs and our customers’ strategic initiatives.”

Integrated, granular security

Pantheon appreciates that Google Cloud services are built for public cloud, with granular security as a core design and development requirement. Employees simply use their G Suite credentials to gain access to Google Cloud Platform infrastructure and services.

“We’ve been a G Suite shop for years because of the paperless collaboration benefits,” says Josh. “G Suite connects our distributed company, and it was very natural to use those same logins for Google Cloud Platform.”

Staying competitive and productive

Moving to Google Cloud Platform opens up new possibilities for services Pantheon can offer to customers in the future, including machine learning and big data analytics, to give them a more complete view of how digital experiences are driving their businesses. Internally, engineers can move faster, do more effective capacity planning, and provide better service as Pantheon moves its products upmarket.

Pantheon expected to save 20% on cloud infrastructure costs by moving to Google Cloud Platform, but was able to double that savings with resource optimization and managed services.

“Since moving to Google Cloud Platform, our platform is more secure, reliable, and scalable than ever. We reduced our cloud infrastructure costs by 40%, and our customers’ sites run 45% faster than industry benchmarks,” says Niall. “Our engineers are Google fans for a reason—they’re happier, more efficient, and more productive on Google Cloud Platform.”

Enterprises want to embrace the pace of innovation and an operational business model of the modern-day cloud, but they don’t want to disrupt their existing IT landscape or upgrade all their legacy applications. This presents a conundrum, as most legacy applications were not designed to run in the cloud, and migrating them can be challenging, risky and cost-prohibitive. 

The complexity, cost and risk of cloud migration often stand in the way of a successful digital transformation. At Google Cloud, we work to meet you where you are and help you craft a migration strategy that lets you run what you want, where you want, how you want. This is why we’re excited to introduce Bare Metal Solution: to jumpstart the migration of applications that have been holding back your cloud adoption. 

Bare Metal Solution consists of all the infrastructure you need to run your specialized workload such as Oracle Database close to Google Cloud. This infrastructure is connected with a dedicated, low-latency and highly resilient interconnect, and connects to all native Google Cloud services. Bare Metal Solution uses OEM hardware that is certified to run multiple enterprise applications, most of which can be migrated to this infrastructure with little or no change, minimizing the risk of migration while simultaneously increasing its velocity.

Bare Metal Solution also comes with automation tools to help you onboard your environment quickly—provisioning your applications, relational databases, configuring popular operating systems and setting up services such as backups and monitoring. The management interface will be familiar to your existing IT teams or systems integrator, allowing you to leverage your investments in existing tools, processes and personnel.

In addition to providing state-of-the-art infrastructure and integrated low-latency access to Google Cloud, Bare Metal Solution also provides the following:

• Completely managed hardware infrastructure: End-to-end infrastructure management such as compute, storage and networking, as well as fully managed and monitored environments such as power, cooling and facilities.
• Google Cloud support and billing: A seamless support experience with support for infrastructure, including defined SLAs for initial response. 24X7 coverage for all Priority 1 and 2 issues. Unified billing across Google Cloud and Bare Metal Solution. 
• Service Level Agreements: Defined enterprise-grade SLA for hardware uptime and interconnect availability.

As an integral part of Google Cloud, Bare Metal Solution lets you off-load provisioning, managing and monitoring of your infrastructure to Google, so you can focus on your own data center modernization. 

The nitty gritty 

Many of the workloads that run on Bare Metal Solution have demanding CPU and I/O requirements. Bare Metal Solution provides a cost-effective and scalable architecture based on state-of-the-art x86 servers and high-performance, resilient storage. The servers are offered in several fixed configurations that support most mainstream enterprise operating systems:

• Dual-socket x86 systems
  • 16 core with 384 GB DRAM
  • 24 core with 768 GB DRAM
  • 56 core with 1536 GB DRAM
• Quad-socket x86 systems
  • 56 core with 1536 GB DRAM 
  • 112 core with 3072 GB DRAM

Bare Metal Solution servers can be used for standalone applications, or configured with application native or database native clustering technologies for high availability. Storage is offered in 1TB volume increments of either hybrid or all-flash disk. For applications that require custom compute shapes or special-purpose hardware, we also offer bespoke hardware configurations. 

Bare Metal Solution uses OEM hardware that is certified for many ISV software applications as well as custom built applications, including those built on Oracle Database. These hardware configurations are offered as a subscription, billed monthly with a preferred term length of 36 months. There are no data ingress and egress charges between Bare Metal Solution and Google Cloud in the same region; customers are responsible for provisioning adequate bandwidth for their business needs.

Legacy apps are no barrier to cloud

Google Cloud is the preferred destination for organizations building cloud-native applications as well as migrating existing on-premises applications, bringing reliable infrastructure, leading data analytics capabilities, and a culture of innovation to your IT environment. Whether you’re looking to build new applications in the cloud or to overhaul your infrastructure, we’re here to help you reach your digital transformation goals. And now, with Bare Metal Solution, we’re excited to extend the power of Google Cloud to specialized, legacy workloads. To learn more please visit our website.

In fast-moving organizations, it’s not uncommon for cloud resources, including entire projects, to occasionally be forgotten about. Not only such unattended resources can be difficult to identify, but they also tend to create a lot of headaches for product teams down the road, including unnecessary waste and security risks. 

To help you prune your idle cloud resources, we’re excited to introduce Unattended Project Recommender. It’s a new feature of Active Assist that provides you with a one-stop shop for discovering, reclaiming, and shutting down unattended projects. With actionable and automatic recommendations, you no longer have to worry about wasting money or mitigating security risks presented by your idle resources. Unattended Project Recommender uses machine learning to identify, with a high degree of confidence, projects that are likely abandoned based on API and networking activity, billing, usage of cloud services, and other signals. This feature is available via the Recommender API today, making it easy for you to integrate with your company’s existing workflow management and communication tools, or export results to a BigQuery table for custom analysis.

Thousands of projects can be unattended in large organizations, presenting major security risks

Your cloud projects can go abandoned or unattended for a number of reasons — ranging from a test environment that’s no longer needed, to project cancellation, to project owner switching jobs, and more. Not only can such projects contribute to your cloud bill (waste) but they may contain security issues such as open firewalls or privileged service account keys that attackers can exploit to get a hold of your cloud resources for cryptocurrency mining or, worse, compromise your company’s sensitive data. These security risks tend to grow over time because the latest best practices and patches are usually not applied to unattended projects. 

We experience this issue here at Google, too… In fact, it has been on Google’s internal security team’s radar for some time now, so we joined forces and looked into this problem together, starting with our very own “google.com” organization cloud projects. We quickly found some projects that were unattended, but remediating this issue was easier said than done due to challenges in several areas:

• Detection: With lots of signals available to you via sources like Cloud Monitoring, what are the right ones you should look at (e.g. API, networking, user activity)? How can you tell the difference between an unattended project and a project that has a low level of activity by design (e.g. a “shell” project that holds an auth token)?
• Remediation: Once you have identified a project that seems abandoned, how do you go about ensuring that it’s indeed an unattended project? How do you reduce the risk of deleting something that might be essential to a production workload, causing irreversible data loss? How do you solve this at the scale of your entire organization, beyond a one-time cleanup? 

Over the course of 2021 we built and tested a Google-internal prototype first, cleaning up many of our internal unattended projects, and then worked with a number of Google Cloud customers to build and tune this feature based on real-life data (thank you to all of our early adopters for working with us and your generous feedback that helped us shape this feature!) It was not uncommon for us to come across organizations with thousands of unattended projects, and we’re very excited to bring Unattended Project Recommender to all customers, in public preview.

Discovering and acting on unattended project recommendations

Unattended Project Recommender analyzes usage activity across all projects under your organization, including the following data:

• API activity (e.g. service accounts with authentication activity, API calls consumed)
• Networking activity (ingress and egress)
• Billing activity (e.g. services with billable usage)
• User activity (e.g. active project owners)
• Cloud services usage (e.g. number of active VMs, BigQuery jobs, storage requests)

Based on these signals, it can generate recommendations to clean up projects that have low usage activity (where “low usage” is defined using a machine learning model that ranks projects in your organization by level of usage), or recommendations to reclaim projects that have high usage activity but no active project owners. Here’s what an example post-processed summary list of recommendations can look like for the “foobar” organization that has 3 projects:

  Project ID: demo-project-307815
Recommendation: CLEANUP_PROJECT

Project ID: new-project
Recommendation: N/A

Project ID: bobs-playground-project
Recommendation: RECLAIM_PROJECT

In addition to the recommendations, you can also examine the underlying project activity insights that the recommendations are based upon. The insights provide additional information that can be useful for integration with your organization’s existing workflows and automation (e.g. send an auto-generated email or chat message to project owners based on the list provided by the owners field). Here’s an example insight payload:

  content:
  activeAppengineInstanceDailyCount: 0
  activeCloudsqlInstanceDailyCount: 0
  activeGceInstanceDailyCount: 3
  activeServiceAccountDailyCount: 1
  apiClientDailyCount: 18922           // Daily average API calls produced
  bigqueryInflightJobDailyCount: 0
  bigqueryInflightQueryDailyCount: 0
  bigqueryStorageDailyBytes: 0
  bigqueryTableDailyCount: 0
  consumedApiDailyCount: 0             // Daily average API calls consumed
  datastoreApiDailyCount: 0
  gcsObjectDailyCount: 11
  gcsRequestDailyCount: 0
  gcsStorageDailyBytes: 2663548
  hasActiveOauthTokens: false          // OAuth tokens used in the last 180 days
  hasBillingAccount: true
  numActiveUserOwners: 1
  owners:                              // List of project owners
  - activeOnProject: false
    member: user:user1@example.com
  - activeOnProject: true
    member: user:user2@example.com
  serviceWithBillableUsage:
  – Cloud Storage
  - Compute Engine
  vpcEgressDailyBytes: 264456938       // Daily average VPC egress bytes
  vpcIngressDailyBytes: 392435047      // Daily average VPC ingress bytes
  usagePercentile: 20                  // Level of usage relative to other projects

GCP projects are used in many different ways and for many different purposes. In case you get a recommendation to delete a project that’s being used in a way that’s out of the scope for this feature, you can dismiss the recommendation and it will stop showing up for the given project. 

Restoring deleted projects

When you choose to shut down a project using the projects.delete() method, it gets marked for deletion. After a project is marked for deletion, it becomes unusable, all resources within that project are shut down, and a 30-day wait period for the project and all of its data to get fully deleted begins.

In case a useful project is accidentally shut down, you have the option to restore the project within that 30-day wait period. Since restoring allows you to recover most but not necessarily all of your project data and resources, we recommend carefully examining the utilization insights associated with a project and considering any additional utilization signals that may not be captured by the Unattended Project Recommender before taking the cleanup action.

Early customer success stories

A number of enterprise customers are already using Unattended Project Recommender to keep their organizations clean of unattended projects and resources.

Decathlon, a French sporting goods retailer, is excited for the insight Unattended Project Recommender will bring to their environment, and are already deploying it as a part of their latest cloud security initiatives.

“After a thorough test of this feature and the validation of our CISO, we ended up deleting our first 775 projects, and no one complained! A great help to improve our security. The next step for us will be to operationalize it at scale, and implement a company wide policy for unattended resource management.” —Adeline Villette, Cloud Security Officer

For Veolia, one of the world’s largest water, waste and energy management companies, not only does this feature reduce security risks and waste, but also helps drive cultural shift and alignment with its ecological transformation strategy.

“This feature allows us to reduce our costs and security debt on assets that are no longer in use, and is also fully in line with Veolia’s philosophy of limiting its carbon footprint. After having tested Unattended Project Recommender on more than 3,000 projects throughout our organization, we are looking to bring it as proactive alerts to our project owners at scale.”—Thomas Meriadec, Product Manager

Box, a secure cloud content management provider, views it as a foundation for building a repeatable process to remediate unused resources.

“Unattended Project Recommender is a great fit for us. It gives us a unified view of project usage across our entire organization and enables us to address security risks of legacy projects in a systematic and organized manner, ensuring an even safer environment.” —Matt Bowes, Staff Security Engineer

Getting started with the Unattended Project Recommender

To help you get started, we’ve prepared a Cloud Shell tutorial (source code) that you can use to find unattended project recommendations within your own Projects/Folders/Organization. Click this button to clone the tutorial from GitHub and run in your Cloud Shell environment:

As you can see, listing recommendations for your projects only takes a few clicks with the tutorial (special thanks to Lanre Ogunmola, Security & Compliance Specialist, for making this look so easy)! For additional detail on using the gcloud CLI or API to discover unattended project recommendations, please refer to the documentation page.

You can also automatically export all recommendations from your Organization to BigQuery and then investigate the recommendations with DataStudio or Looker, or use Connected Sheets that let you use Google Workspace Sheets to interact with the data stored in BigQuery without having to write SQL queries.

As with any other Recommender, you can choose to opt out of data processing at any time by disabling the appropriate data groups in the Transparency & control tab under Privacy & Security settings.

We hope that you can leverage Unattended Project Recommender to improve your cloud security posture and reduce cost, and can’t wait to hear your feedback and thoughts about this feature! Please feel free to reach us at active-assist-feedback@google.com and we also invite you to sign up for our Active Assist Trusted Tester Group if you would like to get early access to the newest features as they are developed.

We recently launched Vertex AI to help you move machine learning (ML) from experimentation into production faster and manage your models with confidence—speeding up your ability to improve outcomes at your organization.

But we know many of you are just getting started with ML and there’s a lot to learn! In tandem with building the Vertex AI platform, our teams are dropping as much best practices content as we can to help you come up to speed. Plus, we have a dedicated event on June 10th, Applied ML Summit, with sessions on how to apply ML technology in your projects, as well as grow your skills in this field. 

In the meantime, we couldn’t resist a quick lesson on hyperparameter tuning, because (a) it’s incredibly cool (b) you will impress your coworkers (c) Google Cloud has some unique battle tested tech in this area and (d) you will save time by getting better ML models into production faster. Vertex Vizier, on average, finds optimal parameters for complex functions in over 80% fewer trials than traditional methods. 

So it’s incredibly cool, but what is it?

While machine learning models automatically learn from data, they still require user-defined knobs which guide the learning process. These knobs, commonly known as hyperparameters, control, for example, the tradeoff between training accuracy and generalizability.  Examples of hyperparameters are the optimizer being used, its learning rate, regularization parameters, the number of hidden layers in a DNN, and their sizes.

Setting hyperparameters to their optimal values for a given dataset can make a huge difference in model quality. Typically, optimal hyperparameter values are found via grid searching a small number of combinations, or tedious manual experimentation. Hyperparameter tuning automates this work for you by searching for the best configuration of hyperparameters for optimal model performance. 

Vertex Vizier enables automated hyperparameter tuning in several ways:

1. “Traditional” hyperparameter tuning: by this we mean finding the optimal value of hyperparameters by measuring a single objective metric which is the output of an ML model.  For example, Vizier selects the number of hidden layers and their sizes, an optimizer and its learning rate, with the goal of maximizing model accuracy.
2. When hyperparameters are evaluated, models are trained and evaluated on splits of the data set. If evaluation metrics are streamed to Vizier (e.g. as a function of epoch) as the model is trained, Vizier’s early stopping algorithms can predict the final objective value, and recommend which unpromising trials should be early stopped. This conserves compute resources and speeds up convergence.
3. Oftentimes, models are tuned sequentially on different data sets. Vizier’s built in transfer learning learns priors from previous hyperparameter tuning studies, and leverages them to converge faster on subsequent hyperparameter tuning studies.
4. AutoML is a variant of #1, where Vertex Vizier performs both model selection, and also tunes architectures/non-architecture modifying hyperparameters. AutoML usually requires more code on top of Vertex Vizier (to ingest data etc), but Vizier is in most cases the “engine” behind the process. AutoML is implemented by defining a tree like (DAG) search space, rather than a “flat” search space (like in #1). Note that you can use DAG search spaces for any other purpose where searching over a hierarchical space makes sense.
5. There are times when you may wish to optimize more than one metric. For example, we would like to optimize model accuracy, while minimizing model latency. Vizier can find the Pareto frontier, which presents tradeoffs for multiple metrics, allowing users to choose the appropriate tradeoff. Simple example: I want to make a more accurate model, but would like to minimize serving latency. I do not know ahead of time what’s the tradeoff between the two metrics. Vizier can be used to explore and plot a tradeoff curve, so users can select on the most appropriate one. For example, “a latency decrease of 200ms will only decrease accuracy by 0.5%”

Google Vizier is all yours with Vertex AI

Google published the Vizier research paper in 2017, sharing our work and use cases for black-box optimization—i.e. The process of finding the best settings for a bunch of parameters or knobs when you can’t peer inside a system to see how well the knobs are working. The paper discusses our requirements, infrastructure design, underlying algorithms, and advanced features such as transfer learning that the service provides. Vizier has been essential to our progress with machine learning at Google, which is why we are so excited to make it available to you on Vertex AI.

Vizier has already tuned millions of ML models at Google, and its algorithms are continuously improved for faster convergence and handling of real-life edge cases. Vertex Vizier’s models are very well calibrated and are self-tuning (they adapt to user data), and offer unique power features, such as hierarchical search spaces and multi-objective optimization. We believe Vertex Vizier’s set of features is a unique capability to Google Cloud, and look forward to optimizing the quality of your models by automatically tuning hyperparameters for you.

To learn more about Vertex Vizier, check out these docs and if you are interested in what’s coming in machine learning over the next five years, tune in to our Applied ML Summit on June 10th, or watch the sessions on demand in your own time.