Google Cloud Platform Technology Nuggets - October 1-15, 2024 Edition
Google Cloud Platform Technology Nuggets - October 1-15, 2024 Edition
Welcome to the October 1–15, 2024 edition of Google Cloud Platform Technology Nuggets.
The nuggets are also available in the form of a Podcast. Subscribe to it today.
Containers and Kubernetes
GKE’s IP allocation strategy might just stump you and result in a “IP_SPACE_EXHAUSTED” error. Node capacity in GKE in influenced by Cluster primary subnet, Pod IPv4 range and maximum pods per node. Check out this blog post to understand better and ensure that you don’t face this issue or better still, understand what could have caused it. Below is the output from the Network Analyzer tool, that can detect IP exhaustion issues.
The power of NVIDIA AI and Google Kubernetes Engine (GKE) is a powerful combination to serve AI models efficiently. Google Cloud announced the availability of NVIDIA NIM, part of the NVIDIA AI Enterprise software platform, on GKE and discoverable via Google Cloud Marketplace, letting you deploy NIM microservices directly from the Google Cloud console. Check out the blog post on the announcement and step by step of deploying this from the Marketplace. NIM is a set of easy-to-use microservices for accelerating the deployment of foundation models on any cloud platform.
Identity and Security
Google Cloud networking and security are key differentiators of Google Cloud, that helps customers run their workloads securely. It is designed to help prevent DDoS attacks. To understand how it prevents these attacks and to get started (either via the Cloud console or Terraform script), check out the blog post.
Continuing the theme of security, one of the industries that has strict security and regulatory compliance needs is the Telecom industry. Check out this post, that highlights how Telecom providers that wish to migrate to Google Cloud, will be provided with tools and compliance needs that will help them operate at scale, securely and within the regulatory requirements.
Cloud Key Management System (KMS) provides a managed service to create, store, and perform cryptographic operations such as code signing with keys via Cloud hardware security modules (Cloud HSM). The latest update includes the ability to sign Microsoft Artifacts with SignTool, while protecting your keys with Cloud HSM.
There have been several updates to Confidential Computing, which keeps your data safe while it is being processed and used, via a hardware-based Trusted Execution Environment (TEE). Check out the blog post that highlights additional machine series (C3D, C3 with Intel DTX and more) on which Confidential VMs are now available.
Machine Learning
If you have been in the Generative AI space in the last year and have built applications, chances are good that you have taken a shot at Retrieval Augmented Generation (RAG) to ground the LLM to your corpus of data. However, its not an easy task and you would have had your fair share to provide more accurate semantic search. There are various techniques to improve results and this new blog post focuses on a new task type embedding in Vertex AI, which promises to significantly improve the accuracy and effectiveness of your RAG system. Check out the detailed post.
If you are still in your early attempts to build out a RAG solution using a set of technologies, then this post can be a great way for your to step by step build out a RAG solution using Llama-index, Streamlit, RAGAS, and Google Cloud’s Gemini models.
Looking to fine-tune Gemma to cater to your specific data and instructions? A blog post titled “Fine-tuning Gemma, the journey from beginning to end” mentioned the following statement “To fine-tune Gemma for a chat-style experience with pytorch, we used Low-Rank Adaptation (LoRA), a Parameter-Efficient Fine-Tuning (PEFT) technique with Transformer Reinforcement Learning’s (TRL) Supervised Fine Tuning (SFT).” That’s definitely a lot of terms but check out the article to understand how it was done.
Databases
First up, we cover a summary post that gives you all the goodness and new features that have been announced in Google Cloud databases. If there is just one post that you read for Databases, then check out this post titled “Whats new for Google Cloud databases”.
Use multiple Google Cloud databases across your applications? Enter Database Center, an AI-powered, unified fleet management solution in preview that provides the following:
Comprehensive view of your entire database fleet.
Proactively de-risk your fleet with intelligent performance and security recommendations.
Optimize your database fleet with AI-powered assistance.
The last feature enables you to use a natural-language chat interface to ask questions and quickly resolve fleet issues and get optimization recommendations. Check out the blog post for more details.
There are a couple of Memorystore updates:
Valkey 8.0 on Memorystore is now available in preview.
Cross-region replication and Single-shard clusters (cost-effective high availability for smaller workloads) are now available in preview for both Memorystore for Redis Cluster and Valkey.
In Database Migration news, SQL Server migrations from on-premises and other clouds to Cloud SQL for SQL Server are now generally available as part of Database Migration Service (DMS).
In AlloyDB updates, ScaNN for AlloyDB index is now general availability (GA) and helps build scalable, performant, and accurate Gen AI and search applications. The ScaNN for AlloyDB index is the first PostgreSQL-compatible index that supports more than a billion vectors and maintains performance. Check out the blog post for more details on how to get started.
Data Analytics
BigQuery as a platform for all your data needs continously to see rapid features being added to it. If you’d just like to read one single blog post that captures several features that have been added in the last several weeks, check out this post that highlights BigQuery platform enhancements. Read on in this section for a few key features that are worth highlighting.
Complex nested queries in BigQuery might be a thing of the past with the introduction of “pipe syntax”. You visually separate different stages of a query with the pipe symbol (|>), which makes it easy to understand the logical flow of data transformation.
An example of the pipe syntax is shown below and you can clearly see how it makes each step, self-contained and easy to understand.
Check out the blog post that highlights this new syntax, along with other enhancements introduced in BigQuery. One of the use cases that is well suited to this syntax is log analytics, and the pipe syntax for log analytics is available in preview.
AI models are only going to be as good as the data that is provided for them to learn from. But real-world data brings with itself, several valid concerns around data privacy. Given that the need for data is real but at the same time, protecting PII data is important and hence the rise of Synthetic Data generation, which comprises artificially generated datasets that statistically mirror real-world data. If you are looking to generate Synthetic Data within BigQuery, you can check out the integration that Gretel provides in conjunction with BigQuery Dataframes. Check it out.
If you are going to base your complete data strategy on the BigQuery platform, then data governance is going to be needed. This includes data discovery, data quality checks and more. Dataplex is the solution for that and it has added several new features towards that. This includes Automated cataloging, AI-powered data insights and more. Check out the blog post to learn more.
If you are running applications across multiple systems in a multi-cloud environment, aggregating logs, indexing and then doing analysis/visualization on them is an important requirement but comes with its own challenges vis-a-vis an architecture optimized for speed and cost. Check out this solution that proposes BigQuery Omni as part of a solution that supports the TCO reduction (engineering, cost, complexity and more) of Log Analytics workloads in a multi-cloud environment.
In what is another excellent engineering feat, consider remaining in BigQuery, using the power BigQuery SQL and yet being able to query your Spanner transactional tables as if they were BigQuery datasets. That is exactly what BigQuery external datasets for Spanner means. As the documentation states “It is a connection between BigQuery and an external data source at the dataset level. It lets you query transactional data in Spanner databases with GoogleSQL without moving data from Spanner to BigQuery storage.” Check out the blog post for more details and note that the feature is available in public preview.
What is history-based optimizations in BigQuery? The definition states “History-based optimizations use information from already completed executions of similar queries to apply additional optimizations and further improve query performance such as slot time consumed and query latency.” This feature has been available in public preview and has seen instances of a 100x improvement at times. Check out the details on this feature and how it has been implemented over here and try out a sample query too.
Developers and Practitioners
Google has been highlighting key features of Gemini models like larger context size to help completely change how use cases around coding and understanding large code bases are done.Gemini Code Assist Enterprise is now available for customers to try. Key features include:
Local codebase awareness: Long context in Gemini is put to good use here by allowing you to prompt it for generating / understanding / fixing code that spans across multiple files.
Code Customization : Received suggestions from your code, which takes into consideration your coding styles, best practices and more.
Check out the blog post for more details and steps on how to enable your project for previewing these features.
Storage and Data Transfer
Gemini is making its way into multiple Google Cloud services to help customers understand the current workloads, ways to optimize the current deployment and more. One such service is the Storage Insight service, powered by Gemini that does 3 things as mentioned in the blog post:
eliminate manual data analysis and get answers rapidly
proactively find potential security and compliance risks
identify possible cost-savings opportunities to optimize your storage spend
Check out the blog post for more details and how to get started.
Application Modernization
What does an Application Platform look like and especially one that is designed to take care of modern day AI workloads and gives your developers what they need to build these new class of applications, now and in the near term for sure. In a new paper that highlights what this Application Platform looks like, with key new areas that you for modern day AI applications, get a head start today to understand where your platform engineering needs to go.
Cost Management Tools
Who does not want to save on Cloud Costs? In fact practicing FinOps efficiently is a core requirement now in running workloads in the Cloud. With AI being injected into every service, it comes as no surprise that the heavy lifting of understanding current costs, doing anomaly analysis is now best driven by tried and tested models to provide information back to the operators. Cost Anomaly Dtection, now available in public preview, at no additional cost, is a tool to “identify unusual spikes in cloud spending, across all products and services, by automatically monitoring your cloud projects and displaying any spikes in your billing console.” Check out the blog post that highlights how to detect, investigate and setup alerts. As the post indicates, combining this knowledge with Budgets is a powerful combination.
Assigning Labels to resources is often reported as one of the key things to do to help report across your cloud resources. They go a long way in identifying teams, projects, or services that are driving your expenses. Google Cloud has introduced a Cost Attribution Solution, labelled “GCP Labels for Enterprise” that provides suite of tools and best practices is designed to improve your cost metadata and labeling governance processes, enabling data-driven decisions so you can ultimately optimize your cloud spending. The blog post describes the solution with a Github repository for the solution.
Learn Google Cloud
Let’s first learn about Supervised Fine-tuning (SFT). SFT is an approach that helps us train a model to address domain specific and other nuances that the standard model does not? But do we need to go for SFT v/s Prompt Engineering, RAG and other approaches? That is exactly what this article covers.
Dataflow is fully managed streaming platform with AI capabilities. One of the best ways to understand this is to see how solutions have been put together, ranging from real-time clickstream analysis, log replication and analytics, real-time ETL and more. Check out this guide that discusses 5 Dataflow solutions.
Stay in Touch
Have questions, comments, or other feedback on this newsletter? Please send Feedback.
If any of your peers are interested in receiving this newsletter, send them the Subscribe link.