[Originally published on LinkedIn 13 June 2024]
Jeff Welsch, Product Management Leadership, Google Cloud presented to Cloud Field Day 20 on Google’s Modern Infrastructure Cloud.
Modern enterprise workloads are changing. You can break down workloads into three categories:
- AI driven: language, vision, speech, recommendation, etc.
- Cloud-first: containers, serverless, modern databases, streaming services, etc.
- Traditional enterprise: SAP, VMware, Microsoft, HPC, etc.
To support this, Google is rearchitecting and disaggregating the traditional server. Google has developed the Titanium offload processor. Using PCIE, Titanium offloads storage and networking, enabling the scaling of compute, storage, and networking independent of each other.
Google claims massive benefits such as 80% improved CPU performance, three times more packets per second network throughput, and non-disruptive maintenance.
The modern infrastructure cloud provides bare metal general purpose machine instances and specialized workload instances where the system is optimized for compute, memory, storage, or AI acceleration.
In the true sense of modernization, Google is expanding beyond x86. The Axion-based C4A instance uses Axion ARM processors providing up to 50% better performance and 60% better power efficiency compared to x86. These instances still use the Titanium IPU for offloading storage to Hyperdisk.
What about Microsoft applications?
Google understands that much of the modern enterprise still uses Microsoft-based workloads and believes that Google’s modern infrastructure cloud provides:
- Savings via cost optimization: custom machines, custom visible cores, scheduled suspend/resume and right sizing recommendations.
- Reliable, performance, secure infrastructure with sole-tenant infrastructure with over commit, live migration, multi-write HA for SQL server.
- Ease of modernization with managed services for GKE/Windows, managed AD, Cloud SQL for SQL server, NetApp volumes.
Google has partnered with NetApp and is supporting NetApp volumes on Windows for user and group unstructured file shares, VDI, and MS-SQL shared storage. Policy-based auto-tiering is available and is transparent to applications as is moving to/from active/inactive tiers.
In the Modern Infrastructure, is VMware Still a Thing?
Yes, it is, and Google provides the complete Cloud-integrated VMware experience, with all the capabilities you expect but running in the Google Cloud. And the Cloud is built for massive scalability providing the largest clusters.
Importantly, Google Cloud is the first cloud provider to support VMware Cloud Foundation (VCF) license portability, enabling you to quickly and easily move your VMware environment to Google.
The benefits include 58% faster app migrations
SAP on Google Cloud?
Google it is supporting SAP including HANNA. The benefits include:
- Performance and scalability: up to 96TB scale out
- Reliability: 99.95% uptime LSA, live migration
- Easy business transformation: AI out of the box, connecting to Gemini, Vertex AI, etc.
- Security: Automated SAP security analysis, encryption at rest and in flight, DDOS protection and, most importantly, Mandiant managed services
- Manageability: SAP Workload Manager and Assured Workload for SAP
In a true eat your own dogfood world, SAP runs their own instances on Google Cloud!
What does this mean for you?
The compute industry is cyclical. General purpose CPUs get overloaded, so we develop offload processors for networking and storage. And then the general purpose CPUs get faster and cheaper and the network and storage processing moves back inside the CPU. And then we create heavier workloads and go back to offload processors.
But we seem to have reached the size, speed, power, and cost limits for general purpose CPUs. That plus Google is operating at a scale were they can produce cost efficient offload processors.
The benefit, of course, is that you can scale networking and storage independently of compute. And this makes the modern infrastructure cloud very interesting for large workloads.
Equally interesting is that Google really wants you to move your enterprise workloads to Google Cloud. Thus, they are making a concerted effort to support both existing and new workloads.
I’m sure the engineers can be sold on the technical benefits of Google Cloud. But can Google build a business case for moving off your own infrastructure (or another CSP’s infrastructure) to Google?