Tanel Poder and Kellyn Gorman team up on a webinar to talk Oracle on Azure

Silk recently hosted a webinar, Lift, Shift and Evolve: Moving Oracle onto Azure, with two experts on high performance Oracle databases and Microsoft Azure:

Tanel Poder, a well-known blogger, entrepreneur, self-proclaimed database performance geek, and co-author of Expert Oracle Exadata, and Kellyn Gorman, Principal Cloud Solution Architect at Microsoft, SME in Oracle on the Azure cloud, with a specialty in Oracle Exadata.

You can watch a recording of the conversation here. But this post will cover some of the highlights of what was discussed:

Are Big Oracle Databases Going Away?

When Tanel and Kellyn first met 25 years ago at the Rocky Mountain Oracle Users Group, the cloud as we know it did not exist. If you wanted to build a big business, you went to Sun Microsystems and bought servers, or you went to Oracle for your database, Tanel explained. The world has changed a lot since then. In the cloud there are more options.

So, what’s the “state of the state” for big Oracle databases on the cloud?

Monolithic environments that many people see as archaic and going away… ARE NOT, remarks Kellyn. These systems are foundational for many companies and are the main data sources being utilized in the cloud for Machine Learning and AI. Instead, the question becomes: How do you lift-and-shift these workloads into the cloud and do it right?

At Microsoft, Kellyn works with Oracle customers who often believe Azure cannot handle their workloads. It’s her job to demonstrate how it can be done. While customers think of CPU and memory as requirements, it’s throughput that needs special attention for high I/O workloads to succeed on the cloud.

Rewrite Apps or Lift, Shift and Evolve?

Ten years ago, we started hearing a lot about NoSQL databases and distributed, scale out approaches. Instead of using a big server that can do a lot of I/O, customers can use a NoSQL database and scale out instead. But in order to use these new databases, customers need to rewrite existing applications to work with the new database.

How realistic is it to take 100 databases, export the data, and rewrite the SQL?

Kellyn explained that some applications may be simple to rewrite, while others will take years and years (and hundreds of thousands of hours). She estimates that 3 out of every 5 workloads simply cannot move off Oracle.

But that’s not the end of the road. Customers can still take the data and bring it into the cloud through an Extract, Load, Transform (ELT) process, put it into data lakes, and extract more value. This is the concept behind lift, shift and evolve.

Is Oracle a Fit for PaaS or IaaS?

Kellyn specializes in decoupling workloads from Oracle Exadata and bringing high I/O workloads into Azure. While every cloud provider’s goal is to let customers do more with what they have, going to the cloud doesn’t mean you need to use PaaS services. IaaS, many times, is the goal that you must have for complex, transactional systems that demand fast performance.

Kellyn remarks that with Exadata implementations, many customers used hardware (or “threw iron at the solution”) to get the performance needed. When looking to make a move to the cloud, the performance demands can be 900,000 IOPS and 10,000 NICs per second. This is where Silk comes in and makes a big difference for high I/O database workloads on IaaS. Silk has proven performance of over 1 million IOPS, 20 GB/S throughput, and consistent sub-millisecond latency to handle the most demanding workloads coming from Exadata. Silk’s data compression is also an important feature. When moving from Exadata to Azure (or the non-Oracle cloud), customers lose support for Oracle Hybrid Columnar Compression (HCC), and the I/O explodes. Silk compresses data to enable Exadata-level performance.

Benefits of the Silk Platform

Tanel tested the Silk Platform for Oracle workloads on Azure. The results showed Silk delivering 5 GB per second of data scanning speed per single VM in a single cloud instance. With the launch of the Azure v5 VM with the latest CPU and network cards, Silk delivered 10 GB per second of scanning speed and 6 GB of writing speed for a single VM.

The Silk Platform is an infrastructure solution that integrates into the cloud vendor’s environment. Silk leverages the compute network (as opposed to the storage network) and thus can deliver significantly faster and consistent database performance to support high I/O workloads. As the cloud vendors continue to improve compute network speed, Silk will continue to boost performance alongside network gains with customers reaping the benefits.

Is Oracle RAC Needed on Azure?

When moving Oracle workloads from Exadata to Azure, the inevitable question comes up: What to do about Oracle RAC?

There is RAC bare metal that customers can do inside of Azure. It is a colocation. But it is not supported by Oracle. RAC is only supported by Oracle inside of the Oracle Cloud.

Kellyn advises the most essential question to ask when discussing Oracle RAC is whether you really need it on the cloud. Of the 2,000 Oracle customers on Azure, only 3 customers needed RAC. Some customers associate RAC with high availability when RAC is really designed for instance resiliency and scalability. If using RAC for rolling patches and upgrades, for example, customers can use Active Data Guard instead, which is the high availability portion of the product.

Kellyn points customers to the Maximum Availability Architecture whitepaper from Oracle. She notes that Data Guard works very well with Azure and that Azure has always-on availability groups for SQL Server and Azure SQL to seamlessly switch over between environments.

Tanel and Kellyn agree that the discussion of RAC vs. Data Guard is an entirely different conversation when talking about what’s available on the cloud compared to on-prem systems with physical hardware.

In discussions with customers, it’s now about architecting for the cloud and using the right solutions for the cloud.

If you’re interested in going deeper into the discussion, you can watch the webinar on-demand here.

Be sure to also check out these resources to learn more: