X

Oracle Tests Better in Performance than Amazon Web Services

Andrew Reichman
Director of Product Management

Oracle Cloud Infrastructure Compute bare metal instances is shown in independent testing by StorageReview to have a 2X-5X performance advantage with comparable or dramatically lower pricing, compared to similar configurations from Amazon Web Services (AWS) across a wide range of workloads. 

The Testing: End-to-End Workload Performance

In March 2018, StorageReview gave Oracle an Editor’s Choice award for the performance and innovation that they saw when testing Oracle Cloud Infrastructure bare metal and virtual machine instances. At the time, Oracle Cloud Infrastructure was the only cloud that they had tested, but the results compared favorably to on-premises configurations running the same workloads. In August 2018, StorageReview tested AWS i3.metal bare metal instances across the same range of workloads they ran for Oracle previously, and the results were a strong validation of the Oracle Cloud Infrastructure performance proposition for customers.

The testing done in the lab at StorageReview covers more than storage. The testing is end-to-end workload performance testing, and it measures all the components that make up the user’s experience on the tested platforms. The results provide an aggregate measurement of performance across compute, storage, and network components, and is about as close as a lab can get to estimating the performance that’s likely to be seen by a user.

The Results: Oracle is Up to 5X Faster than AWS

In the testing, Oracle demonstrated up to 5X the performance when running on remote block storage, and double the performance when running workloads on local SSD storage. Every workload tested, including Oracle Database, Microsoft SQL Server, 4k random read and random write, 64k sequential read and sequential write, as well as a variety of virtual desktop workloads, all showed a similar performance advantage for Oracle Cloud Infrastructure in comparison with the results for AWS.

Additionally, the latency recorded at peak performance was far lower on Oracle, and the percentage of recorded performance with latency below 1ms, the common threshold for application usability, was far higher. Latency has a powerful impact on variability of performance. Customers running performance-sensitive systems of record need performance consistency, one of the key design points of Oracle Cloud Infrastructure, and these results show that Oracle can deliver a higher level of consistency than AWS in addition to the higher level of performance.

Superior Oracle Database Workload Performance

When we designed Oracle Cloud Infrastructure, we knew that a primary use case for our customers would be Oracle Database and the critical business applications that run on top of our database, so we knew we had to deliver exceptional results for these demanding workloads. The results showed we hit the mark. For performance-intensive database workloads, Oracle Cloud Infrastructure offers performance results that are head and shoulders above the capabilities offered by AWS. 

The results with a configuration that uses remote block storage, network connected to bare metal instances on both clouds, shows the most dramatic advantage for Oracle. Oracle provides 5X the performance, as seen here:

How does Oracle get such a big advantage over AWS? With the remote block storage configuration, the answer comes down to the unique cloud architecture we've built to address the needs of enterprise users, and more specifically, how we built our network and our block storage service. Oracle has a next-generation cloud network that connects our cloud components, including between servers and the block storage sub-systems. The network has no resource over-subscription, so performance doesn't get compromised when the network gets busy.  Further, we used a flat network topology, which reduces the number of hops and the associated latency between any two devices. Off-box network virtualization offloads the effort from the server, which reduces the performance tax that customers would see without such an approach. Finally, storage traffic uses the full 25-Gbps pipe to the server, while AWS confines the storage traffic to their EBS optimized link that’s limited to 10 Gbps for their bare metal instance. The Oracle Block Volume Service is designed for maximum performance with all-SSD capacity and rates the highest IOPS per GB and IOPS per instance metrics of any block storage service in the cloud.

One of the key things that you can see in the performance comparison for remote block storage is that a higher percentage of the IOPS Oracle delivers is usable, with latency below 1 ms, the common threshold for application latency tolerance. In this graph, the percentage of unusable IOPS of the peak recorded for Oracle is 10%, while Amazon records 25% of its peak IOPS at unusable latency levels, both represented by the hashed bars at the top of the peak IOPS levels. Higher levels of latency contribute to variability of performance at high levels of performance. Part of Oracle's design point in cloud is to cap performance before latency becomes a major issue, making the performance we deliver less variable, delivering better results for critical workloads that need consistency as much as they need high performance.

With the local SSD configuration, the Oracle performance advantage for Oracle Database workloads is slimmer, but still significant. In this case, Oracle provides double the performance, but also gives customers more than 3X the local storage capacity, making this extremely high performance configuration far more usable for workloads that need to scale capacity over time. The comparative performance for local storage configurations can be seen here:

Fewer factors go into the performance difference when local NVMe SSD storage is used. Both vendors are using a similar media type, and there's no network connection between server and storage that impacts performance since the storage sits on-board the bare metal server. In this case, the Oracle advantage comes from the SSD drive itself, which has built-in cache that increases performance enough to drive the 2X performance benefit demonstrated. In addition to twice the performance recorded when running on SSD, Oracle offers 51TB of SSD on our bare metal instances, while AWS offers just 15TB, meaning that it's much more likely that customers can accommodate big scale applications, as well as the capacity needed for data redundancy and ongoing growth of darta on local SSD with Oracle than with AWS.

Superior Performance for SQL Server, Virtual Desktop, and General Workloads

While we built Oracle Cloud Infrastructure to be optimized for Oracle Database, the enterprise optimized infrastructure we built also has significant performance advantages over AWS for all the other workloads that StorageReview tested.  Customers with demanding performance requirements for any category of workloads will clearly find a good home with Oracle Cloud.

Here are the results for running Microsoft SQL Server, with Oracle delivering double the performance on local SSD and more than 5X with remote block storage, along with far better usable IOPS:

Here's what StorageReview measured for a 4K random write workload, with Oracle showing more than double the performance on local SSD and just under 5X on remote block storage:

And finally, this is how it broke down for a virtual desktop infrastructure (VDI) workload, a test of initial login, with Oracle showing 2.6X the performance on local SSD and almost 5X with remote block storage:

Price for Oracle Block Storage is 19X lower for up to 5X higher performance

The last thing is price. Although Oracle is delivering a huge performance advantage, the cost is lower than AWS in most cases, as has been validated in other independent analysis. For block storage, StorageReview built the highest performance configuration possible on AWS so that it would compare as favorably as possible. The problem with that, however, is that AWS makes customers pay for the amount of input/output performance that they consume, which drives up the cost dramatically. For this series of testing, Oracle delivers 4-5X more performance at 19X lower cost.  In the configuration that StorageReview tested, the total cost for the AWS solution was $69,794 per month, driven largely by the cost of storage performance which customers must forecast and pay for on Amazon's high performance storage offering, Elastic Block Storage Provisioned IOPS. The Oracle Cloud Infrastructure configuration with higher performance across all workloads cost $3,697 per month, with Oracle's Block Volume Storage service delivering superior performance without charges for IOPS consumption.

In the local storage configuration, Oracle costs slightly more than AWS, by about 25%. However, Oracle also offers double the performance, more memory and 3.4X the local storage capacity, meaning that we can run bigger workloads and accommodate more workload growth over time. For customers that care about performance, this is an equation that delivers tremendous value.

We built Oracle Cloud Infrastructure to deliver consistent high performance for demanding enterprise workloads of all kinds, and we’re thrilled to see the advantages of our design demonstrated so clearly. We invite users to try Oracle Cloud to see how it can help them solve their biggest business challenges with the confidence of industry-leading performance that doesn’t break the bank.

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.Captcha