X

Oracle Bare Metal Cloud - DenseIO Virtual Machine Performance Analysis

Hi everyone, I’m Lee Gates and I'm excited to update you on the awesome performance of our new DenseIO Virtual Machines. My colleague Omair posted an overview of the new compute shapes, and I will review the results of my testing. When you need high performance local NVMe storage, but only a few terabytes or a few cores, this is the right service to consider. The table below shows how the new compute shapes compare to the Standard VM shapes we have been offering to date:

 

Shape

Core

RAM

(GB)

Networking

Storage

Dense I/O VM shape

 

 

 

 

VM.DenseIO1.4 (New)

4

60

High

3.2 TB NVMe SSD

VM.DenseIO1.8 (New)

8

120

High

6.4 TB NVMe SSD

VM.DenseIO1.16 (New)

16

240

High

12.8 TB NVMe SSD

Standard VM Shape

 

 

 

 

VM.Standard1.1

1

7

Moderate

Block only

VM. Standard1.2

2

14

Moderate

Block only

VM. Standard1.4

4

28

Moderate

Block only

VM. Standard1.8

8

56

High

Block only

VM. Standard1.16

16

112

High

Block only

 

High performance and storage hungry workloads are difficult to architect, deploy, and operate. We're focused on helping deliver a seamless experience, arguably removing the traditional hardest problem: sizing a difficult performance requirement.

Now let’s dig into the numbers. Our local NVMe block volumes are designed to deliver sub-millisecond latency, high peak performance, and repeatable consistent performance. These are the same NVMe volumes we delivered with our Bare Metal Cloud launch delivering up to 4 million IOPS in bare metal compute instances configured with 9 NVMe block volumes.

New DenseIO Virtual Machine instances use local NVMe volumes to deliver up to 1,800,000 IOPS, hitting the target of a high performance virtual machine and your choice of cores and networking. When focusing on sizing your high performance or demanding workload, understanding the sustained maximum performance is the key ingredient for delivering workload success. I've talked with several customers who are delighted to hear our performance and next generation NVMe architecture consistently deliver.

Here's how I think about these new compute shapes:

  • 4 Core Dense VM - Our smallest high performance virtual machine with one NVMe block volume
  • 8 Core Dense VM - 2x RAM, 2x NVMe, best for increased capacity or transaction performance
  • 16 Core Dense VM - x4 RAM, 4x NVMe, best for highest transaction performance or advanced architectures

We're testing the local VM performance using the same test used for our Block Volume Update, this time running against local NVMe storage. For these tests, we'll use a wide range block sizes and the most common read and write patterns. These tests and charts were generated with the Gartner Cloud Harmony test suite. All of these tests are run consecutively 5 times to achieve steady state results.

We'll start with the 16 core DenseIO Virtual Machine. The first chart is for using all sixteen cores against a four local block volumes we see 1,800,000 read IOPS. The second chart and test shows when we assign all sixteen cores to a single local NVMe block volume. We see that the results showing 800,000 read IOPS for a local block volume. In the charts below the X axis describes the block size tested, ranging from 4KB to 1MB, the Y axis describes the number of IOPS delivered and the Z axis describes the read/write mix tested, ranging from 100% read to 100% write. This test shows many of the storage configurations applications use in one chart.

Before running any tests, protect your data by making a backup of your data and operating system environment to prevent any data loss.

WARNING: Do not run FIO tests directly against a device that is already in use, such as /dev/sdX. If it is in use as a formatted disk and there is data on it, running FIO with a write workload (readwrite, randrw, write, trimwrite) will overwrite the data on the disk, and cause data corruption. Run FIO only on unformatted raw devices that are not in use.

Test Details Observed Performance: 16 Core DenseIO Virtual Machine, 1,800,000 IOPS
NVMe Block Volume Capacity: 2.9TB x 4
Direct I/O
Host Shape: DenseIO Virtual Machine
Region: Phoenix
Reproduction Steps
  1. Provision 16 Core DenseVM
  2. Run Gartner Cloud Harmony Block Storage
    1. ~/block-storage/run.sh --nopurge --precondition_once
      --target /dev/nvme0n1,/dev/nvme0n1,/dev/nvme0n2,/devnvme0n3
      --skip_blocksize 512b
16_core_4_drives_workloads.png
16_core_workloads.png

 

Let's look at the 8 core DenseIO Virtual Machine. The first chart is for using all eight cores against a two local block volumes we see nearly 1,000,000 read IOPS. The second chart and test shows when we all 8 cores to a single local NVMe block volume. We see that the volume has quite a bit more performance than the 4 core test, showing 800,000 read IOPS for a local block volume.

Test Details Observed Performance: 8 Core DenseIO Virtual Machine, 1,000,000 IOPS
NVMe Block Volume Capacity: 2.9TB x 2
Direct I/O
Host Shape: DenseIO Virtual Machine
Region: Phoenix
Reproduction Steps
  1. Provision 8 Core DenseIO Virtual Machine
  2. Run Gartner Cloud Harmony Block Storage
    1. ~/block-storage/run.sh --nopurge --precondition_once
      --target /dev/nvme0n1,/dev/nvme0n1
      --skip_blocksize 512b
8_core_2_drives_workloads.png

 


8_core_workloads.png

 

For a single drive with a 4 core DenseIO Virtual Machine, we see 580,000 read IOPS. We focus on the 4K block size since it is used for the largest number of workloads.

Test Details Observed Performance: 4 Core Dense VM, 580,000 IOPS
NVMe Block Volume Capacity: 2.9TB
Direct I/O
Host Shape: DenseIO Virtual Machine
Region: Phoenix
Reproduction Steps
  1. Provision 4 Core DenseIO Virtual Machine
  2. Run Gartner Cloud Harmony Block Storage
    1. ~/block-storage/run.sh --nopurge --precondition_once
      --target /dev/nvme0n1 --skip_blocksize 512b

DenseIOVM_4_core_workloads.png

Using the Gartner Cloud Harmony test suite to generate these test results presents a great overview of every question you might have about local block volume performance. If you wanted to get these results yourself, we're still running the $300 free credit promotion. These new virtual machines are in our cloud store cost estimator. To run a 4 core Dense VM is $15/day, the 8 core is $30/day, and the 16 core estimate is $60/day.

Our DenseIO Virtual Machine service hits the sweet spot of incredibly high performance NVMe storage paired with 4, 8, and 16 core virtual machines. Each NVMe block volume delivers up to 800,000 read IOPS, and meets demanding requirements from database and high performance applications. We are providing our customers with the ability to validate our numbers and our competitor’s offerings with an "apples to apples” comparison test. This combination of performance and value is incredible and makes sizing straightforward as you consider your move to cloud.

Please feel free to send your feedback and most challenging performance questions. I've attached the full test reports for each of the configurations mentioned here. We are delighted to deliver this best in class service with a perfect match to begin considering your mission critical workloads and difficult performance problems. We want to ensure your success, if you want more information on our performance methodology, have questions on specific workloads or need help achieving similar results, please reach out to me at lee.gates [-at-] oracle.com

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.