NetApp all new “All Flash FAS” (AFF)
Launched in June 2015, All Flash FAS (AFF) is NetApp’s new all-flash array family that is optimized for flash and built on clustered Data ONTAP. It delivers high performance and consistent low latency as well as superior flexibility and best in class data management. Priced aggressively, and ideal for customers who want to accelerate the speed of their business while not compromising on overall efficiency, reliability, and flexibility of IT operations. AFF is ideal for workloads that require consistent low latency with high IOPS random reads and writes, robust data management, and scale-out flexibility. Many database applications need this combination, as do VDI and VSI deployments where high load and fast response times are requirements.
The AFF8000 series leverages a high-performance architecture (with multicore Intel chipset, DDR3 memory, and PCIe v3) that boosts throughput while maintaining low latency. The parallelism of Data ONTAP better utilizes the increased number of cores present in the controllers, optimizing them to take full advantage of the high throughput of the SSDs. In other words, AFF systems do an excellent job of “lighting up all the cores”!!!
We recently deployed an AFF HA pair with an existing Data ONTAP cluster. This cluster already had a hybrid array FAS HA pairs. This deployment now enables us to non-disruptively move volumes and LUNs between the different nodes in the cluster. Doing this gets our data onto the optimal storage tier based on performance requirements and cost considerations. Before deploying the new AFF HA pair, we ended up running some Performance tests using IOmeter. Here are some of the high-level performance tests we ran on a NetApp’s new AFF8020 storage.
Physical Setup Storage: NetApp AFF8020 18+2 SSD Aggregate ISCSI LUNs
Windows Hosts: Running IOmeter workload IOmeter Queue depth of 32
The purpose of these tests was to smash the AFF controller to its maximum performance threshold, with the existing storage.
So we started with a single host running IOmeter with four workers and gradually added workers. Eventually, we ended up having four hosts (3 Physical + 1 Virtual) running eight workers each when we could see AFF8020’s CPU utilization hitting 100%.
**Both the tests were performed with the transfer size of 4KB.
We mainly focused on two tests: 4KB 100% Random reads 4KB 30% Write and 70% Read (random)
4KB 100% random reads:
When we ran the 100% Random read test we were able to get around 115k IOPS (in total) on the Windows Host side. The latency observed on all 4 hosts was well below 10 milliseconds. The screenshot below shows the Storage “sysstat -x 1” output while this test was running. Here we can see that the CPU’s is running around 100%. Also, disk utilization is above 90%.
4KB 30% Write and 70% Read (random): To gauge how the AFF8020 would perform in the production environment we ran a 30% write, and 70% read (both random) test. Here we were impressed with the IOPS achieved.
We were able to squeeze 80,000 IOPS with a latency hovering around 11 milliseconds on the each of the four hosts. Storage CPU was again around 100% with this test.
This testing demonstrates how the NetApp All-Flash FAS8020 system can supply low latencies with high IOPS demands. NetApp provides you the ability to offer flash-to-disk-to-cloud, unified protocol support, non-disruptive operations, and integrated data protection on a single storage system.