Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.
Insight and analysis on the data center space from industry thought leaders.
How to Optimize Flash Storage in the Data Center
The ability to mix high density with traditional flash in a real-time auto-tiering manner will dramatically reduce the cost of flash, enabling its broader use for all workloads across the data center.
Industry Perspectives
May 23, 2016
3 Min Read
Rajesh Nair is CTO of Tegile.
All-flash storage systems are about to spark a renaissance within the data center. Until very recently, the higher costs of all-flash storage has prevented the technology from being adopted across a majority of data center workloads. However, the recent introduction of high-density flash will certainly change this situation, as the ability to mix high density with traditional flash in a real-time auto-tiering manner will dramatically reduce the cost of flash, enabling its broader use for all workloads across the data center.
Flash is still uncharted territory for many data center executives. As the use of all-flash becomes more common, it will be more important than ever to know how to assess key performance parameters and assess any cost challenges encountered along the way.
Key Flash Performance Parameters
Flash storage has turned the traditional concept of IOPS (Input/Output Operations per Second) upside down. In the past, customers with legacy storage systems (e.g., spinning disk), needed to buy as many spindles as possible in order to achieve a good IOPS rate. This strategy is no longer necessary – storage administrators can get killer IOPS performance with smaller storage footprint, as well as much better OPEX from inline dedupe and compression.
There are three main performance parameters (along with IOPS) that every storage administrator should be aware of: block size, latency and bandwidth/throughput (IOPS x IO-size = bandwidth). Using these as the guiding parameters, IT administrators can then identify the most common missteps based on those that occur inside the storage array, between the array and the server, the network, and inside the server.
Inside the Storage Array
Storage administrators often experience issues within the storage array when server CPU utilization is running high. When running with small block IO (e.g., 4k), every read/write segment is running CPU cycles at the protocol layer.
Chances are you are running up CPU capacity. There are a few solutions to this issue:
Add more CPUs. When running applications servers beyond what they are designed for, IT administrators must decrease the amount of tasks or add more CPUs. Otherwise, they will never get the sub-millisecond latency that they crave.
Check for unaligned write IOs. Many of today’s operating systems are designed to automatically help with partition alignment. If you have an older one, these too can be configured with the right flags set.
Between the Storage Array and the Server
The primary performance issue that arises between the array and the server is caused by disk drive latency. If a data center is not achieving optimal performance, storage administrators can look to flash as a way to fully utilize the HBAs or CNAs and horsepower in each application server.
Inside the Server
If your entire server is having issues, you can troubleshoot your configuration on multiple levels –both on the hardware level and software level:
Rearrange your hardware. Configure all adapters so that they follow rules of affinity, on the same side of the NUMA bus.
Read into your own operating system's multi-pathing practices. Every operating system has its own set of multi-pathing tools (e.g., Perfmon on Windows, iostat, vmstat and sar on Linux).
Utilize hypervisor and in-guest tuning. There are many tools available that will help tune arrays (such as ESXtop) to an optimal setup, as well as guidelines – for example, IT administrators can tune a guest OS for lower CPU utilization by using PVSCSI instead of alternative SCSI controllers.
This is by no means an exhaustive list of issues, parameters or solutions; but they are the most common issues an administrator will face. All in all, flash technology is a boon for data center administrators and businesses, as it gives businesses the ability to compete in today’s real-time, increasingly data-intensive world.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.
About the Author
You May Also Like
