In the last few years, flash storage has turned from very expensive into
quite affordable. Vendors that sell all-flash arrays advertise the extremely low latencies, and those are indeed truly impressive. So it may feel like all-flash systems will solve all your performance issues. But reality is that even with game-changing technological advances like flash, the complexity of the entire infrastructure makes sure that there are still plenty of problems to run into.
Here’s our top 5 reasons why all-flash arrays won’t magically solve all your problems:
1. Shifted bottlenecks
In traffic, opening a new lane on the freeway helps solve congestion where the lane is added, but it tends to shift the weakest link. After a while, the delays will pop up elsewhere. In your IT infrastructure it works exactly the same way. Fast flash arrays will solve disk drive bottlenecks, but it shifts the bottleneck to an unknown new location. Eventually, other components inside the storage infrastructure run a risk of becoming overloaded, like channels, ports, adapters, or internal bandwidth. These new bottlenecks lead to queuing, long response times, and even application outages because of timeouts.
2. Suboptimal tiering
Most enterprises will not have everything on all-flash, but will keep certain applications on spinning drives, or on a combination of spinning drives and flash storage. This is called tiering, and this can be done within an array, or across different storage arrays. The decisions that you make when you choose which tiers to buy, and how to utilize them greatly influence performance, but visibility is often reduced because of the added layer of complexity.
3. Sequential streams
Flash drives still have a relatively low maximum sequential throughput; this can become a bottleneck especially with bursty I/O or when you do not pay close attention to maintaining a well-balanced environment. For application workloads that are highly sequential, spinning drives still offer the best value.
4. Remote link bottlenecks
Even if your arrays offer very fast backend disk response times, if you have a secondary site with insufficient link throughput, it will hurt you. In case of a synchronous remote copy solution it will impact your front end response time, and in case of an asynchronous solution it will lead to a secondary copy that is not up to date enough. Both pose a threat to your continuous availability.
5. Configuration errors
Errors, anomalies or unwanted changes in the configuration can lead to lack of redundancy and threats to availability, regardless of your drive configuration. Having flash doesn’t change any of that. So regardless of your drive technology, you always need to be able to quickly identify configuration errors in the environment before they cause production issues.
In summary; flash is great, but not the answer to all your problems. Even when you use all-flash arrays, you will still need to manage your IT infrastructure intelligently to avoid issues that impact application availability.
If you are interested in storage performance considerations when purchasing new storage this white paper may interest you. We also have a deep technical whitepaper on which performance and capacity metrics are relevant when managing SAN storage. For more information please contact us.
Flash Performance in High-End Storage
This paper analyzes the performance of Flash and SSDs when deployed in high-end storage, using RMF measurements from production systems.
Should I Disable SVC Write Cache for Flash?
It has been suggested that in order to maximize throughput for large sequential operations on SVC volumes residing entirely on IBM Flash systems, you should consider disabling the write cache.
How to Determine Flash Candidates in Your Environment
If you are getting ready to add flash, it is important to see how much of your capacity is a good candidate for flash so you aren't throwing away money on flash you don't need