How do you make an informed decision to buy flash storage?
Flash storage is over 28 times faster than a spinning disk! The advantages of this speed for improved application performance, and also making your use of data and ICT more efficient are impressive.
But technology is rapidly developing, a whole range of options are available, and there are many suppliers of all-flash storage in Australia including Pure Storage, HP, HDS and Oracle.
So, how would you make an informed decision when buying flash storage?
As always, you need to work through the facts or myths on flash degradation so you can understand how an all-flash storage array can really help you.
First, you need to know the Input/Output (I/O) profile of your applications, and particularly what your requirements are for ‘hot vs. cold’ data. Then, you need an inventory and road map of both your current needs and your future operations, to look at how your current requirements will grow into the future state.
Figuring out your I/O footprint is crucial! Which data is ‘hot data’ and needs to be accessed the most by your organisation? What about the data that you don’t touch on a regular basis (‘cold data’)? It’s important that you always optimise where your data resides and off-load that ‘cold data’, data that needs to be online but not required on a regular basis. Cold data shouldn't be kept on a high-performance flash layer, but instead on a high-capacity lower performance location. That way you always get the best performance out of the flash modules for your hot data, while spending wisely with storing your cold data on cheaper disk.
As an example, a supermarket might want to send more supply runs of trucks out to stores to restock shelves. Doing so requires performing their inventory and stock replenishment reports more often each day. Hence, they would need to read each store’s live inventory data more often, but not write it as often. The solution? Having that data reside on flash storage would dramatically improve the system performance of running those reports.
Another important consideration is the block size of I/Os that your applications use. Some storage arrays can handle variable block sizes and mixed workloads, others need to be manually configured on a per LUN basis, and some just do not cope with this at all! Different systems and applications have varying needs, from 4k writes all the way to multi-megabyte stripes. A VDI environment is different to an OLTP database or a research platform. Understanding your I/O profile is crucial here, as purchasing the wrong array can lead to a wasted investment that does not meet your business requirements.
This leads to a very simple, yet important question: Where is the real bottleneck in my infrastructure stack? Sometimes it’s not the storage layer that is holding up your I/Os. It might be the network or resource constraints on the servers. Sometimes it’s the actual application software itself or a poorly indexed database. Never just look at one area to solve your problem: look at the bigger picture. You should always be searching for better ways to optimise your entire technology stack, or else you may waste good money on solutions that don’t really address the particular problem you’re experiencing.
Finally, data protection and disaster recovery is required for almost all organisations as the reliance on data becomes more critical. Most organisations have strategies already in place to mitigate the risk of losing data, and data replication is a key technology for doing that. Now, when you introduce flash storage into the mix, your data replication can quickly become a bottleneck to your overall performance and undermine the benefits you’ve invested in. This is because the sub-millisecond write performance you get from a flash storage array is curtailed due to a higher latency inter-site link (ISL) between your production and disaster recovery facilities. If your ISL links for data replication do not match the expected performance of the storage arrays themselves, then the replication itself will hurt the IOPS that is delivered back to your applications (especially when synchronous replication is used). Make sure you cater for the massive leap in storage performance where your disaster recovery strategies are concerned.
Interested in learning more? NTT ICT has just released an e-book to help our clients understand flash storage and the key issues surrounding this technology. Download your free copy here.