Intelligent CIO Middle East Issue 03 | Page 80

FINAL WORD We will let you in on a secret: Most hard disk arrays, which have appeared on the market in the last five years are actually hybrid arrays. Provided the active data inventory and dataset fits into the highspeed flash tier, hybrid arrays can achieve high performance. But as soon as the hybrid flash tier is bypassed or overrun due to an increase in random access or a change in the datasets, the performance and more importantly the latency, will suffer enormously, because the read-write speed of hard disks is very slow (in the doubledigit millisecond range). An all-flash array can sustain high performance and consistent submillisecond latency in such environments because there is no tiering layer as all data is kept on flash storage. The lowest common denominator to performance and latency is the all-flash tier. Flash storage makes the difference The performance benefits of flash storage are already well-known: The short latency times and the capability to supply hundreds of thousands of IOPS ensure that virtual desktop infrastructures (VDIs) respond significantly quicker, virtual server infrastructures (VSIs) can be managed on a more predictable and simpler basis, and databases and business analytics applications can be expedited. MOHAMMAD DUALEH Regional Sales Manager, MENA, Pure Storage To achieve these performance benefits and to be more reliable and also more affordable, than mechanical disk, flash storage cannot be treated like mechanical hard disks. Particularly in terms of reliability, the error behaviour of the two storage media, for instance, differs completely. If you examine how flash storage is worn down with respect to bit error rates, you can determine the following: • In the case of hard disks, mechanical defects are caused by vibration, constant motion, voltage fluctuations and heat. Bit errors frequently occur in close proximity to each other, and multiply exponentially in proportion to the capacity and age. 80 INTELLIGENTCIO • In the case of flash storage, bit errors can be attributed to wear (program erase cycle). The medium simply wears out over time. First it is only possible to read the data, then comes complete failure. The rate of uncorrectable bit errors, on the other hand, remains relatively constant on hard disks in relation to the number of writes processes. In case of flash storage, read processes are free and write processes expensive – behaviour which is the opposite of that on a hard disk. The read and write processes are largely asymmetrical, www.intelligentcio.com