However, demand on higher memory performance brought about by multiple-core CPUs, data growth, and the consumerisation of IT, is contributing to what Heoh said is the storage I/O crisis. This is further exacerbated by randomisation of the virtualised or consolidated memory architecture due to virtualisation, data consolidation and different cloud architectures. In addition, demand for ever decreasing seek times and higher rotating disk speeds will add further pressure on data storage.
There is also the need to understand the application's I/O "fingerprint". By that, Heoh meant the app's requirements like I/O load, memory block size, locality of access, access pattern, and latency sensitivity.
Flash may be of help to boost overall performance. Heoh highlighted some issues that one would need to consider in evaluating flash alternatives. One is the cost of the memory (cost per gigabyte for example). Others include performance of the drive itself (read/write input/output per second or IOPS); power requirements and physical size of the drive; protection consideration in application such as high availability, backup and managing drive loss; integration issues like fitting in to existing architecture; and operational simplicity.
He also talked about different approaches to flash architecture, namely, server-attach PCI flash, array with flash cache, array with flash tier, and an all-flash LUN or array.
In conclusion, Heoh reminded the audience that data centres are facing a storage I/O crisis due to increasing demand on memory storage, randomisation of memory architecture, and the shrinking IOPS per terabyte. He stressed that CIOs should understand their application's I/O "fingerprint", which is key to choosing the best flash strategy for one's environment. "Flash should be evaluated on several dimensions; it's not all about performance," Heoh said. "Look at the full picture and go beyond evaluating just the raw cost per gigabyte to build your flash ROI."
Object storage systems
SNIA South Asia's education instructor Wong Tran gave a talk on the underpinning of cloud and big data initiatives. After describing what constitutes a cloud (according to the U.S. National Institute of Standards and Technology, NIST), he described the five essential cloud characteristics: on-demand self-service; broad network access; resource pooling; rapid elasticity; and measured service.
"Things like Google, Gmail, YouTube are examples of cloud services," said Tran. "Being able to access these systems is the characteristic of cloud computing."
But what kind of storage can meet these criteria? "If you look at the evolution of storage, traditional spinning drives with block storage led to storage arrays and separate file systems that rely on tree structures to handle files instead of raw bits like in a storage array, which makes storing large amount of data a challenge," said Tran.
Sign up for Computerworld eNewsletters.