IBM sees NVMe gains in V5100F, but more to come with better apps

IBM’s new NVMe-equipped all-flash array – the midrange V5100F – will offer a 2.4x performance boost in terms of reduced latency. But that could be doubled yet again when application makers optimise their software stack for NVMe flash storage.

That’s the view of IBM execs following the launch of the V5100F last week, which IBM claimed as the first midrange storage array in the market to come with NVMe flash.

The Storwize V5100F has end-to-end connectivity to hosts with NVMe-over-fabrics (NVMf), via Fibre Channel. With data reduction, IBM said the V5100F systems will support 2PB of usable flash in a 2U rack with commodity or IBM FlashModule NVMe SSDs.

The new IBM flash systems natively embed IBM Spectrum Virtualize software-defined storage and IBM Storage Insights analytics. 

With end-to-end NVMe/NVMf, IBM claims a 2.4x improvement through reduced latency.

According to Thomas Harrer, EMEA chief technology officer for IBM hardware sales, that reduction in latency could be doubled again when application makers optimise their software for use with NVMe flash.

“If applications were transformed, we would see a much better than 2x improvement in latency. Apps are made for SCSI, which is a software stack of many layers that translates into a lengthy code path,” said Harrer. “We are ready with the back-end storage, but the applications are not ready yet. It is not yet reality.”

The cited 2.4x boost in performance is versus IBM’s V5030F all-flash array with SCSI-based flash storage.

NVMe emerged as the latest incarnation of flash storage a couple of years ago. It had been usual to connect flash storage via protocols that use SCSI – eg, SAS and SATA – which was developed in the era of spinning disk. However, this didn’t allow systems to fully utilise the performance benefits of flash.

So, NVMe was developed, which did away with the SCSI stack and hugely increased the number of input/output traffic channels and queues possible with potential performance gains boosted by orders of magnitude.

This led to a number of startups emerging that offered storage based on NVME, but none could offer the same storage features available in “traditional” arrays.

That was because the huge increase in disk performance now meant there was a bottleneck at the storage controller and there just wasn’t the processor power to spare to provide data protection, data reduction, thin provisioning, encryption, etc.

So, like here with IBM, suppliers have settled for a performance boost from NVMe that is in the single figures of x, and have left the much greater gains to specialist system startups that target niche high-end use cases.