Storage vendors make us think that underlying storage can never have too much performance. You know, this makes sense! Just imagine having the storage that allows running applications slightly faster than any of your competitors. Customers will enjoy your responsive and fast services, and therefore they will prefer your company over any other guys on the market. This being said, it is a good idea to build in an all-flash array, or just add some NVMe disks: they will pay off anyway.
However, don’t hurry with investing in NVMes! First, they still cost you a dime. So, you need to carefully think over the purchase so that the resulting storage infrastructure won’t be overkill. Second, you need to change the protocols that you use; otherwise, you just won’t full throttle your NVMe disks.
StarWind introduces support for NVMe over Fabrics (NVMe-oF) – the protocol tailored to squeeze maximum NVMe performance at minimum latency and CPU utilization.
Performance-hungry applications don’t take the full advantage of NVMe storage. This happens because NVMe devices are connected to CPU via legacy iSCSI and iSER protocols that work good for slow spindle drives but cannot ensure the decent performance for NVMes. Thus, whatever the potential of your NVMe storage is, your applications still don’t get the good part of its I/O. This means that even with superfast drives and high-end CPUs on board, your applications won’t benefit that much from them.
Another thing that IOPS-hungry applications need is close-to-zero latency. iSCSI and iSER protocols can’t ensure such latency simply by their design. They have a single short (64) command queue that decreases network throughput.
StarWind NVM-oF Target delivers your applications the maximum IOPS that high-performing NVMe storage could ever provide. This protocol has been purpose-designed for fast communication with the NVMe storage over high-speed networks with a reasonable portion of server CPU power involved. So, both NVMe devices and CPUs are utilized efficiently bringing you high ROI.
Latency-dependent applications get a consistent microsecond-scale latency with StarWind NVMe-oF Target. The protocol features multiple parallel command queues, 64 thousand commands each. Moreover, once combined with high-speed RDMA, NVMe-oF Target enables your applications to achieve almost the instant request response. Speaking of numbers, this protocol enables to build almost 8 times more efficient NVMe storage than if you use any SCSI-derived protocol.
With StarWind NVMe-oF Target, your mission-critical applications get the fastest storage on the market. No bottlenecks in protocols – both CPU and NVMe storage power are unleashed. As a result, you get high ROI and can stop worrying about storage performance.