Tech moves fast! Stay ahead of the curve with Techopedia!
Join nearly 200,000 subscribers who receive actionable tech insights from Techopedia.
A storage performance platform is a new architecture paradigm for virtualized environments. The premise of this new approach to storage is the use of server-side media controlled by hypervisor-integrated software able to provide distributed or clustered platforms with accelerated storage read and write I/O operations. These new platforms make use of flash- or RAM-based storage solutions within the server itself as opposed to slower disk-based ones implemented over a network fabric.
A storage performance platform makes use of two things that allow it to provide superior performance to current network-based storage solutions. First is the server-side computing resources such as RAM or flash, since they provide many advantages in terms of performance compared to network-based storage solutions. One advantage is that these resources are already very close to the applications themselves so they can truly leverage the performance of the media. Compared to current network-based storage solutions that may reside in a different room, building or geographic location entirely, server-side resources provide better performance because of shorter I/O paths. Shared storage implemented across a fabric cannot compete with the performance delivered by the RAM itself that sits right next to the CPU, or by flash storage that can be connected using high bandwidth interconnects. As an example, current PCIe flash storage devices can provide 250,000 IOPS actual with microsecond latency.
The second part of this new storage platform is a hypervisor-integrated software installed as kernel modules. This means that I/O access to the accelerated storage resources is done as part of the kernel and not as a separate filler driver inside another virtual machine or a virtual machine itself. In such a manner, scheduling and contention issues are avoided completely, which means that there is more throughput per I/O traffic. And since the software is a kernel module, the I/O path is ensured to be as short as possible from the application towards the requested data, which results in the best possible performance. Current shared storage that is accessed via a network fabric would never be able to compete with the throughput of something that is sitting right inside the host/server itself.