This week at Flash Memory Summit 2018 we got to see the next iteration of Toshiba KumoScale. We first covered Toshiba KumoScale in our piece New Data Center Toshiba SSDs and KumoScale NVMeoF Solutions during Open Compute Summit earlier this year. Toshiba is seizing the trend toward NVMe over Fabrics (NVMeoF) to move up the storage stack. As NVMeoF disaggregates storage and compute, opportunities for market shifts occur.
Toshiba KumoScale Adds TCP for NVMeoF
When we first saw Toshiba KumoScale it was based on RDMA networking. RDMA is likely the future, however, standard TCP is widely deployed. From an interoperability perspective, TCP is still easier to use. At Flash Memory Summit 2018 we saw the solution running TCP with relatively little loss enabled by software and the Marvell FastLinQ3 100GbE NICs. One of the keys to this solution is the TCP offload enabled by the Marvell FastLinQ3 hardware. For clients, they do not need new RDMA hardware which means that Toshiba can deliver faster performance to legacy servers.
Excerpt from the Toshiba KumoScale Press Release:
Here are two key paragraphs from the Toshiba KumoScale TCP accelerated NVMeoF press release:
First introduced earlier this year, Toshiba’s KumoScale software enables the use of NVMe-oF to make flash storage accessible over a data center network. KumoScale provides a simple, flexible abstraction of physical disks into a pool of block storage, all while preserving the high performance of direct-attached NVMe SSDs. While the benefits of NVMe-oF are well known, cloud adoption can be impeded by the expense and complexity of the high-performance, low latency networking needed to deploy it. NVMe-oF on TCP enables an unprecedented price/performance balance.
TCP, the most widely-used networking protocol in today’s data centers, has not been a true option to transport NVMe-oF – until now. By leveraging the trusted TCP protocol, KumoScale enables NVMe-oF to be deployed over existing Ethernet networks and opens the door for the broader deployment of NVMe-oF networks and clouds.
