Shulou(Shulou.com)06/01 Report--

In this issue, Xiaobian will bring you about how NVMe over Fabrics helps Ethernet unify the data center network. The article is rich in content and analyzes and narrates from a professional perspective. After reading this article, I hope you can gain something.

Yesterday we talked about iSER. From the final survey results, many netizens think that iSER cannot meet the requirements of all-flash data centers in the future because it still needs to use SCSI protocol. They all felt that NVMe over Fabrics (NVMeF) could be the biggest contributor to helping Ethernet unify the data center network.

We don't have much time today. Let's take a brief look. Is it possible?

Let's start with the history of NVMeF. NVMe has become a successful standard because it does not use SCSI, the overhead of the protocol is greatly reduced, and it is perfect for flash memory. However, due to the scalability and management of PCIe, it has not yet become a mature fabric for data centers. The goal of NVMeF is to run NVMe using existing mature fabrics.

Without PCIe, NVMe needs to perform protocol conversion with SCSI based protocols to exchange data, which increases latency and diminishes the advantages of NVMe.

Therefore, NVMeF is used to solve this problem, so that NVMe devices can communicate remotely as if they were local, and the added latency of remote communication is controlled within 10us.

In addition to PCIe Fabric, NVMeF can also utilize all fabrics that support RDMA, and even support FC, Intel OmniPath, etc. in the future. Don't think about NVMe and think you have to use PCIe Fabric, it's not, after all RDMA Fabric is more mature in the data center.

Like iSER, its RDMA Fabric can choose Infiniband, iWARP, RoCEv1 (no routing), RoCEv2 and other transport protocols. That said, NVMeF can also take advantage of mature Ethernet technology, preferably DCB (Lossless Ethernet).

NVMeF's long-term goal is to be a transport-independent specification that can leverage new data center Fabric technologies such as Intel OmniPath.

At present, some manufacturers have done some demos. In addition to Intel, Mellanox is actually very enthusiastic. Anyway, it mainly sells the underlying RDMA Fabric, which runs NVMe or iSER.

According to the plan, NVMeF 1.0 specification should be released by the end of 2015. Just now, Brother Watermelon checked and found that the specification had not been released yet. It seemed that it would have to be postponed. Because foreigners are on holiday, these two days should not be published.

In fact, NVMeF has been warming up for a long time, and all-flash array vendor Pure Storage announced early that it would use this technology in the future.

Of course, the hottest all-flash array in the market, EMC's DSSD is also said to use NVMeF technology, and has been demonstrated at VMworld 2015. It is said that the product will be officially launched at the end of 2015, and now it seems to be postponed.

From the current storage underlying transmission technology, there should be the following 6 kinds of Fabric, from the speed point of view, SAS and FC lag behind. PCIe is still early as a data center Fabric due to insufficient scalability. The most mature ones are Ethernet and Infiniband, but OmniPath has the most speed advantage.

Ethernet is the most popular, with multiple speed options, wide transmission range and highest cost performance. Although DCB is still relatively expensive, iSER and NVMe can also run on ordinary Ethernet. FCoE must be supported by DCB, so it is out of the market, but iSER and NVMe are on top.

The above is how NVMe over Fabrics shared by Xiaobian helps Ethernet unify the data center network. If there are similar doubts, please refer to the above analysis for understanding. If you want to know more about it, please pay attention to the industry information channel.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.