Storage Spaces Direct (S2D)

Microsoft and Mellanox set rules of the game in the hyper convergence market

Hyper convergent infrastructures (HCI) become one of the most perspective technologies for creation of private and hybrid clouds. The idea of HCI startups, before others offered the leading positions which realized a possibility of radical changes and in time to take in this segment of the market. However and the industry heavyweights who as usual are delayed on start, but having sufficient scientific and technical potential do everything possible to reach the position appropriate to them.

General situation is as follows: irrespective of under what brand this or that competitive HCI is made, it consists of four interconnected components – servers, DWH, network and virtualization of DWH for the purpose of creation of the general pools (Storage Pool). In this respect, perhaps, the only exception represents three-component approach to formation of HCI based on Intel Xeon Scalable with a chipset of Intel C620 and DWH of Optane SSD.

Gartner Magic Quadrant 2016

The companies providing functionality, namely two first of the listed above four components with rare exception act as the supplier of HCI. They can make servers and/or DWH, or can purchase them, practically without having own production, i.e. to be as speak, fabless.

As for two other components (virtualizatsionny software and network equipment) which it is possible to call backbone, a picture absolutely other here. In these segments there are several undisputed leaders and they set rules of the game.

The duet became the most popular and from the system point of view setting the standard de facto Microsoft and Mellanox. Microsoft provides the means for creation of pools of storage of Storage Spaces Direct (S2D) entering into Server 2016, and Mellanox is a complete stack of network technologies. Such universal vendors as HPE Dell-EMC, Supermicro and dataON specializing in HCI, for example, give to this combination the preference.

Example of use of products of Microsoft and Mellanox in HCI from dataON company

Technologies of Microsoft

Software-defined approach to DWH (Software-Defined Storage, SDS) and technology of virtualization of disk space for the first time appeared in Windows Server 2012 under the name of Storage Spaces. Windows Server 2016 included the updated and more successful Storage Spaces Direct (S2D) option. S2D is capable to collect in the general pools (repositories) and, as it appears from the name, to provide direct access to spaces on the disks SATA, SAS and NVMe placed in servers.

The essence of the concept of disk pools consists that separate physical disks of the server integrate in one or several pools based on which volumes with different performance levels and fault tolerances are created. In one pool there can be disks differing on the amount, performance and the interface of connection. Virtual disks serve as logical volumes. They can be:

• simple (simple) analogs of RAID-0, providing optimal use of capacity of disks. Being the most productive, they does not assume any fault tolerance;
• mirror (mirror) analogs of RAID-10 or 1E for an odd number of disks. Such virtual disk is protected from loss of one or two disks respectively;
• with control of parity (parity) analogs of RAID-5 and RAID-6 VD, allowing loss of one or two disks respectively.

S2D supports:

• two-story data storage (Storage tiers), i.e. placement of data on virtual disks of Storage Spaces on SSD and HDD. For this purpose during creation of volumes (virtual disks) the combination of volumes of SSD and HDD on each of tiers is specified, the file can be tied to a SSD tier for ensuring the guaranteed performance;
• cache for record (write-back cache). For compensation of performance by low HDD a part of SSD in a bullet is used for record caching, at the same time the rest of capacity of SSD can be used for level storage.

Quantity characteristics S2D:

• Up to 240 disks in one, pools can be a little.
• To 80 disks in a bullet, to 4 pools on a cluster.
• To 64 virtual disks in one pool.
• The general capacity of a pool — up to 480 Tb.

S2D is implemented in two options, on five or six-level models. In the drawing the five-level model including is shown below:

• Level of the servers connected according to the network protocol of the application layer of SMB (Server Message Block) serving for remote access to files, printers and other network resources and for inter-process communication. In the first edition it was called the Single file system of the Internet (Common Internet File System, CIFS) and was developed by IBM, Microsoft, Intel and 3Com in the 1980th years. SMB was implemented through NetBIOS (over NBF, IPX/SPX or NetBIOS over TCP/IP) and was used in MS-NET and LAN Manager networks for DOS and also in Windows for Workgroups.
• The bus Software Storage Bus covers all servers making a cluster. When using Software Storage Bus each server can "see" all disks on all servers in a cluster, providing connection in network with full-meshed topology.
• Pool of storage (Storage pool) — a set of physical disks. The disks differing on the amount, performance and the interface of connection can enter it.
• ReFS (Resilient file system) – as well as SMB, a thing not new. This fault-free file system, the to roots PS2 going to OS/2 and IBM.
• Local disk spaces (Storage Spaces) are virtual disk volumes.
• Virtual machines. This sixth level is present if hyper convergent architectural option is implemented. If are limited to convergent architecture or as it is called sometimes disaggregation, then five bottom levels are used. The choice between disaggregation and hyper convergence is defined by user requirements. The first option is more preferable to configurations smaller in terms of the scope of, and the second to larger.

Stack of Storage Spaces Direct technologies

Mellanox technologies

The program component of the architecture described above is the share of technologies of Microsoft, and directly data exchange provides Mellanox. The company locates a complete stack of technologies necessary for this purpose. Advantage of Mellanox is that the products offered by this company correspond to the recommendation of Microsoft concerning use of direct access in memory of RDMA at data transfer rate between cluster nodes not less than 10 GB/sec.

The local technology of direct memory access Direct Memory Access (DMA), i.e. address to memory without violation of work of CPU, is known long ago. It is widely used for connection of peripheral devices. But with the advent of high-speed network technologies at first an opportunity for remote direct access of RDMA (Remote Direct Memory Access), and then and RoCE (RDMA over Converged Ethernet), i.e. on the advanced network Converged Enhanced Ethernet (CEE) differing in higher capacity and smaller delays opened. Application of RDMA began with high-performance computing of HPC and extended to HCI.

The technology is implemented in network controllers of Mellanox, since ConnectX-3 Pro and further. The new family of ConnectX-4 network interface cards allows to simplify implementation of RoCE and makes technology available to the cloud systems and DWH.

ConnectX-4 Lx EN with capacity of 1/10/25/40/50 Gbps is designed for solving of tasks of the virtualized infrastructures and differs in performance, highest in the class, for different applications and the markets. In these adapters this equipment room isolation of input-output is combined with unique scalability and efficiency. ConnectX-4 Lx EN supports speeds of 1, 10, 25, 40 and 50 GbE, less than 1 microsecond and message transmission rate of 75 million packets per second provide a delay.

The extensive package of proposals including the Mellanox switches on the Open Ethernet platform consists of two SwitchX-2 and Spectrum families. The Open Ethernet family includes a various portfolio of switches with 1RU form factor, the number of ports from 12 to 64 and interfaces from 1 to 100 Gbps, the blockings allowing to design a specialized data processing center of any size with any value. At a line of Mellanox company there are high-performance systems of switches with architecture of Open Ethernet and interfaces of 10/25/40/50/56/100 Gbps which provide maximum efficiency of work of data processing centers, solutions for cloud computing, data storage, highly effective calculations and Web 2.0 in any scales.

Mellanox offers a wide choice of cables of different length, capacity and types of carriers. For physical switching of HCI the certified cables and Mellanox LinkX transceivers are used, they conform to all requirements of the IEEE standards and also passed 100% of tests for ensuring optimal integrity of a signal and end-to-end performance. The Mellanox cables are suitable for solutions based on Ethernet, InfiniBand, cloud infrastructure and Web 2.0 and also connections the server - the rack switch, the server DWH, the switch switch.

Stack of Mellanox technologies

Future HCI - Microsoft Azure Stack

Azure Stack is very powerful request of Microsoft to the place of the leading solution provider for hybrid clouds. The offered strategy assumes failure from amateur performance during creation of own platform and for private, and hybrid clouds. Instead ready HCI, delivered by either Microsoft, or a limited circle of partners of OEM are offered (HPE, Dell, Lenovo and some other). Such systems will be completely compatible to Microsoft Azure cloud working for RoCE. The equipment from Mellanox, including the network adapters ConnectX, the Spectrum switches, transceivers and cables (copper and optical) by LinkX will be used to networks in them.

Use of the checked HCI solutions provides seamless approval between a global cloud of Azure and a hybrid cloud of Azure Stack. And the fact that the hardware will arrive from Microsoft and its partners, guarantees all its necessary operational indicators.

Potential success of Azure Stack lets some observers speak about future changes in a gartnerovsky quadrant.

Possible changes in Gartner quadrant

196

Notes