Architecture Analysis for storage devices

The disk array is the most common storage device, from the simplest single control disk array to the complex matrix switching structure of the multiple-control array, can meet the different levels of application requirements. The single control disk array composition is simpler, its core is a RAID controller, in a cabinet can configure more than 10 disks, through the RAID controller implementation raid function. This disk array typically supports the ability to manage logical volumes, can pool RAID devices, and then export different LUN devices based on application requirements. In this piece of the export Protocol, iSCSI protocols are usually used to meet the needs of low-end ipsan applications.

The biggest advantage of a single-controlled disk array is low cost, but poor reliability. The persistent storage service cannot be provided when there is a problem with the RAID controller. Therefore, a dual-control disk array is developed on the basis of single control. Dual-Control disk array can be divided into active-standby,active-active work mode, the former mode of work is relatively simple, Standby controller needs to monitor the active controller, once the active controller has found a problem, You can take over the role of this controller and continue to provide storage services to users. Active-standby enhances system availability, but does not lead to increased throughput. Active-active not only enhances system availability, but also improves storage-system performance. On the implementation, the latter is a lot more complex, not only to solve the problem of storage sharing, but also requires two controllers for memory synchronization. In order to solve the problem of efficient memory synchronization between controllers, Pci-express technology can be used.

Typically, the Ipsan disk array We see is a box with a limited storage capacity that requires a higher capacity and performance of storage devices for midrange users. For this reason, a storage device with a stack of enclosures is present, and the structure of the device is shown in the following illustration:

The head is connected to all the enclosure by multipath way, and all the enclosure is connected by means of a circular connection. This system is relatively simple to connect, the problem is that the number of enclosures will affect the system IO performance. In the old system, the backend is interconnected via FC bus, and in many existing systems, it is connected by SAS bus. This storage system has a separate control device, so it has a strong processing power, not only to achieve raid in a single disk device, volume management, but also to achieve storage virtualization, snapshots and file systems and other complex functions. However, a single controller can easily introduce a single point of failure, so many storage device vendors provide more than one device for dual-control node architectures. For example, EMC's VNX storage device uses a dual-control architecture such as:

The dual-control multi-cabinet method can meet the demand of capacity, but it is difficult to meet the needs of some high-performance applications. Expansion of the disk cabinet allows for capacity expansion, but the system's export interface and the back-end enclosure serial structure can be the system's IO performance bottleneck. For this reason, the architecture of dual-control multiple cabinets is difficult to apply in high-end storage applications.

In order to obtain high performance, only the back-end of the enclosure serial network is broken, and the use of a multiple-controller approach to achieve overall system performance. Cc-numa structure is an important architecture in high performance computing, and it is also used in storage field. The famous Symetrix DMX and Vmax are typical cc-numa structures that can be described as follows:

From the structure, we can see that the essence of the architecture is a dual-control or a single device cluster, and through the cluster interconnection bus shared memory, composed of NUMA system, in the software to see a unified memory, simplifying the software implementation. In Vmax equipment, two controllers constitute an HA pair, shared disk Shing to avoid the single point of failure in the above schematic. All controllers are interconnected through the RapidIO bus so that each controller can access local memory quickly, and remote memory can be accessed via RapidIO. Vmax This interconnection bus is called Matrix interface, the biggest disadvantage of this interconnection is that scalability is poor, there is no way to extend enough nodes. The diagram of the memory structure of the two control nodes in Vmax is as follows:

This column more highlights: http://www.bianceng.cnhttp://www.bianceng.cn/database/storage/