Zhengzhou server? Is it good to use SSD for the lease configuration of Zhengzhou VM? Can the server use SSD? For server host users, the read/write speed of the hard disk will directly affect the response time of the server. Therefore, many users hope that their servers can use solid state disks. However, SSDs are expensive and difficult for users to make a choice. How should users evaluate their needs for SSDs?

First, the reason for choosing flash memory is not always obvious

Suppose you have a matrix diagram showing specific application cases, such as virtual desktop infrastructure (VDI), server virtualization, big data analysis, etc. It's really great to identify the ideal flash option on one axis and on the other. But this is almost impossible, because there are too many variables in the same application environment, and these variables will affect each other.

Second, array based flash memory

If the controller of the storage array is always idle, it indicates that the storage system is waiting for disk drives (an inherent problem with storage media), so adding solid-state storage to the storage array is a useful solution. However, if the array design does not support flash memory, adding solid state storage may not be effective, because a drive bay full of solid state storage may turn the storage controller into a new bottleneck. Similarly, if the utilization rate of the controller is close to saturation, it is useless to put the solid-state storage disk into the storage system. If the network is not a problem, a better solution should be to invest in another storage system that supports solid-state storage disks, such as hybrid arrays or all flash memory arrays. However, if the network bandwidth is limited, or you cannot purchase another storage system, the other option is to install flash storage on the host server

Third, server flash

The flash memory on the host server side can be a solid state storage disk in the form of a drive (SATA or SAS interface), a PCIe flash card, or even a flash memory connected to the memory bus through a dual in-line memory module (DIMM). All of these methods make the performance of flash memory closer to the application processor, rather than the storage array connected to the network, thus providing better storage performance by reducing latency. According to past experience, solid state storage disk is the most economical of the three forms. PCIe cards provide better performance than solid state storage, but generally, the cost per GB is also higher. However, the new form of placing flash memory in memory now provides another way of low latency, which may open some new application modes.

Fourth, network transmission

Deploying solid-state storage on servers instead of network attached storage systems has some advantages, namely, reducing SAN network transmission. If the application can obtain data from the flash cache or flash layer of the server, it no longer needs to request data through the network. This reduces the workload of the shared storage array and frees up more resources to support other servers. Therefore, the reduction of network transmission makes server flash memory a better choice than buying another shared storage system.

Fifth, start with storage bottlenecks

Solid state storage usually solves application performance problems by improving the data processing speed of the server. In essence, there must be a bottleneck in a certain link of the storage infrastructure. Finding the bottleneck through analysis is the first step to determine the solution.

Sixth, whether it is layered

After the deployment location is determined, it is necessary to determine the selection of storage type (the actual use of solid-state storage). In addition to the full flash array, the key to flash implementation is to put the most appropriate data into flash before use, and keep this state in the background. Essentially, tiering technology creates a high-speed storage area for the most critical data sets and subsets (such as database indexes or change logs), and populates flash memory according to selected policies.

Seventh, the dispute between MLC and SLC: is it still so important

When flash memory entered this stage for the first time, a key purchase indicator was which flash technology was used. Single tier storage (SLC) is more reliable, faster, but also more expensive. Multi tier storage (MLC) has a short service life and slow performance, but the unit price per GB is much lower. Enterprise multi tier storage is somewhere in between.

Eighth, how much flash memory is enough

Hierarchical technology requires enough flash memory to store complete applications, or at least the most critical data sets, so it is relatively simple to determine the capacity required for this method. However, the capacity used by cache technology is difficult to estimate. It is good to start from the rule of thumb, but the actual environment test is more helpful to judge whether the flash memory capacity is sufficient and will not be wasted. An interesting example was given by a flash and cache software vendor. The customer is a large telecommunications enterprise that runs multiple super large data centers and supports multiple VMware clusters and hundreds of virtual machines.

Ninth, flash cache

Cache software is usually included in the characteristics of storage systems, which can maximize the capacity of flash memory in traditional storage arrays. If this function is available, it can play a great role, because it is completely transparent to users and usually requires little configuration work. Cache technology is also applicable to PCIe flash memory cards installed on the host server side.

Tenth, data growth, risk and high availability

There are other constraints that have nothing to do with performance in the Flash deployment decision process. One of them is that the bottleneck caused by the existing infrastructure needs to be solved by flash memory. The other is risk. The partial write cache mode may pose a risk before the data is safely written to the primary storage area. Consider specific flash memory