Hybrid Flash memory Array sourcing Guide

The hybrid flash array is superior in performance to an array based on traditional hard drives, with lower latency and cheaper prices than most pure flash arrays. However, their prices are still higher than those based on traditional hard disks, and their structures are much more complex. Given the price and complexity, what kind of applications and environments can derive the most benefit from a hybrid flash array?

First, let's look at the purpose of developing a Flash array: High data growth rates have had a huge impact on the data center. The problem can be analyzed from "3V", i.e. volume (volume), speed (velocity) and diversity (produced).

Many data centers are increasing at a rate of more than 100% a year. These excess data present a serious challenge to maintaining the storage capacity and performance of the data center and controlling procurement, maintenance, data center footprint and energy costs. Data growth is also fast. With data growing at an annual rate of more than 100% per cent, it is hard for corporate IT departments to keep up with this growth rate. With such data maintaining rapid growth, it is extremely difficult to develop strategic storage decisions.

Now add to the problem of data diversity, the variety of data it needs to store, including text and digital documents, audio and video, mobile data growth, and one of the biggest sources of data growth ever: machine-generated data. For example, a flight traveler produces several gigabytes of data on flight details, including his identity, security, seating, flight status, weather, airport, etc. By the tens of thousands of passengers flying each day, it is possible to figure out additional daily passenger data by multiplying them.

Data and IO

Input/output (IO) activity occurs whenever data is moved between servers and stored. Fast-growing volumes of data, faster performance, and various data types increase the need for the number and speed of IO for the network and storage systems.

In a server virtualization environment, this problem is even more serious. Virtualization technology is appropriate for managing server and application growth issues, but virtualized servers generate more IO activity with a variety of storage resources. Advances in technology usually increase the number of IO activities generated by the server, but storage devices cannot grow at the same rate due to mechanical constraints.

There are problems with IO bottlenecks at every level from the server to the storage device, including the controller, the optical fiber, and the physical server itself. But so far, the main bottleneck remains the storage device IO bottleneck due to the mechanical structure of traditional hard disks.

The application sends IO requests to the operating system, which then assigns IO resources to the storage system. The storage system chooses a suitable location on the disk to use for the new IO activity, and then moves the disk header to the location where the new IO activity occurs. At the same time, the disk begins to rotate under the head of the disk.

There is a lot of physical activity that needs to happen in that process, which can have a huge impact on storage performance. In addition, IO writes typically cause an IO activity to affect the results of multiple IO activities because it requires disconnected storage units. Because the search speed is slow, the delay will be lengthened, thus affecting the reading and writing data.

Traditional hard drives, however, are very limited in terms of improving performance speeds. Because they are mechanically structured, their iops can only reach 100-200, and the search time can only reach 3.6 milliseconds. What happens if you increase the IOPS of one hard drive to 200? It requires multiple technologies for stripping (striping), clustering (clustering) and parallelism (parallelizing), as well as multiple disk resources and storage controllers.

Even the most advanced storage technology today can only do this. The result is that there are serious bottleneck problems in storage devices because of the relationship between seek time and disk rotation delay. However, at any given point in time, thousands or even millions of IO requests arrive at the storage controller.

Can flash memory solve this problem?

Flash technology is now recognized as the best solution for IO bottlenecks. Flash technology has been used in a number of areas, including: Some environments will PCIe flash memory card as the server side of the cache technology to use, servers, networking or storage at every level can see the solid state of the hard disk.

The IO speed and latency of different flash arrays may vary greatly. Hybrid arrays combine the flash cache and flash levels with solid-state or traditional hard drives. These systems have a maximum performance speed of up to 100,000 IOPS, a delay of 3 to 5 milliseconds, and a guarantee that there is enough storage space available in the array. The data processing speed of pure flash array is usually faster, can be far more than 100,000 IOPS, the delay is usually less than 1 milliseconds, but its storage capacity is much less than the mixed array.

Price is also a problem. In general, the faster the flash system is, the more expensive it is. Flash prices are falling, and some array vendors claim that their pure flash arrays are cheaper than mixed arrays. However, not all flash memory is the same, and the speed of a pure flash array using a consumer-level flash memory may be slower than a mixed array. In any case, IT departments must consider the service level requirements before cost.

Critical applications that require enhanced storage processing also need to buy flash and pure flash arrays on the server side, but IT departments may not be able to buy enough flash memory to store the vast amounts of data generated by applications such as databases, OLTP, SharePoint, Exchange, and virtual servers and desktops.

There are also a number of key applications that require faster speeds than those based on traditional hard disks, but do not necessarily use a very expensive pure flash array. This is why the hybrid flash array exists. Moreover, even a hybrid flash array can increase the cost and complexity of the storage environment.

Is it worth it? The answer is: it depends.

Question 1: Is application ability affected by storage Io bottlenecks?

If the answer is no, continue using the array based on the traditional hard drive as much as possible. But if application can start to fall, consider using flash storage technology, including: Flash cache on the server side, virtualization HBAs, pure flash arrays, or hybrid arrays.

2: In addition to performance requirements, do you still face high storage capacity requirements?

The best application environment for Flash and pure flash arrays on the server side is a business-critical application with high requirements for IO and latency, where the data for business-critical applications needs to remain active for a long time and without the need to share storage resources with other applications. Some applications require more IO processing power than traditional hard disk arrays can provide, and they produce a large amount of data, in which case it is best to use a mixed array.

Question 3: are traditional arrays aging and need technology upgrades?

If you plan to replace an existing storage array, try to add more flash memory to your environment. In many cases, although the price will be higher, but it is worth doing so, because it can improve performance, reduce latency and provide more storage capacity.

Hybrid array Vendors

If you want to talk about mixed array vendors, you may have the feeling of confusing flowers with attractive eyes. It's important to think about what you need before you buy it. Many hybrid array vendors also offer more expensive, pure flash arrays, and vendors are certainly more willing to sell you a pure flash array. If all you need is low latency, high IOPS performance and high storage capacity, don't listen to them selling pure flash arrays.

IBM, Hewlett-Packard and EMC have both launched hybrid Flash array products. IBM's XIV Storage system provides a hybrid array model in which IBM SAN Volume Controller (SVC) centrally manages its flash devices. EMC offers pure flash and hybrid Flash array products under its Vmax, Vnx/vnxe, and Isilon brands. EMC's key advocacy is that its customers can easily upgrade from a mixed-disk array to a hybrid array to a pure flash array, which is certainly good if its customers want to keep using EMC's products.

HP offers an array of traditional hard disks, mixed arrays and pure flash arrays under its 3PAR brand. NetApp's FAs 3250 is a hybrid flash array that serves a target of more than 150,000 IOPS and is capable of receiving more than 5 milliseconds of delayed performance.

Oracle's ZFS Storage ZS 3 is a high-performance hybrid array product. It combines flash memory with DRAM to achieve higher IOPS performance and less than 1 milliseconds of read data latency, but Oracle gives it a much lower price than a pure flash array of comparable performance.

This area is by no means the only big-brand storage vendors. Nexsan's NST 5000 provides a flash layer that integrates high-performance flash cache with DRAM to achieve high-performance flash processing capabilities. Anglo Megatrends 3500i is divided into two configurations of pure flash or hybrid flash.

Tegile Zebi HA 2800 provides expansion slots for solid-state and traditional hard disks, so it can be converted from a pure flash array to a large-capacity hybrid array. Nimble's CS series array uses a multilevel unit flash cache and a cost-economical background SATA hard drive. Tintri provides a hybrid storage array with a traditional SATA hard drive, but its target market is a VMware virtualized environment that can send IO requests directly to the virtual machine.

Some PCIe Flash card vendors are also starting to take action. NexGen is a hybrid Flash server storage product launched by the flash card manufacturer Fusion-io. NexGen combines the PCIe Flash card with the flash memory and the SAS hard drive in the server. The Nutanix NX series integrates PCIe flash memory cards with Flash and Intel processors in virtualized environments. The 3000 series has increased storage capacity using SATA solid-state drives and traditional hard drives.

Hybrid flash arrays can balance performance and capacity requirements under relatively low cost conditions, and are therefore suitable for many environments. Today, there are a number of alternative hybrid flash array products available, offering enterprise users a variety of options so they can pick out the best products for their performance and capacity needs.