Resource of SSD disks for servers

The SSD resource on NAND flash memory is determined by the number of write and erase cycles that the drive can withstand in its life cycle. Regardless of the type of memory, the name of the manufacturer and the SSD controller, such internal processes as garbage collection and write amplification are involved in the degradation of NAND cells. The most effective way to extend the service life of solid-state drives is to increase the area of ​​service cells, overprovisioning.

Garbage collection ( GC)

NAND memory is divided into blocks, which are divided into pages. It is impossible to directly write data to the flash memory at the physical address. The SSD controller writes updated data to free pages, marking the output as invalid. The address translator redirects the logical addresses of updates to new physical addresses. It is not possible to delete a page with data that has lost relevance (garbage), only a whole block. In order to free, make available to the OS the memory space under the invalid data, it is necessary to first move the valid data of the neighboring pages of the freed block to pages in another block. This is GC, the process of collecting valid data into working blocks followed by erasing garbage ones.

Write amplification ( WA)

When there are few free blocks as the SSD fills up, each write operation leads to a series of data movements by the controller - to collect them into consistent blocks and carve out a reserve of space by removing garbage. As a result, the volume of writing to the flash memory cells (including copying of valid data during GC) can be many times greater than the volume of the host's write package. The gain of the WA record is the ratio of these volumes. It depends on the workload profile. As a rule, random writing of short fragments multiplies records much more than sequential writing of large ones: random access data is more fragmented, and more operations are required to collect scattered small garbage.

Overprovisioning ( OP)

For the freedom of maneuver of the "garbage collector", an OP area is allocated from the general SSD memory space, accessible to the SSD controller, but inaccessible to the host. This reserve of NAND service cells is used as a staging area for GC data migration, wear leveling, and dead cell replacement. The SSD manufacturer assigns the OP size itself, reducing the capacity of the SSD user. There are generally accepted denominations (like 256GB or 240GB), but there can be any:

This explains the variety of SSDs, when drives on the same controller come from the hands of the same vendor, but with different passport indicators.

The logic of the OP marking is easy to imagine using the example of the game of fives. If there was not one free cell on the field, but more (for example, a 5x5 box instead of 4x4), the same dice from 1 to 15 would be arranged in fewer moves. So it is with SSD recording enhancement: the larger the service memory reserve is, the smaller it is.

Wear cells

The user is not interested in the mechanics of the process, but rather in the prediction of SSD longevity - in terms of volume of overwrites per day (DWPD) or, more correctly, total volume of overwrites during the life cycle (TBW). The average statistical resistance to wear (endurance), which we see in SSD passports, is calculated according to JEDEC standards. For server drives – on corporate templates with random access loads.

TBW prediction increases with increasing OP, but non-linearly. From a certain point, the addition of OP does not lead to a noticeable increase in the resource, rather it reduces the bandwidth of the SSD. The appropriate size of the reserve area for actual use is chosen by evaluating the nature and intensity of loads, including in similar tasks on old servers.

Productivity SSD

Performance of random recording is maximum on a fresh drive. As it fills up, data becomes more and more fragmented, and the SSD controller spends more time on garbage collection, a major factor in performance degradation. Increasing OP reduces WA, the controller frees up pages for writing faster and quickly processes host requests when the SSD is close to being saturated with write operations.

Everything is in the hands of the user

OP size is not a constant property of SSD, an insurmountable obstacle for resource management and performance. All major vendors have utilities that allow you to control various SSD parameters. Including changing the size of the area of ​​service cells of server drives - just as it is done at the factory.

• Overprovisioning is the practice of reducing the useful capacity of an SSD, increasing its resource and performance. Resource overallocation is set by the Host Protected Areca (HPA) parameter.
• This feature is only supported by Kingston Data Center series SSDs
• First delete all SSD partitions. Clean the SSD with diskpart. Safe erase is recommended for best performance.
• Enter the desired capacity in gigabytes (GB) and the last four digits of the SSD serial number. After modifying the HPA and rebooting, the system recognizes the new disk capacity.

Know the measure

Software management tools for SSD resources and performance increase the user's freedom of action - a valuable quality in times of poor availability of different denominations of drives. Instead of waiting a long time for the delivery of exotic Write Intensive SSDs (5-10 DWPD), it is enough to buy affordable Mixed Use (1-3 DWPD) and mark them for a smaller capacity.

The optimal OP size depends on the application. For workloads with intensive random access writing, it is useful to sacrifice some capacity for increased lifetime and performance gains. Where reading prevails, extravagance is inappropriate.

How can we help?

The Server Solutions company sells Dell PowerEdge R760 and Dell PowerEdge R760xs servers throughout Ukraine, among our customers are small, medium and large businesses. If you or your company needs advice and the purchase of high-quality server equipment, then you should contact us.