Device Affinity

What is Device Affinity?

Device Affinity is a term to describe the synergy between our SSD software and hardware which delivers superior performance, endurance and, in some cases, more capabilities to the end user. Device Affinity manifests itself at multiple layers of the “stack”.

Device Affinity differentiates our products.

It can be seen in our CellCare® technology for adaptive flash access, signal processing, data management algorithms as well as error checking and correction (ECC). This gives us the ability to offer 5-year warranties on our SSDs even as Flash geometries shrink and the raw endurance of Flash decreases.

Device Affinity is also seen in our SSD device drivers where we closely integrate so that extra steps and extra media read/write activity is eliminated. This can improve performance and endurance by a factor of 2.

Finally, Device Affinity also occurs with our advanced software. It can also deliver benefits to software, e.g. databases, analytics tools, or data-intensive applications, while accessing data stored on our devices.

Let’s take the example of HGST Virident ClusterCache.

This is especially important when considering 3rd party software solutions that lack device-level integration with flash media. The chart below illustrates the current paradigm (left) of caches without device affinity, like SanDisk FlashSoft, compared to an integrated caching with Device Affinity like HGST Virident ClusterCache (right).

ViridentClusterCache
Current paradigm of caches without device affinity, compared to an integrated caching with Device Affinity like HGST Virident ClusterCache

As you can see, the traditional approach requires intermediate artefacts due to a “layered” approach which disconnects the layers, thus losing valuable information. For example, the traditional approach on the left shows the need for a cache index for recovery, effectively doubling the amount of data written to Flash (amplification). In addition, for garbage collection, the traditional approach has 3x amplification of traffic. When Device Affinity comes into play, the resulting 6x efficiency boost can be leveraged to deliver higher performance and endurance in a flexible manner.

Example of raw cache writes per transaction (lower is better).
Source: MySQL 5.6, TPCC-like workload, 2500W dataset, 65GB buffer pool, 64 clients, 8 hour runti

 

When we map this to a real-world workload like a MySQL database, you can see a dramatic difference. Highlighted left above is an example of raw cache writes per transaction (lower is better). On the right we show effective life of Flash media using ClusterCache compared to commercial alternatives. Our technology consistently lowers the numbers of writes producing a longer-lasting Flash.

Web2.0 is a del.ico.us schema, 90% read, 10% write workload, 250GB dataset, 65GB buffer pool. Reporting the INNODB TPM (transactions per minute)
Web2.0 is a del.ico.us schema, 90% read, 10% write workload, 250GB dataset, 65GB buffer pool. Reporting the INNODB TPM (transactions per minute)

 

Compared to commercial caching applications, HGST Virident ClusterCache shows a higher number of queries per minute illustrating how our integrated Device Affinity approach delivers superior performance.

HGST ClusterCache considers both performance and longevity into a single integrated solution with Device Affinity. Alternative approaches use commercial software and generic Flash cards. Due to the complexity of this integration compared to the HGST approach, it is clear that FlashMAX with ClusterCache is a lower risk approach with proven results.

Hopefully, this blog has given you a quick introduction to HGST’s “Device Affinity” and how it makes a difference in our products and solutions. If you have questions or comments, please share it below or on any of our social channels.

About Walter Hinton

avatarWalter Hinton is Sr. Global Director for Marketing at Western Digital Corporation. He brings over 25 years of experience in storage, networking including roles in Product Management at McDATA, Chief Strategist for StorageTek and CTO of ManagedStorage International.