At StorONE we recently launched the All-Flash Array.next. While our use of Optane gets all the headlines, how we manage QLC is a real technical achievement that can benefit data centers of all sizes.
QLC offers enterprises excellent read performance, good write performance, but low endurance. Low endurance can be deceiving, though. The way a storage solution writes data to the QLC SSDs can improve endurance. Large sequential writes are less disruptive to the SSD than random writes and reads have almost no impact on endurance.
The capacity of the QLC also helps. The higher the capacity per drive, the longer it takes for each drive to reach the notorious drive writes per day threshold. If the storage solution is provided with enough high capacity QLC drives and can write to those drives sequentially (with a version of data that won’t change much) the enterprise will experience very high QLC life expectancy.
QLC Needs A Write Shock Absorber
The challenge is dealing with all the “ifs.” In the enterprise, QLC needs to partner with another technology to make sense. It requires a tier of storage to act as a write shock absorber so the data can come to rest and be sequentialized. The problem for storage vendors is finding the right technology to use as the tier above QLC.
The ideal partner for QLC is Intel’s Optane technology, as we’ve seen in the StorONE All-Flash Array.next. Optane shares as many similarities with DRAM as it does QLC. It has very high write performance, 63X greater than flash, and very high durability, 34X greater than flash, but like flash it is nonvolatile, meaning data written to it can survive a power outage. In some ways, it is the exact opposite of QLC: low in capacity, high in cost, and high in performance. When combined together correctly, Optane can significantly lower storage infrastructure costs.
How to Use Optane with QLC
The challenge for storage system builders is how exactly to use Optane as the shock absorber for QLC. The storage system must take full advantage of Optane’s write performance. It is also critical that the tier is large enough to enable the data to “settle” so it can be sequentially written to the QLC tier while also having enough time to reach a near-permanent state.
The Cache Option
Using Optane as a cache won’t achieve the goals of taking advantage of Optane’s high write performance and provide time for data to settle so the storage software can write sequentially to QLC. When storage vendors use Optane as a cache, that likely means one Optane drive. A single drive means there isn’t enough capacity to enable the data to settle. It also means the customer needs to be concerned about a lack of data protection. What happens if one drive fails before it has flushed data to flash even though it has acknowledged the application’s write?
Vendors and their customers can’t afford the risk of data loss in a single drive cache configuration. That means that most prudent vendors will force their customers to use the Optane cache in a read-only setting. While using it as read-only does better protect data, it also eliminates the primary advantage of Optane; write performance.
Getting full benefit from the Optane investments means that the storage software must be able to extract the maximum performance of each Optane drive, it also means it must use Optane as storage, not as cache. Using Optane as storage also means that the software must apply protection to the Optane tier, so if a single QLC drive fails, data is still accessible. Optane as storage enables the application to benefit from the incredible writing performance and, easily as important, low latency that Optane provides.
Once the environment is receiving the full benefit of the Optane technology, the final step is to move data to the QLC tier after it has had a chance to settle. The Optane tier needs to be large enough that the data has time to come to rest. However, because the read performance of QLC, especially when written to sequentially, is quite good, the solution does not need to focus on the access date, only on a modified date, which makes the Optane tier capacity even more efficient.
Many vendors try to improve performance by leveraging RAM as a cache. Their practice of using RAM as a cache leads to complications, especially in clustered environments. Given the write performance of Optane, one needs to question the logic of also using RAM and its potential complications. Instead, if the software is capable, writing directly to the Optane tier makes more sense. It enables the user to enjoy all the benefits of Optane performance without the flaws of a write cache.
The net impact of combining Optane and QLC? A higher-performing, lower-cost solution that matches what enterprises get today from all-flash arrays. In order to deliver on this promise the vendor must first deliver close to the potential of each Optane drive installed in the system. Most organizations can’t justify 24 Optane drives delivering only 15% of the potential performance. The storage vendor must provide closer to 85% of the per drive performance, especially in the case of Optane.
The vendor must also have the ability to tier data—only a few do. It must also have the ability to execute the tiering between Optane and QLC. QLC is different from hard disk drives, so tiering to it requires specific optimizations. Most vendors don’t support tiering at all, and those that do only support tiering from TLC to HDD.
[…] once we transfer the data to QLC, it can also be read directly from that tier. We also manage QLC so that it is truly enterprise class. The result is high performance in almost any state, and no […]