All-Flash Array (AFA) vendors position their hardware as the cure-all for database storage but administrators know that databases require more than all-flash. The most important requirement of any database storage system is that it must maintain data integrity, an area where most all-flash arrays fail miserably. Another database requirement is to protect data from eventual failures of media and storage hardware. Most AFAs fair better in this area but make you compromise on performance or flexibility. A final requirement is to help database teams better manage storage. Database environments are notorious for requiring multiple copies of databases for test/dev, QA and reporting. Most AFAs simply can’t help in this area, so customers are forced to buy additional storage and even additional software to help them manage the copies.
Databases Require More Data Integrity than All-Flash Provides
Most AFAs use RAM as a cache for both reads and writes. Most database architects design their database servers with massive amounts of internal RAM for read caching. Caching reads again on the AFA provides no additional value and a double cache adds to latency. Most AFAs also provide a write cache on the system, since one of the weaknesses of flash is its poor write performance. While this may provide some performance benefit, the technique does put data at risk because write acknowledgment occurs before the data is on media. If there is a failure while data is in a transient state, there will be data loss. A loss that may be unidentifiable since the database itself is working under the assumption that the write completed successfully.
StorONE’s All-Flash Array.next (AFAn) provides high performance without the need of a RAM cache for either reads or writes. All write IO is sent, and confirmed, to its Intel Optane tier prior to acknowledging write completion to the database. Because the AFAn uses Optane as storage and not as a cache, the media is protected from drive failure so storing data on this tier until it becomes less ephemeral is perfectly safe. Read “Caching vs Tiering” to learn more about the differences between the two technologies.
Databases Require Better RAID
Modern databases are too critical to be dependent on outdated RAID algorithms. Database architects require better protection from failure than the one or two drive redundancy that RAID5 and RAID6 provide. At the same time they don’t want to go to the extreme of a full mirror. The answer to the RAID dilemma can be found in erasure coding. It provides full flexibility in drive redundancy levels, supports mixed drive sizes within the same group and does not require dedicated hot spares. It also provides much more rapid, parallel rebuilds than RAID. The problem with erasure coding is its typical performance is not acceptable for production workloads.
StorONE’s AFAn uses vRAID, an advance form of erasure coding, to deliver all the capabilities of erasure coding with no impact on performance. In our database testing benchmarks, vRAID is active in all tests, yet the AFAn still achieved over one million IOPS.
Databases Require Better than Snapshots
Most AFAs have snapshots, which should enable database architects to address the needs of multiple copies of their databases. They should be able to take a snapshot and feed that to test/dev, QA, reporting, and backup processes. The reality is most AFAs can only maintain a small number of snapshots per system. A database environment can easily require hundreds of snapshots of the database at different intervals with indefinite retention needs. A typical AFA would suffer serious performance impacts if asked to replace stand-alone database copies with snapshot copies.
StorONE’s AFAn can literally deliver an unlimited number of snapshots and retain those snapshots indefinitely, without impacting performance. StorONE’s S1:Snap technology enables database architects to take snapshots as frequently as every three minutes, and retain those snapshots for years. The snapshot versions of databases can be restored as read/write volumes and then used by any of the various processes listed above.
Databases Require Better Performance
Modern databases, in relation to the storage system, contain a much higher percentage of write IO than they did a few years ago. Today, IT purchases large quantities of RAM and places it inside the database servers so that the majority of the read IO is serviced from the server’s RAM. The use of large amounts of RAM means that the majority of the server IO going to the storage infrastructure is write IO, not read IO. The storage infrastructure needs to support very high write IO performance, a weak spot for traditional AFAs.
The use of large amounts of RAM to overcome poor network performance is IT’s attempt to address another storage challenge, the inefficient read performance of all-flash arrays and their poor utilization of network bandwidth. Using large amounts of RAM is an effective workaround but it is also expensive.
The AFAn’s use of Intel Optane as a storage tier is ideal for RAM-heavy database servers. Optane provides a significant write performance advantage over flash so it can respond quickly to inbound write commands. The AFAn can deliver over 300k sustained write IOPS from its Optane tier while maintaining sub-millisecond latency.
Databases Require a Better Storage ROI
Data Integrity is critical and so is performance but as some point the storage system needs to justify its investment. AFAs do very little to reduce the cost of database storage infrastructure; in most cases they increase it. To deliver anywhere close to acceptable performance the AFA requires dozens of flash drives. The problem is most databases don’t need the capacity that these dozens of drives deliver. The database architects find themselves dealing with the same problems they used to deal with during the days of hard drives.
The key to a better database storage ROI is to extract the full potential of the media the storage system uses. Maximizing potential becomes even more important when using Intel Optane technology. The AFAn is able to deliver over 1 million read IOPS and 300K write IOPS from just four Optane drives. The AFAn further improves ROI by intelligently using QLC for less ephemeral data and older snapshots to further drive down the cost of capacity. The QLC tier however provides excellent read performance and as a result delivers automated tiering without performance impact.
With AFAn you get the best database storage ROI in the industry while improving data integrity and data protection.
Next Steps
To learn more about what is next in Database Storage, join our live webinar tomorrow at 1:00pm ET / 10:00am PT:
To learn more about how StorONE can revolutionize your database storage infrastructure check out our StorONE for Databases page
You can also register to get a copy of our AFAn performance report:
