Flash’s first deployment in the data center was often as a cache implementation. A cache is a staging area that holds a small amount of the most recently accessed data. Caching software enables active data to be accessed from flash storage instead of hard drive storage. Even though flash pricing has come down, making all-flash a reality, more performance is still needed but it has to be something like a more advanced version of cache, IO Acceleration.
Caching vs. IO Acceleration
Caches are small and volatile; IO Acceleration is large and durable. Caches were designed when memory based storage was very expensive. If the organization could access 50% of its IO operations from cache that was considered effective but it still meant that 50% of the traffic had to cross the network and access data from hard disks.
In-memory caches are not durable, meaning that power loss means data loss. The potential for data loss meant they were not safe for write caching so all writes had to go to the hard disk tier. While writes make up less of the IO distribution of the typical environment, they are the slowest part of the IO chain. Flash writes data slower than reads and each write often has an additional set of writes associated with flash management and data protection.
The typical sizing of IO Acceleration, on the other hand, allows it to service 90% or more of all read requests and its design lets it work with a variety of storage devices including all-flash arrays and even the cloud. IO acceleration also has durability; protecting data outside of the system on which the acceleration software is installed so that power failure or even server failure does not result in the loss of data. Durability requires much more than just making a copy of each write, it also needs to be able to retrieve those copies intelligently in the event of a failure. The combination thought allows for both read and write acceleration.
The Future is the Cloud
A key next logical step for IO Acceleration solutions is to integrate with the cloud. Unlike cloud gateways, IO Acceleration solutions can bring flexibility to the location of the primary store. The problem with cloud gateways is that a cache miss means retrieving data from the cloud, which if it is located thousands of miles away, also means significant latency when accessing that data.
An IO Acceleration solution could deliver an alternative. If, for example, the IO Acceleration solution could work with a regional VMware cloud service provider and that provider is less than 100 miles away, then access to that provider’s storage involves much lower latency, rivaling on-premises network latency. The software could then add the capability to replicate data from the regional cloud service provider to a public cloud provider, creating a DR copy as well as enabling the organization to use cloud compute to run analytics or reporting on that data.
Introducing JetStream Software
JetStream Software has been delivering IO Acceleration software since 2010. The company started as FlashSoft, which SanDisk acquired in 2012. After Western Digital acquired SanDisk, they spun the company and the FlashSoft IP back out as a new company, JetStream Software. JetStream Software continued improving the core technology during this time and now the company is readying a migration/replication solution to complement the IO Acceleration software, now called JetStream Accelerate.
Today JetStream Accelerate can answer the problem of how to derive more performance from existing hybrid and all-flash investments by installing IO Acceleration software on the server and eliminating network IO as well as storage system software overhead. Accelerate combined with an NVMe Flash drive(s) installed in the server, will deliver unprecedented performance and be far less expensive than upgrading the storage system to NVMe.
In the future, JetStream plans to extend Accelerate to be able to provide an elastic storage solution. The organization will have active data on an on-premises acceleration appliance with their primary storage being located at a regional cloud provider. Again, if that provider is within 100 miles or so, the latency impact of retrieving data from the cloud is negligible. The solution will then incorporate JetStream Migrate to replicate data from the regional provider to a public provider like Amazon or another data center operated by the cloud provider. Once data is at the distant cloud, IT can leverage services like disaster recovery and analytics powered by cloud compute.
IO Acceleration is an ideal solution for organizations looking to get more performance out of their current hybrid and all-flash storage solutions. The concept of acceleration from a tier abstracted from the storage system can also extend to a cloud use case enabling the data center to get out of the storage architecture business altogether.