So maybe not 33, but there are so many different ways to approach flash in a data center environment. The correct approach depends on multiple factors, and understanding the differences between the varying architectures is critical. It is also important to remember the soft criteria. Your solution should simplify administration, reduce risk of data loss, improve application performance, and in most cases provide better economics.
Starbucks recently put up new signage as part of the push around their new Flat White beverage to show a drinks tendency towards milk, vs. espresso. Flash can be thought of in a similar manner. Instead of milk and espresso, you can think of it as server and array.
Server Side Flash:
One one side of the spectrum there are a number of different iterations of flash at the server side. You can introduce a software based cache like Pernix that takes advantage of server side SSD drives to provide better read and write performance. You can also leverage software like EMC ScaleIO that aggregates these drive resources across servers and provide a clustered pool of storage. On the hardware side, you can of course leverage SSD drives for your OS volumes, or your Hypervisor boot images, but you can also introduce Server-Side cards like EMC XtremSF for your high I/O, low latency applications like database servers. Or, working in conjunction with an array on the back end and with software like EMC XtremCache you can provide huge gains in performance by tiering data access across the array and all the way out to the server.
Flash Cache:
One advantage in arrays of all types are the large pools of DRAM that they provide to cache data objects for subsequent reads. This same technology can be extended down to SSD drives to provide even larger pools of very responsive cache for applications to take advantage of. Hybrid arrays are focused around this concept and can use automated tiering software suites to simplify the placement of data objects across the array.
All Flash Arrays:
At the other end of the spectrum is the all flash array like XtremIO. These arrays are designed from the ground up with new architecture to take advantage of the performance and unique characteristics of SSD drives. Not having a defined cache and storage layer that is separate allows the entire device to achieve the best performance profile available. The simplified disk arrangement and pooling provides the simplest platform to administer. And advanced data services provide redundancy across the array delivering the best data protection available.
So how do you take your flash? It’s clear that it is the future of data center storage, and the spinning disk will soon go the way of the tape drive. It will never go away completely, but it’s role will change. No longer will we manipulate and coax performance out of technology built with spinning and moving parts.
Digital IRL 