Tier 1 applications are often the most demanding applications in the data center. These are the applications that are directly tied to income, so there are good business reasons to want them at their best in every dimension. Application availability is one dimension and performance is another. Many Tier 1 applications are infrastructure intensive; they use a lot of resources to deliver value to the business. In particular, database applications depend on storage performance to provide great application performance. For consistently awesome performance, SimpliVity recently released all-flash configurations of its hyperconverged infrastructure platform. These new all-flash nodes will deliver peak and predictable performance for applications using large data sets.
SimpliVity’s main products use a hybrid storage configuration: RAM for write performance, predictive caching and flash for read performance and hard drives for bulk storage. Deduplication reduces the amount of every type of input/output (I/O) and makes each storage tier effectively larger. Processing of deduplication functions is offloaded to SimpliVity’s special-purpose PCIe card, the OmniStack Accelerator, which often boosts performance by eliminating I/O as a bottleneck. For most Tier 1 applications, this tiered configuration provides great storage and application performance.
But like all tiered storage, the performance tiers have a finite size. Performance is only optimal when the active application data fits in the performance tiers. Any I/O that cannot be satisfied by the performance tier will experience slower response as the data comes from the bulk tier. Different applications and even different databases have very different needs.
Most Tier 1 applications work with smaller amounts of data—a few tens of gigabytes—and are well suited to a hybrid storage architecture. But some applications have huge amounts of active data and so will always be accessing the bulk tier for a significant amount of I/O. Other applications will access the bulk tier less often, but will suffer much worse performance as they are very sensitive to latency. To address these applications, SimpliVity has launched all-flash models of its hyperconverged OmniStack platform.
By its physical nature, a hard disk is going to take longer to access data because it has rotational latency that a flash drive does not. Accessing flash is 10 to 100 times faster than accessing a hard disk. The new models replace the hard disks with Intel flash drives. An interesting design choice was using 1.6TB Intel solid-state drives (SSDs). This means there is no capacity compromise for all-flash compared with the hybrid models, many of which use 1TB hard disks. Another is that there is only a single tier of SSD, which means every I/O is the same speed—fast. In some ways, the all-flash nodes have done away with the bulk tier and massively expanded the performance tier in its place. The net result is that worst-case storage latency goes from tens of milliseconds to one or two milliseconds. The variation in storage latency goes away with an all-flash architecture. Every I/O is fast, even when the application reads its database from one end to the other.
SimpliVity’s hybrid storage configurations provide excellent performance at a great price-performance ratio. Adding all-flash configurations delivers great storage performance for Tier 1 intensive workloads, without a massive increase in cost.