Over the past five years, storage vendors have made some impressive improvements in I/O performance, achieved almost entirely through using lots of high-RPM spindles, lots of cache, and a generous helping of SSD drives for “Tier 1” data. Automatic migration of data to the appropriate tier has been a very hot topic, and has generally been accomplished with proprietary algorithms with varying degrees of success. The downside of all this improved performance is that it also comes with pretty hefty price tags.
All of this high-powered hardware was being held back by the simple fact that IO was still being done the same old way, small block random I/O, resulting in a great deal of disk activity for every transaction. Enter the Nimble Storage Cache Accelerated Sequential Layout (CASL) system. Nimble Storage’s unique architecture, CASL, is the enabling technology that makes converged storage possible.
CASL starts with an inline compression engine that compresses data in real time with no added latency, reducing the amount of data stored 2-4x. This is possible because of two key Nimble Storage advantages:
- Nimble Storage’s software was designed upfront to leverage the powerful multi-core CPUs used across the Nimble Storage array family for instant, high performance compression.
- Nimble Storage is the ONLY primary storage array that natively supports variable size blocks.
Next, CASL groups random blocks of data into larger segments before writing it to flash and disk. These fully sequential writes maximize the performance (and lifespan) of flash, which does not handle randomly written data effectively. Sequenced writes also maximize the performance of low RPM drives, which do not like to seek but can handle fast sequential streams. A copy of all active or “hot” data (and even “semi-active” data) is held in the large flash layer, enabling very fast random read performance. Inactive or “cold” data resides only on compressed high-capacity hard drives, further reducing costs. CASL’s intelligent index tracks hot data blocks in real time, adapting to hot spot and application workload changes within milliseconds, while even the most advanced tiered systems require a day or more.
So very simply, here are a couple of take-aways about the Nimble Storage “Special Sauce”:
- SSDs are used only as read cache, never as a Tier One storage layer. This takes advantage of the best quality of an SSD drive, fast reads, and does not rely on them for writes, which SSD’s do not perform very fast. And in using them as cache, the data on these SSD’s is not vulnerable to loss as it is already written to disk.
- Using cache accelerated sequential writes greatly reduces the number of disk I/O’s needed to write data and allows data to be written very efficiently in large sequential blocks. What this means is that the disk is no longer the bottleneck.