Nimble arrays use the unique, CPU driven, Cache Accelerated Sequential Layout (CASL) storage architecture to deliver performance benefits. Let’s break down how this works for read and write operations.
Each controller has built in NVRAM (Non-Volative Random Access Memory). NVRAM retains its information when the array loses power. Initiators connected to the storage array send writes which are placed into the NVRAM of the active controller, the contents of the NVRAM are mirrored to the standby controller. Once this is complete the write is acknowledged. Different applications send writes in different block sizes, these are all written to NVRAM whether the array setup is all-flash or hybrid, providing much faster response times and low latency. This is the reason why even Nimble hybrid arrays are able to outperform other all-flash setups.
Next the contents of NVRAM are copied into DRAM (Dynamic Random Access Memory). Different block sizes are compressed to form a single sequential stripe which is written to disk (4.5MB stripes for hybrid, 10MB stripes for all-flash) with triple-parity RAID protection. Hybrid arrays place hot accessible data or volumes configured for cache pinning onto SSD, other data is indexed and written to spinning disk. All-flash arrays include inline and variable block deduplication, providing datadedupe for different block sizes before writing data to disk, but after committing to NVRAM. Nimble deduplication uses a fingerprint index secured by SHA2-256 encryption. You can read more about inline deduplication and variable block deduplication here.
As well as inline compression CASL provides higher levels of capacity utilisation by by implenting thin provisioning and sequential stripe writes to deliver fast, consistent, and efficient writes. Over time as stale blocks are left from the deletion of data and snapshots; background sweeping processes remove these blocks to reclaim free space and create new stripe writes, ensuring that full stripe writes are always used and the file system avoids fragmentation.
Read operations also benefit from performance delivered by the methods outlined above. Furthermore CASL uses an intelligent algorithm for data placement. New writes are committed next to each other on disk, meaning writes can often be retrieved together during a read operation. Data that is written and read straight away benefits from read operations direct from NVRAM. If the data is not present in NVRAM or DRAM the cache (if applicable) is checked, all data is indexed to SSD for quick look up. In hybrid solutions data can be manually pinned to the cache, in conjunction with Nimbles adaptive cache design that is able to adapt to changing workloads.
In hybrid solutions CASL populates the cache with recently written hot data and associated metadata. As time goes on data in the cache that has not been accessed is downgraded until it reaches the point where it becomes cold data; is evicted from the cache, and written to spinning disk. This makes full use of both the speed of SSD and the capacity and cost savings of HDD.
Resources – Nimble Technology Overview.