Home > Storage > PowerScale (Isilon) > Product Documentation > Storage (general) > PowerScale OneFS SmartFlash > Write caching – SmartCache and the coalescer
Writes to a cluster are placed into a protected write cache, or SmartCache, and immediately acknowledged to the client. When several small writes to the same disk are received, SmartCache coalesces them before flushing these writes to disk. This coalescing reduces the number of I/Os required to write the data to physical media.
Being a distributed file system, OneFS can make informed, strategic decisions about where to place written data. That is, on which nodes and their specific drives within the cluster to place data. OneFS decides upon placement using a process known as safe write. Safe write develops a write plan, writes the data to the local and remote cache, and then commits it to the journal. Writes are kept in the journal until the data is written to disk.
With Dell chassis-based hardware, each 4RU enclosure houses four nodes. Each compute node blade contains its own CPU, memory, PCIe slots, storage, and so on.
A node’s file system journal is protected against sudden power loss or hardware failure by the Power-fail Memory Persistence (PMP) feature. PMP automatically stores both the local journal and journal mirror across both nodes in a node pair, as illustrated in the following figure:
This journal de-staging process is known as “vaulting.” During vaulting, a dedicated battery in each node protects the journal until it has been safely written on both nodes in a node-pair. Gen6 hardware uses a 32 GB journal, providing ample space to coalesce and optimize writes.
With PMP, constant power is not required to protect the journal in a degraded state because the journal is saved to SSD and mirrored on the partner node. In contrast, the self-contained PowerScale F-series all-flash nodes protect the journal with a 16 GB DDR4 battery-backed NVDIMM-N instead of using PMP.