How does ZFS intent log (ZIL) improve write performance?
In the world of high-performance storage, there is a constant battle between speed and safety. Databases demand "Synchronous Writes"—meaning the application won't move to the next task until it is 100% sure the data is safely on a physical disk.
The ZFS Intent Log (ZIL) is the specialized mechanism that allows ZFS to satisfy that safety requirement without dragging performance down to the speed of a spinning disk.
Normally, when an application writes data, ZFS stores it in a RAM-based buffer called the ARC (Adaptive Replacement Cache). Every 5 seconds, it flushes that data to the main disks in a "Transaction Group" (TXG).
The problem? If the power fails at second 4, that data in RAM is gone. For a database, this is catastrophic. Therefore, databases use Synchronous I/O, forcing ZFS to commit the data to stable storage immediately. Without a ZIL, ZFS would have to stop everything and do a "Random Write" to the main data pool for every single tiny transaction—which is incredibly slow.
The ZIL acts as a persistent log of all "in-flight" synchronous transactions. Instead of waiting for the main 5-second flush, ZFS writes the transaction details to the ZIL immediately.
The Request: The application sends a Sync Write.
The Log: ZFS writes the data to the ZIL.
The Acknowledgement: ZFS tells the application, "The data is safe."
The Cleanup: Later, during the standard 5-second TXG flush, the data is written permanently to the main pool, and the ZIL entry is cleared.
By default, the ZIL lives on the same physical disks as your data. This means your "Log Writes" are competing with your "Data Writes," which creates a bottleneck.
To truly improve performance, we use a SLOG (Separate Intent Log). This is a dedicated, lightning-fast physical device (like an NVMe SSD or Optane drive) used only for the ZIL.
Without SLOG: The ZIL is on slow HDDs. Latency is high.
With SLOG: The ZIL is on an SSD. The application gets its "Success" message in microseconds, even if the main data pool consists of slow, high-capacity spinning disks.
It is a common myth that adding an SSD for the ZIL makes all writes faster.
Asynchronous Writes: (Standard file copies, backups) These bypass the ZIL entirely. They go straight to RAM and wait for the 5-second flush. Adding a ZIL will not help these.
Synchronous Writes: (Database commits, NFS, Virtual Machine disks) these depend on the ZIL. Adding a fast SLOG will make these significantly faster.
| Feature | Standard Pool (No SLOG) | Pool with NVMe SLOG |
| Sync Write Latency | High (Disk Seek Times). | Ultra-Low (Flash Speeds). |
| Data Safety | High (All writes logged). | High (Same safety, higher speed). |
| Workload Suitability | General file storage. | Databases (Oracle, SQL), NFS, VMs. |
| Cost | Baseline. | Additional cost for high-end SSD. |
The ZIL is ZFS’s way of saying "I’ve got it" before the data actually hits the final resting place on disk. By offloading the ZIL to a dedicated SLOG device, you remove the latency penalty of synchronous writes, allowing your databases to perform at the speed of flash while maintaining the integrity of ZFS.