I added an SLOG to my test pool. Even though most of the workload I will generate will be asynchronous, an SLOG shouldn't hurt and at most sit idle. (For Science!)
While loading the pool with a steady stream of writes at 5-15 MB/s, I am monitoring the data in and out using arcstat.py, zilstat, gstat, and top for good measure. As the workload is quite mild for what the pool is capable of, the two disks (2x4TB HDD mirrored) generally only activate every 5 seconds to write the current transaction group from RAM (vfs.zfs.txg.timeout default value).
What I don't understand is that I see a regular burst of writes to the SLOG, lasting less than 1 second, once every 30 seconds, like clockwork. Usually in the range of 20-70 MB. (And also the odd 1-2 ops at random intervals, more seldom.)
My tunables and mostly all settings are default. I'm accessing CIFS shares only, from Windows 7 and Windows 10 clients.
Is there a ZFS mechanism which is issuing synchronous writes every 30 seconds, and which mechanism is it?
While loading the pool with a steady stream of writes at 5-15 MB/s, I am monitoring the data in and out using arcstat.py, zilstat, gstat, and top for good measure. As the workload is quite mild for what the pool is capable of, the two disks (2x4TB HDD mirrored) generally only activate every 5 seconds to write the current transaction group from RAM (vfs.zfs.txg.timeout default value).
What I don't understand is that I see a regular burst of writes to the SLOG, lasting less than 1 second, once every 30 seconds, like clockwork. Usually in the range of 20-70 MB. (And also the odd 1-2 ops at random intervals, more seldom.)
My tunables and mostly all settings are default. I'm accessing CIFS shares only, from Windows 7 and Windows 10 clients.
Is there a ZFS mechanism which is issuing synchronous writes every 30 seconds, and which mechanism is it?
