Hi Folks,
Just upgraded my FreeNAS with additional RAM (32GB total now) an L2ARC SSD and a couple of Kingston SSD Now V300s for mirrored SLOG (I think mirrored SLOG isn't strictly needed anymore - is this correct). Anyway, to cut a long story short, one of the SSDs appears to be much slower than the other.
They are drive IDs ada4 and ada5 respectively. My SLOG partitions are 4GB on each drive.
If they are mirrored and I run sustained sync write tests (from an ESXi VM on an NFS datastore) I can see in gstat that ada4 is 100% busy and ada5 is about 20% busy - a 2 GB write runs at about 18Megabytes/s. If I drop the mirror and run the same write test with ada5 it stays around 60% and the transfer saturates my Gigabit (117Megabytes/s).
The really odd thing is that they were both performing equally well before. The only thing I did was delete some old tunables that had been created by autotune before the RAM upgrade. After rebooting, performance on ada4 tanked, but ada5 remained good.
They are both on the mobo Sata 3 ports (Supermicro X10) and they are both syncing at the same bus speed according to dmesg.
Any ideas what might be going on here? Seems very odd.
Just upgraded my FreeNAS with additional RAM (32GB total now) an L2ARC SSD and a couple of Kingston SSD Now V300s for mirrored SLOG (I think mirrored SLOG isn't strictly needed anymore - is this correct). Anyway, to cut a long story short, one of the SSDs appears to be much slower than the other.
They are drive IDs ada4 and ada5 respectively. My SLOG partitions are 4GB on each drive.
If they are mirrored and I run sustained sync write tests (from an ESXi VM on an NFS datastore) I can see in gstat that ada4 is 100% busy and ada5 is about 20% busy - a 2 GB write runs at about 18Megabytes/s. If I drop the mirror and run the same write test with ada5 it stays around 60% and the transfer saturates my Gigabit (117Megabytes/s).
The really odd thing is that they were both performing equally well before. The only thing I did was delete some old tunables that had been created by autotune before the RAM upgrade. After rebooting, performance on ada4 tanked, but ada5 remained good.
They are both on the mobo Sata 3 ports (Supermicro X10) and they are both syncing at the same bus speed according to dmesg.
Any ideas what might be going on here? Seems very odd.