Intel and Seagate both have web pages explaining over-provisioning and noting that it has a positive effect on random write performance. But after running sysbench random write test on a couple of configurations on my Intel SSD (both not partitioning some space and making it unadressable, by enabling host-protected area), the only effect I saw was a marginal performance regression when compared to a partition taking up the whole drive.
Is it possible I’m testing wrong, or does my Intel 545s’ firmware just work differently from Intel’s data centre drive they show gaining write iops when user over-provisioned? Seagate doesn’t name their drive, they only mention it’s based on a SF-2281 controller, which appears to be used widely in client drives.
I’m ran my tests on Linux, on an ext4 partition.
The difference shouldn’t show until the drive is getting near full; or if they employ a dynamic SLC write buffer, after a large amount of writes in a short time.
3 Likes