Was this helpful?
I/O Settings
The effective disk block access unit is:
group_size * block_size
This value needs to be sufficiently large; even a single magnetic disk now needs a value of 2 MB to get reasonably good sequential throughput. If you have a multi-disk (RAID) system, you need to multiply this efficiency target by the number of disks, because the requests will be spread across all disks in equal chunks.
Finally, current SSDs can be efficient with somewhat smaller sizes. Group_size*block_size should be at least 512 KB for SSDs.
For example, in a disk configuration consisting of 8 SAS drives in RAID5, you could opt for 16 MB disk transfers, using a group_size of 8 and a block_size of 2 MB:
[cbm]
bufferpool_size = 4G
group_size = 8
block_size = 2M
[memory]
max_memory_size = 12G
In a RAID 0,5 system, group_size*block_size should ideally be an exact multiple of the RAID stripe size times the number of disks. In a RAID 1,10,01 system, group_size*block_size should be stripe size times half the number of disks.
Last modified date: 01/30/2023