Merge function

Data can be written to or read from drives in a RAID in sequential or random order. When incoming requests are sequential and the recording block sizes are small, it is beneficial to wait for them to accumulate and form a large block. Then, merge these requests and write or read data from the RAID in large blocks. Applying such access patterns can improve system workload performance, as this approach reduces the number of read-modify-write operations ¹ on syndromic RAIDs.

The merge parameters for requests accumulation for a RAID can be configured manually using xicli raid create and xicli raid modify commands.

The --merge_read_enabled parameter activates the Merge function for incoming read requests, allowing their accumulation. You can specify a waiting time between incoming read requests in sequential areas using the --merge_read_wait parameter. At the end of the waiting time, the requests are merged together if possible. You can use the --merge_read_max parameter to specify the maximum waiting time for request accumulation. Usually, large I/O sizes require large values for these parameters.

# xicli raid create -n media5 -l 5 -d /dev/nvme0n1 /dev/nvme1n1 /
dev/nvme2n1 /dev/nvme3n1 -ss 64 -mre 1

The --merge_write_enabled parameter activates the Merge function for incoming write requests, allowing their accumulation. You can specify a waiting time between incoming write requests in sequential areas using the --merge_write_wait parameter. At the end of the waiting time, the requests are merged together if possible. You can use the --merge_write_max parameter to specify the maximum waiting time for request accumulation. Usually, large I/O sizes require large values for these parameters.

# xicli raid modify -n media5 -mww 500

If the access pattern is mainly random or request queue depth is small, the waiting time will not allow merging requests.

The function only works when the condition is met:

data_drives * strip_size ≤ 1024

where

• “data_drives” is a number of drives in the RAID (for RAIDs 5, 6, 7.3 or N+M) or in one RAID group (for RAIDs 50, 60, or 70) that are dedicated for data;

• “strip_size” is a selected stripe size for the RAID (strip_size value) in KiB.

The “data_drives” value depending on a RAID level:

Deactivate Merge when the request queue depth of user's workload is not enough to merge a full stripe. Activate Merge, if

iodepth * block_size >= data_drives * strip_size

where “block_size” is a block size of the RAID (the block_size value in RAID parameters) in KiB.

xiRAID offers an algorithm that automatically selects the waiting time for accumulating write requests. This algorithm determines whether the merging of write requests is enabled (--merge_write_enabled parameter), the wait time between write requests (--merge_write_wait parameter) and the maximum wait time for stripe accumulation (--merge_write_max parameter). The time between incoming requests is set to 0-3000 ms, and the maximum wait time for stripe accumulation is set to to 20000 ms.

You can enable the Automatic Merge function during RAID creation and modify it later using the --adaptive_merge parameter. Once the Automatic Merge algorithm determines the waiting time, you can prevent it from changing these settings later on using the --single run parameter.

Example: Enable the Adaptive Merge function for the RAID 5 named "media5" and prevent it from changing the determined settings later on. After the settings have been determined, the Adaptive Merge is automatically deactivated and displayed as 'False' in the xicli raid show output.

# xicli raid modify -n media5 --adaptive_merge 1 --single_run