I got this new Sandisk X210 512gb. And from what I glimpsed from the Internet, I decided to test enabling dedup on a zpool with an SSD.
The testing was as follows:
First I created the pool. Then I set atime=off, compression=lz4, checksum=sha256 and dedup=on.
I just used
With already some 48.1G of data on the pool.
With the pool was almost three quarter full.
After the above test was done, I revert some settings, checksum=fletcher4 and dedup=off. Then I populated the pool with some new data:
I also used
From the above data, I observed that when dedup was running, there were approximately 1000 of 4KB reads from the SSD and 1000 of 128KB writes on the SSD. This may explain why there were spikes in service time and the reduced write throughput.
In conclusion, dedup on an SSD is also very demanding as there are mixed workload of reads and writes of different sizes.
The testing was as follows:
First I created the pool. Then I set atime=off, compression=lz4, checksum=sha256 and dedup=on.
I just used
zfs send and zfs recv to populate the new pool with data. The result that I got from zpool iostat -v was as follows:With already some 48.1G of data on the pool.
Code:
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 48.1G 396G 911 1.77K 3.56M 217M
gpt/x210-140785400341 48.1G 396G 911 1.77K 3.56M 217M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 51.5G 393G 1.15K 1.65K 4.59M 203M
gpt/x210-140785400341 51.5G 393G 1.15K 1.65K 4.59M 203M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 54.8G 389G 1.18K 1.60K 4.74M 197M
gpt/x210-140785400341 54.8G 389G 1.18K 1.60K 4.74M 197M
----------------------- ----- ----- ----- ----- ----- -----With the pool was almost three quarter full.
Code:
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 289G 155G 1.51K 1.33K 6.05M 165M
gpt/x210-140785400341 289G 155G 1.51K 1.33K 6.05M 165M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 293G 151G 1.47K 1.37K 5.88M 165M
gpt/x210-140785400341 293G 151G 1.47K 1.37K 5.88M 165M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 295G 149G 1.41K 1.28K 5.64M 157M
gpt/x210-140785400341 295G 149G 1.41K 1.28K 5.64M 157M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 298G 146G 1.42K 1.37K 5.68M 169M
gpt/x210-140785400341 298G 146G 1.42K 1.37K 5.68M 169M
----------------------- ----- ----- ----- ----- ----- -----After the above test was done, I revert some settings, checksum=fletcher4 and dedup=off. Then I populated the pool with some new data:
Code:
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 300G 144G 4 1.09K 18.2K 126M
gpt/x210-140785400341 300G 144G 4 1.09K 18.2K 126M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 307G 137G 6 3.30K 24.2K 368M
gpt/x210-140785400341 307G 137G 6 3.30K 24.2K 368M
----------------------- ----- ----- ----- ----- ----- -----
capacity operations bandwidth
pool alloc free read write read write
----------------------- ----- ----- ----- ----- ----- -----
x210-test2 315G 129G 3 3.43K 15.0K 371M
gpt/x210-140785400341 315G 129G 3 3.43K 15.0K 371M
----------------------- ----- ----- ----- ----- ----- -----I also used
gstat to monitor the SSD. When dedup was running, I observed that there were times when the mean service time would surge past 10 ms and in the mean time, the SSD was always 95%+ busy.From the above data, I observed that when dedup was running, there were approximately 1000 of 4KB reads from the SSD and 1000 of 128KB writes on the SSD. This may explain why there were spikes in service time and the reduced write throughput.
In conclusion, dedup on an SSD is also very demanding as there are mixed workload of reads and writes of different sizes.