Surprised! How can the write IOPS be 10x faster than the cloud native disk performance
In a competitive fio write performance benchmark between 3rd party storage solution and cloud native disk, we noticed that the write performance for the 3rd party storage solution is 10x faster than the cloud native. This usually seems impossible since we would argue that no one can beat the raw disk performance. The following case shows an interesting write performance optimization from the storage solution under test.
Cloud native raw disk performance(fio,4k,write)
From the benchmark result, it shows the IOPS limit of the cloud drive is ~7500. This aligns with the cloud storage SPEC in use.
10x faster IOPS is observed!!!
In the output of fio, the IOPS(>75k) is 10x faster than the result from cloud native raw disk(~7500). This can be verified from the iostat output. The IOPS at logic volume layer is aligned with the fio output.
Iosize matters!!!
When the write request comes to the physical disk from logic volume, the I/O size is changed from 4k to 400k. It indicates the smaller writes got merged to larger writes.
Conclusion
With limited IOPS on the cloud drive, larger IO size really help improve the write performance by reducing the write requests to disk. With proper write performance optimization at logic volume layer, the write performance can be boosted 10x faster with no surprise!!!