Write I/O latency increases significantly

Hello, Lisa!

Have a nice day!

For a LUN configured with deduplication and compression, fingerprint comparison is required before cached data is written to disks. The comparison takes a long time. Therefore, cached data cannot be written to disks in a timely manner, preventing sufficient cache space from being reserved to receive new data. As a result, the write I/O latency increases, and host performance is affected. 

The maximum write bandwidth provided by a LUN configured with deduplication and compression ranges from 80 MB/s to 100 MB/s.

No high bandwidth pressure on controllers at the problem time.

Hope can help you!

Thanks for your reply!
So what should I do?

Lisa, you can refer to the following steps:

1. Migrate data to thick/thin LUNs. (The common write bandwidth reaches the bottleneck.)

2. Limit the number of UNMAP I/Os to be received from the host. (UNMAP I/Os cause the write bandwidth to reach a bottleneck.)

3. Upgrade the storage system to a version later than V300R006C60SPC100/V500R007C50SPC100, and migrate data to deduplication and compression LUNs newly created in this version. The performance is expected to increase by 20% (100 MB/s to 120 MB/s).

4. Persuade customers to use Dorado storage to improve both the performance and deduplication/compression ratio.

Hope can help you！