Features and performance of premium performance disks - ApsaraDB RDS

This topic describes the concepts and performance of premium performance disks to help you select a suitable disk.

Features

Premium performance disks are a new storage type supported by ApsaraDB RDS Custom instances. They are compatible with all features of enterprise SSDs (ESSDs) and support the I/O performance burst and 16 KB atomic write features. These features provide you with more efficient and flexible storage solutions.

I/O performance burst

Method to enable: This feature is integrated into premium performance disks and is enabled by default. No additional configurations are required.
Cost: The feature is provided free of charge.
Maximum IOPS: The I/O burst feature allows the IOPS of a disk to exceed the baseline performance limit. During peak hours, IOPS can reach up to 1,000,000.
Advantages:
- This feature improves the I/O capability to handle workload bursts during peak hours.
- This feature supports memory I/O burst. This allows for the dynamic allocation of additional resources, such as I/O bandwidth and memory, in specific scenarios to effectively handle heavy workloads over a short period.

16 KB atomic write

Method to enable: To enable this feature, see Appendix: Enable 16 KB atomic write.
Cost: The feature is provided free of charge.
Core technology: The doublewrite mechanism is disabled. This feature uses checksum and log rollback mechanisms to ensure the atomicity of each data page write, which securely simplifies the write process.
Performance improvement:
- Reduced I/O writes: This feature optimizes the dirty page flushing process and significantly reduces the IOPS and bandwidth required for instance disk writes.
- QPS increased by more than 40%: This feature significantly improves the query performance of instances for high-concurrency scenarios. For more information about the test, see Write optimization performance testing.

Performance of premium performance disks

Storage capacity of premium performance disks (GB)	Baseline performance (Same as PL1 ESSDs)
100 GB <= Storage capacity < 1,000 GB	I/O bandwidth: 125 to 350 MB/s IOPS: 6,800 to 50,000
1,000 GB <= Storage capacity < 2,000 GB	I/O bandwidth: 350 MB/s IOPS: 50,000
2,000 GB <= Storage capacity < 4,000 GB	I/O bandwidth: 350 MB/s IOPS: 50,000
4,000 GB <= Storage capacity < 16,000 GB	I/O bandwidth: 350 MB/s IOPS: 50,000
16,000 GB <= Storage capacity < 64,000 GB	I/O bandwidth: 350 MB/s IOPS: 50,000

Appendix: Enable 16 KB atomic write

To enable the 16 KB atomic write feature, perform the following steps:

Run the following command to format the disk.
Important
This operation purges all data from the disk. Before you proceed, make sure that you have backed up all important data.
```
mkfs.ext4 -m 0 -C 16K -O extent,uninit_bg,bigalloc -E lazy_itable_init=1 /dev/vdb
```
- -m 0: disables the reserved capacity of the file system. By default, 5% of the disk space is reserved.
- -C 16K: sets the block size to 16 KB.
- -O extent,uninit_bg,bigalloc: enables ext4 features:
  - extent: improves the storage performance of large files.
  - uninit_bg: accelerates file system checks.
  - bigalloc: optimizes the allocation efficiency of large files.
- -E lazy_itable_init=1: delays the initialization of the inode table to accelerate the formatting speed.
Run the following command to adjust the operating system settings. Set the max_sectors_kb parameter for all storage devices to 64 (64 × 512 B = 32 KB).
```
for file in /sys/block/*
do
 echo "before: $file/queue/max_sectors_kb = `cat $file/queue/max_sectors_kb`"
 echo 64 > $file/queue/max_sectors_kb
 echo "after: $file/queue/max_sectors_kb = `cat $file/queue/max_sectors_kb`"
done
```
max_sectors_kb: defines the maximum amount of data allowed in a single I/O request.
Adjust the MySQL parameters. Add the following parameters to the my.cnf configuration file.
```
innodb_use_native_aio = ON
innodb_doublewrite = OFF
```

References

ApsaraDB RDS Custom overview