Feature List
Feature |
Sub-feature |
Feature Description |
Constraint |
Required Software Package |
Supported on VMs |
|---|---|---|---|---|---|
x86-Kunpeng hybrid deployment |
In block/object/file storage services, x86 and Kunpeng servers are deployed in the same storage pool. |
|
- |
No |
|
In block/object/file storage services, Kunpeng servers are used to expand the capacity of an x86 server cluster. |
|||||
Bcache |
In block/object/file storage services, |
|
Patch package: |
No |
|
I/O passthrough |
I/O pass-through improves the storage performance in balanced configuration and 7:3 read/write hybrid scenarios. |
|
RPM package: Contact Huawei technical support. |
No |
|
Certification by commercial-edition Ceph for Ubuntu |
Kunpeng server (model 2280) |
The Kunpeng server (model 2280) has passed the certification by commercial-edition Ceph for Ubuntu. |
- |
- |
- |
Kunpeng server (model 5280) |
The Kunpeng server (model 5280) has passed the certification by commercial-edition Ceph for Ubuntu. |
||||
Data compression for block storage service |
Compared with the open source LZ4 compression, the Kunpeng BoostKit for SDS compression algorithm delivers a 25% higher compression ratio and a 25% lower effective capacity cost per TB. |
|
Binary package: Contact Huawei technical support. |
No |
|
In the balanced configuration, the Kunpeng BoostKit for SDS compression algorithm delivers a 10% higher bandwidth performance than the open source LZ4 compression. |
|||||
Data compression for object storage service |
Compared with the open source LZ4 compression, the Kunpeng BoostKit for SDS compression algorithm delivers a 25% higher compression ratio and a 25% lower effective capacity cost per TB. |
||||
In the balanced configuration, the Kunpeng BoostKit for SDS compression algorithm delivers a 10% higher bandwidth performance than the open source LZ4 compression. |
|||||
Performance acceleration for write operations in block storage and object storage services |
In the balanced configuration, it prevents the random write performance from dropping to zero and increases the random write performance using three-copy data pools by 20%. It also improves the mixed read/write (7:3) performance by 10%. |
|
RPM package: Contact Huawei technical support. |
No |
|
Block storage service |
In the balanced configuration, the EC Turbo (4+2) performance reaches over 80% of the x86 three-copy performance, and the storage cost is reduced by 50%. |
|
RPM package: Contact Huawei technical support. |
No |
|
Object storage service |
In the balanced configuration, the EC Turbo (4+2) performance reaches over 80% of the x86 three-copy performance. The storage cost of large I/Os is reduced by 50%, and the storage cost of small I/Os is approximately the same. |
||||
Block storage service |
In balanced configuration, the 4 KB hotspot read performance is improved by 100%. |
Supported OSs: openEuler 20.03 LTS SP1 openEuler 22.03 LTS SP2 |
Patch package: Contact Huawei technical support. |
No |
|
Database service |
In all-flash scenarios, the mixed read/write performance of 64B/128B, 64B/512B, and 64B/1024B key-values is improved by 30% after this feature is enabled compared with the open source RocksDB database. |
Supported OSs: openEuler 20.03 LTS SP1 openEuler 22.03 LTS SP2 |
Patch package: Contact Huawei technical support. |
No |
|
Reliability service |
Slow HDD detection: SATA HDD FDR > 80%, precision > 70% HDD fault prediction: SATA HDD FDR > 60%, FAR < 0.5% |
Supported OSs: openEuler 20.03 LTS SP1 openEuler 22.03 LTS SP2 |
RPM package: Contact Huawei technical support. |
No |
|
Block storage service |
In a data compression scenario with the balanced configuration, the compression ratio of "Kunpeng + data compaction + data compression" is 1.3 times that of "x86 + open-source data compression." |
|
Patch package: Contact Huawei technical support. |
No |
|
Object storage service |
In a data compression scenario with the balanced configuration, the compression ratio of "Kunpeng + data compaction + data compression" is 1.3 times that of "x86 + open-source data compression." |
||||
EC coding and decoding |
The vectorized EC coding and decoding scheme replaces the high-order finite field multiplication of traditional scalar encoding with low-order binary XOR operations and reuses intermediate calculation results through coding orchestration to reduce the number of operands. |
|
RPM package: Contact Huawei technical support. |
No |
|
CRC16 verification |
The CRC16 algorithm optimized based on the principles of a large-number modulo algorithm is used to replace the standard CRC16 algorithm. It has better Kunpeng affinity, improving system performance. |
||||
CRC32 verification |
The CRC32 algorithm optimized based on the Kunpeng platform is used to replace the standard CRC32 algorithm, improving system performance. |
||||
Object storage metadata reduction |
Object storage service |
Metadata is separated from application object data by setting different storage classes for buckets and objects in the buckets. |
- |
No |
|
SPDK I/O acceleration |
Block storage service |
A Ceph cluster is deployed in containers on Kunpeng servers running openEuler 20.03. The integration of the SPDK, UCX, and KSAL maximizes storage and network performance, answering the need for high throughput and low latency in modern distributed storage. |
The SPDK I/O acceleration feature adopts a container-based deployment of SPDK and Ceph 17.2.7. SPDK + other distributed storage modes are not supported. |
No |