Single service comparison

Huawei splits the AWS EBS scope across two complementary services that are not interchangeable. EVS is the multi-tenant, on-demand per-volume block store attached to ECS/BMS — the closest analogue to EBS gp/io/st volumes, sized per disk up to 32 TiB with classes such as General Purpose SSD, General Purpose SSD V2, Ultra-high I/O, and Extreme SSD. DSS is a physically dedicated storage pool (high I/O or ultra-high I/O) sold as a yearly reserved pool from 13.6/7.2 TB minimum, attached to ECS/BMS/DCC, and targets HPC, OLAP, and compliance workloads needing resource isolation. Map standard EBS volumes to EVS and dedicated/performance-isolated workloads to DSS rather than treating them as equivalents.

Control and data APIs differ in shape. EBS exposes volume, snapshot, and Fast Snapshot Restore operations under EC2/EBS APIs with.aws CLI parity, cross-region snapshot copy, and S3-backed snapshots. EVS and DSS use Huawei-native EVS APIs and DSS APIs against ECS/BMS, with snapshots and backups governed through CBR (Cloud Backup and Recovery) rather than a standalone snapshot service; cross-region copy is a CBR/vault operation. Encryption uses Huawei KMS for both EVS and DSS with AES-256, but key policy, IAM, and tagging models follow Huawei IAM, not AWS IAM, so automation and policy-as-code must be rewritten.

Scaling, HA, and operational responsibility boundaries differ. EBS volumes are zonal resources with single-AZ durability and rely on Multi-AZ deployment patterns, EBS Snapshots, and Application-consistent tooling for HA. EVS and DSS both use three-copy intra-region redundancy advertising 99.9999999% durability, and shared EVS/DSS disks (up to 16 attachments with SCSI-3 reservations) replace EBS Multi-Attach for cluster workloads like WSFC and Oracle RAC. Backup orchestration, retention, and restore SLAs are owned by the customer through CBR; there is no automatic EBS-style snapshot lifecycle, so backup policies, vaults, and replication must be explicitly configured at cutover.

Assess per-volume workload profile, IOPS/throughput, latency, size, attachment (single vs Multi-Attach), encryption, snapshot cadence, retention, and cross-region DR objectives. Map standard gp/io/st EBS volumes to EVS with the matching class (General Purpose SSD V2 for adjustable IOPS, Ultra-high I/O or Extreme SSD for databases), and reserve DSS for workloads needing dedicated pools, predictable isolated throughput, or compliance-driven physical isolation such as HPC, OLAP, RAC, and SAP HANA. Confirm region/availability-zone availability and quotas for both EVS and DSS before target design.

There is no Huawei-provided one-click EBS-to-EVS/DSS migration; treat volume migration as a data movement exercise. Common patterns are host-level replication (rsync, native DB redo/log shipping, or DRBD-style replication) from the on-premises or AWS EC2 source to new ECS/BMS instances on EVS/DSS, or exporting EBS snapshots and restoring object-backed images via CBR where supported. For multi-attach EBS volumes, replicate data and re-establish shared EVS/DSS disk multi-attach (SCSI-3) on the target. Validate that required block device naming, mount options, filesystem, and LVM layouts are reproduced, and that KMS-encrypted volumes are re-encrypted against target Huawei KMS keys.

Validate read/write latency, IOPS, throughput, and consistency under peak load before cutover, comparing observed EVS/DSS metrics against the recorded EBS baseline rather than published peaks. Run application-level consistency checks (fsck, database checksums/recovery) and rehearse restore from CBR backups. Cut over during a maintenance window using a final sync/quiesce checkpoint, repoint DNS and connection strings, and keep the AWS source intact for rollback until sign-off. Update monitoring to Cloud Eye and CES thresholds for disk IOPS, throughput, and health, replacing CloudWatch EBS metrics.

Mind the cost and operational model shift. EBS charges GB-month capacity plus io-class provisioned IOPS and snapshot storage; EVS charges per GiB-month by disk class with GPSSD V2 offering tunable IOPS/throughput, while DSS pools are billed yearly as dedicated capacity (examples from documentation: High I/O pool from ~$39,840/year (~$0.34/GiB-month) and Ultra-high I/O from ~$64,430/year). DSS minimum pool sizes and yearly commit change unit economics versus per-volume on-demand EBS, so consolidate volumes into shared pools where justified and re-model TCO including CBR backup capacity, vault replication, and cross-region transfer. Account for rewritten automation, IAM/KMS policy, tagging, and quotas, and confirm DSS pool lead time and regional availability before committing production workloads.