Request from H3C

Scenario

Description

H3C Proposal

Daos feedback

Priority

If contributed to community

Feasibility and effort estimation

Owner(i.e. who own design who own dev and testing)

Delivery plan

(Q1, Q2, Q3 in 2022)

Risk and other comments/concern

User can create a block storage volume by specific size, name and others attributs

User can delete a block storage volume by specific volume name or uuid

online, offline

The volume size can be expanded in the IO operation service of online or offline cases.

The specific volume could be searched by their volume name or uuid

The volume presented appears to be the full provisioned capacity to the application servers, but nothing has been allocated until write operations occur. 

With thick provisioning, the complete amount of virtual disk storage capacity is pre-allocated on the physical storage when the virtual disk is created. A thick-provisioned virtual disk consumes all the space allocated to it in the datastore
right from the start, so the space is unavailable for use by other virtual machines or instance.

The volume's atributes can by modified, like: name

Recycle Bin is a snapshot recovery feature that enables you to restore accidentally deleted
 snapshots. When using Recycle Bin, if your snapshots are deleted, they are retained in the
 Recycle Bin for a time period that you specify.

You can back up the data on your volumes by taking point-in-time snapshots.
Snapshots are incremental backups, which means that only the blocks on the device that
have changed after your most recent snapshot are saved.
This minimizes the time required to create the snapshot and saves on storage costs
by not duplicating data. Each snapshot contains all of the information that is needed
to restore your data (from the moment when the snapshot was taken) to a new volume.

A clone of a Block Storage  volume is a copy made of an existing volume at a specific moment in time.

You can use quality of service (QoS) to guarantee that performance of critical workloads is not degraded
 by competing workloads. You can set a throughput ceiling on a competing workload to limit its impact
on system resources, or set a throughput floor for a critical workload, ensuring that it meets minimum
throughput targets, regardless of demand by competing workloads. You can even set a ceiling and floor
for the same workload.
A throughput ceiling limits throughput for a workload to a maximum number of IOPS or MBps, or IOPS and MBps.
The workload should CAN'T be out of the +10% to -10% limits scope. 

Use NVMe over Fabric protocol to setup a NVMe block storage.

Automatic cluster deployment

YES

Automatic cluster installation

YES

rollback

YES

YES

online upgrade

YES

offline upgrade

YES

rollback

YES

online monitoring

YES

offline monitoring

YES

YES

YES

YES

block storage pool

The user can ceate a specific block storage pool with size or name. 

online

The specific block storage pool can be expand its size by no interupt with their operation (online case)

Add new node (online)

The cluster can add a new node server to existing cluster systerm without interupt cluster's operation (online case)
The online case means that the storage system still can provide IO service to the custermer.
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 35%. 

delete a node (online)

The cluster can delete a new node server to existing cluster systerm without interupt cluster's operation (online case)
The online case means that the storage system still can provide IO service to the custermer.
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 35%. 

Add a new SSD (online)

The node server can add a new NVMe SSD without interupt cluster's operation (Online case)
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 20% in the 3 nodes server cluster case.. 

Replace a SSD (online)

The node server can replace a new NVMe SSD with original one. (online case)
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 20% in the 3 nodes server cluster case.. 

remove exist SSD (online)

The node server can remove a running or failed NVMe SSD in system. (Online case)
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 20% in the 3 nodes server cluster case.. 

Add a new PMEM (online)

The node server can add a new PMEM without interupt cluster's operation (Online case)
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 20% in the 3 nodes server cluster case.. 

Replace a PMEM (online)

The node server can replace a newPMEM with original one. (online case)
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 20% in the 3 nodes server cluster case.. 

remove exist PMEM (online)

The node server can remove a running or failed PMEM in system. (Online case)
The IO operation fluctuation is allowed, the fluctuation scope CAN'T exceed over 20% in the 3 nodes server cluster case.. 

Network exception robust

The cluster could keep providing IO operation service in the random network failiure case.
Descripton: in 3 nodes cluster system, only 1 node's network should be in the random network failiure case testing.

NIC exception robust

The cluster could keep providing IO operation service in the random node's NIC failiure case.
Descripton: in 3 nodes cluster system, only 1 node's NIC should be in the random network failiure case testing.

SSD Disk Exception is the test driven case

Node Exception is the test driven case

PMEM space optimazation

Currently, the PMEM size is divided with the amount of NVMe SSD, it caused a big PMEM space waste.
The PMEM's space  running out but the NVMe SSD's space still with a lot of usage space.

4TB/H

The data rebuild speed above 4TB/H, in the cluster of 3 nodes each node has 8 NVMe SSD.
The IO operation service should not be considered, in simple case.

IO recovery time

When a node rebooted, how many seconds the IO service could be recovery back to the 100% before it's reboot.
The IO reovery time should be in 300 seconds in 3 nodes cluster system