Hedvig: Notes on the solution design

Hardware recommendations : Hyperscale

Nodes

• For single cluster minimum of 3 storage nodes are recommended.
• To scale performance and capacity , add storage nodes to hedvig .
• Node should be added in multiples of 3

CPU and Memory

• Servers should be dual socket E5/E7 with on board storage, Jbod are also supported but in Jbod or Raid0 mode.
• Hedvig is optimized for scale out. It is preferred to add storage node then to add storage to single server.
• The minimum memory configuration for each physical server for a Hedvig Cluster Node (non-hyperconverged) is usually 48 GB.

Typical deployments use servers with 128 to 256 GB.

Network

• 10G network for internal network and 1 G network for management.

Drives

• No limitation to any drive SAS/SATA/SSD

Note: It should be yet to be verified that SAS 12Gbps will have any advantage as storage tearing is in place.

Hardware recommendations : Hyperscale

• In hyper-converged architecture , storage and compute scale together on commodity servers.
• Hedvig recommends separate controller for VM (hypervisor ,vm storage and client side cache) and Storage.
• One drive, for a datastore for the Hedvig VMs (that is, the Hedvig Cluster Node VM and Hedvig Storage Proxy VMs).
• Additional drives, which are dedicated to the Hedvig Cluster Node and which should all reside on the second controller that can be passed through to the Hedvig Storage Node VM.

Storage node

• Two SSD are used for metadata commitlog and Hblock commitlog and data.
• Further SSD are used server side cache , auto-tiering