Get the App
SLTechnology News&Howtos  ›  Servers  › 

NetApp DataONTAP Cluster Mode Learning Notes 2

Shulou Source: shulou.com Published: 2022-06-03 01:14:07 09月22日 Update

How WAFL works

The basic structure is as follows:

Client sends write request

Controller 1 writes data to system memory and NVRAM and synchronizes data to Controller 2

Controller 2 sends acknowledgement to controller 1

Controller 1 sends acknowledgement to client

Another client sends a write request

Controller 1 writes data to system memory and NVRAM and synchronizes it to Controller 2

Controller 2 sends acknowledgement to controller 1

Controller 1 sends acknowledgement to client

Another client sends a write request to controller 1, which writes memory and NVRAM and synchronizes the data to controller 2

Controller 2 sends acknowledgement to controller 1

Controller 1 sends acknowledgement to client

The controller NVRAM is now half full

At this point, the consistency point is triggered, and the controller writes the data in memory to disk

(Conditions to trigger consistency point: NVRAM is half full; CP is more than 10 seconds since last occurrence; snapshot is created; administrator shuts down system)

Consistency completed, NVRAM cleared

WAFL sends an acknowledgement to the client as soon as data is written to memory. This is done before data is written to disk to optimize performance. For clients, the data has been permanently written to storage. If power is lost, the contents of the system memory will be lost. If data is written only to system memory, an inconsistent state results. NVRAM is non-volatile-it can hold data in the event of a power outage and recover from NVRAM if a power outage occurs before data is written to disk. NVRAM writes data to system memory and writes it to disk from a consistent point. Writes data to both directors in the HA pair so that if there is a takeover HA peer, the data can be written to disk.

When controller 1 fails,

Controller 2 writes the contents of NVRAM to memory

Controller 2 then writes the data in memory to disk

Controller 2 clears NVRAM

Client sends write request to controller 1

Controller 1 writes data to memory and NVRAM and synchronizes it to Controller 2

Controller 2 sends acknowledgement to controller 1

Controller 1 sends acknowledgement to client

Client sends read request to controller 1

Controller 1 first looks for data from memory

If the data exists in memory, move it up to the top

Other data is moved down and the found data is sent to the client

Another client also sends a read request

Control not found in memory, need to read from disk

Data read from disk is placed at the top of memory

Indirect data access: Client sends read request via controller 2

Controller 2 sends the request to Controller 1

Controller 1 retrieves data from disk and places it in memory, and sends it to Controller 2 for presentation to clients

This data will be placed at the top of memory

Tags: Control controller data memory client client disk system consistent and will synchronization top consistency two content principle snapshot performance right and wrong condition Apple Docker Huawei Linux macOS MariaDB Microsoft MySQL NVidia OPPO Reno Linux MySQL OPPO Reno Huawei macOS