Understanding cluster concepts

Cluster diagram

The following figure diagrams how the cluster hardware components are connected.

Cluster components and terminology

The following table describes components and explains terms that apply to clusters.


Term	Explanation
cluster	A pair of F630 filers connected so that one filer can detect when the other is not working and, if so, can serve the failed filer's data.
cluster interconnect	Cables and adapters with which the two filers in a cluster are connected and over which heartbeat and WAFL^TM log information are transmitted when both filers are running.
cluster monitor	Software that administers the relationship of filers in the cluster. You access the cluster monitor with the `cf` command.
failed filer	The physical filer that has ceased operating. It remains the failed filer until giveback succeeds.
FC-AL disk shelves	In normal operation, each filer in a cluster owns the set of disk shelves that are connected to it through the FC-AL A loop.
giveback	The return of identity from the virtual filer to the failed filer, resulting in a return to normal operation; the reverse of takeover.
heartbeat	A repeating signal transmitted from one filer to the other that indicates that the filer is in operation. Heartbeat information is also stored on disk.
takeover filer	A filer that remains in operation after the other filer stops working and that hosts a virtual filer that manages access to the failed filer's disk shelves and network connections. The takeover filer maintains its own identity and the virtual filer maintains the failed filer's identity.
local filer	The filer you are logged in to.
mailbox disk	One of a set of disks owned by each filer that is used to store the cluster state information of the filer. If that filer stops operating, the takeover filer uses the information in the mailbox disks in constructing a virtual filer. Mailbox disks are also used as filesystem disks.
normal mode	The state of a filer when there is no takeover in the cluster.
NVRAM mirror	A synchronously updated copy of the contents of a filer's NVRAM (Non-Volatile Random Access Memory) contents kept on the partner filer.
partner	From the point of view of the local filer, the other filer in the cluster.
partner mode	The method you use to communicate through the command line interface with the virtual filer during a takeover. Partner mode is accessed using the `partner` command.
takeover	The emulation of the failed filer's identity by the takeover filer in a cluster.
takeover mode	The method you use to interact with a filer while it has taken over its partner. The console prompt indicates when the filer is in a takeover state.
virtual filer	A software copy of the failed filer that is hosted by the takeover filer. The virtual filer appears to users and administrators like a functional version of the failed filer. For example, it has the same name as the failed filer.

Normal cluster operation

During normal operation, in addition to serving data, the cluster performs the activities described in the following table.

Action Description
Heartbeat transmission The heartbeat signals transmitted over the cluster interconnect cable tell each filer the state of the other filer. This information is written to disks on each filer that the cluster monitor designates as mailbox disks. The disks provide redundancy.
Monitoring The cluster monitor software looks at the heartbeat signal, commands from filers, and disk connections, and initiates a takeover under the following conditions:

The operator initiates a takeover using the cf takeover command
The partner is determined to have stopped functioning due to a lack of heartbeat activity
The partner's power-on-self-test (POST) fails

Note: If the filer that is not transmitting a heartbeat is rebooting, no takeover occurs.
Takeover may be disabled if NVRAM mirroring is not functioning.

NVRAM mirroring Each filer dedicates half of its NVRAM to a synchronously updated copy of its partner's NVRAM. If a takeover occurs, the takeover filer uses the cluster data in the part of the NVRAM dedicated to the failed filer to ensure that no information is lost.
Time synchronization Filers' times can deviate from each other. A daemon updates the times on filers periodically to keep them synchronized.

Action	Description
Heartbeat transmission	The heartbeat signals transmitted over the cluster interconnect cable tell each filer the state of the other filer. This information is written to disks on each filer that the cluster monitor designates as mailbox disks. The disks provide redundancy.
Monitoring	The cluster monitor software looks at the heartbeat signal, commands from filers, and disk connections, and initiates a takeover under the following conditions: The operator initiates a takeover using the cf takeover command The partner is determined to have stopped functioning due to a lack of heartbeat activity The partner's power-on-self-test (POST) fails Note: If the filer that is not transmitting a heartbeat is rebooting, no takeover occurs. Takeover may be disabled if NVRAM mirroring is not functioning.
NVRAM mirroring	Each filer dedicates half of its NVRAM to a synchronously updated copy of its partner's NVRAM. If a takeover occurs, the takeover filer uses the cluster data in the part of the NVRAM dedicated to the failed filer to ensure that no information is lost.
Time synchronization	Filers' times can deviate from each other. A daemon updates the times on filers periodically to keep them synchronized.

Part No. 210-02334
File: clust1.htm
Last updated: 04/26/99 08:11:41