About Clusters |
A cluster consists of a pair of filers that are connected through an interconnect adapter and cables, and are configured so that both filers share access to a set of Fibre Channel disks, subnets, and tape drives to provide fault tolerance.
When one filer in a cluster undergoes a system failure and cannot reboot, the partner filer in the cluster can take over the failed filer's functions and serve network clients the data on the failed filer's disks. This is known as a takeover. In addition to an automatic takeover, an administrator can issue a manual takeover at any time to force a takeover, for instance, to allow scheduled maintenance to be performed on a filer.
An emulated filer is created on the takeover filer. This is a software copy of the failed filer that is hosted by the takeover filer. The emulated filer appears to users and administrators like a functional version of the failed filer.
After the failed partner is running normally again, you issue a giveback, which returns the identity from the emulated filer to the failed filer, resulting in a return to normal operation.
The filers in a cluster have the following characteristics:
They are connected to each other through a cluster interconnect adapter and cable through which they
Continually check whether the other filer is functioning
Mirror log data each other's NVRAM
Synchronize each other's time
They use two Fibre Channel Arbitrated Loops (FC-AL) through which each filer
Uses the A loop to manage its own disks
Uses the B loop in takeover mode to manage its partner's disks
They each own their spare disks and do not share them with the other filer.
They may reside on the same NT domain or on different domains.
If one filer in a cluster stops functioning physically, its partner filer accomplishes the following operations:
Creates an emulated filer using the failed filer's identity
Accesses the failed filer's disks and serves its data to clients
Maintains its own identity and continues to serve its own data to clients
When the failed filer is functioning again, the following operations occur:
An administrator initiates a giveback that terminates the emulated filer on the partner.
The failed filer resumes normal operation, serving its own data.
The cluster resumes normal operation, with each filer ready to take over for its partner if the partner fails.
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 through the cluster interconnect adapter and cables 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:
Note that if the filer that is not transmitting a heartbeat is rebooting, no takeover occurs. Takeover might be disabled if NVRAM mirroring is not functioning. Automatic takeover is disabled if the partner mailbox disks are not accessible. |
|
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. |