Introduction to Linux Clustering
increase performance when using a TCP-based storage service like iSCSI.
7.2 Access methods for centralised storage
Whilst centralised storage can sometimes provide the data via a network filesystem, it is
more common to use iSCSI or a SAN.
Both iSCSI and SAN provide access to the storage as if it was a local disk. It is then
necessary to run a cluster-capable filesystem on top of them such as GFS.
Note of interest: It is possible to have a non-clustered filesystem on a shared drive, but
seriously bad issues would occur if you accidentally mounted it in two places at one time.
7.2.1 ACCESSING SAN DIRECTLY
A decent SANs allows the administrator to split the SAN into a number of logical hard
drives, and then export only the desired drives to each node.
The drives appear on the node just like any locally connected SCSI/SATA drive.
7.2.2 ACCESSING NAS WITH ISCSI
Like a SAN, many NASes can be configured to split the storage into a number of logical
drives.
Because the NAS is not connected locally, it uses a TCP/IP protocol called iSCSI. This
means iSCSI can be routed, and even transferred over the internet (although the performance
on this would be terrible without a high-speed, low-latency link).
iSCSI is used by attaching an iSCSI target. Once connected the iSCSI export appears just
like a local SCSI hard drive.
It is important to note that the naming of the drives may be change, thus it is important to
use udev to ensure stable naming.
Further information about how to identify and name iSCSI devices using udev can be found
in the scsi_id man page.
7.2.3 ACCESSING NAS WITH OTHER PROTOCOLS
If you are using some other protocol, such as ATA over Ethernet, you will need to run
software on the nodes to make the NAS shares appear as block devices on the server, which
is similar to iSCSI in concept.
© Copyright 2008 Jethro Carr Page 20/33
Komentáře k této Příručce