The VSDs in this scenario are mapped to the raw logical volumes lv_X and lv_Y. Node X is a client of Node Y’s VSD, and vice versa. Node X is also a direct client of its own VSD (lv_X), and Node Y is a direct client of VSD lv_Y. VSD configuration is flexible. An interesting property of the architecture is that a node can be a client of any other node’s VSD(s), with no dependency on that client node owning a VSD itself. You could set up three nodes with powerful I/O capacity to be VSD servers, and ten application nodes, with no disk other than for AIX, PSSP, and the application executables, as clients of the VSDs on these server nodes.

VSDs are defined in the SDR and managed by either SP SMIT panels or the VSD Perspective. VSDs can be in one of five states as shown in Figure 18 on page 192.

Undefined

define undefine

Defined

cfgvsd cfgvsd

Stopped

preparevsd

stopvsd

Suspended

resumevsd

suspendvsd

Active

Available

VSD information is available in the SDR

Open/close and I/O requests fail

I/O requests queued and open/close request serviced

Open/close and

I/O requests serviced

Figure 18. VSD State Transitions

This figure shows the possible states of a VSD and the commands used to move between states. VSD configuration changes, or manual recovery of a failed VSD, require you to move the VSD between various states.

The distributed data access aspect of VSD scales well. The SP Switch itself provides a very high-bandwidth, scalable interconnect between VSD clients and servers, while the VSD layers of code are efficient. The performance

192 IBM Certification Study Guide AIX HACMP

Page 210
Image 210
IBM SG24-5131-00 manual Undefined