NonStop S-Series Server Description Manual (G06.24+)

Introduction

HP NonStop S-Series Server Description Manual—520331-003

1-10

Fault-Tolerant Process Communication

Fault tolerance for processes is accomplished by providing a backup process in

some other processor and providing two or more paths of communication between

them, so that if one path or processor should fail, the other processor and paths will

remain operable. This basic principle of fault-tolerant process communication is

illustrated in Figure 1-5.

In the example shown, process A can be assumed to be the primary process of a

process pair. The backup process, process B, is programmed to accept checkpoint

messages that convey significant changes in the state of the primary process. Upon

any failure of the processor that is executing the primary process, the backup process

can assume execution of the work from the point of the last valid checkpoint.

Dual data paths between the processes assure that checkpoint messages can still be

delivered in the event that one of the paths should fail. Note that process A can send

checkpoint messages to process B either through router X or through router Y. Thus if

a problem should occur on the MFIOB associated with processor 0, process A can still

send its checkpoint messages through the MFIOB associated with processor 1.

Although, for this example, the two processors are represented as being in the same

processor enclosure, in fact the fault-tolerant principle applies even if the primary and

backup processes are running in processors in different enclosures. The ServerNet

architecture is so designed that all X routers are connected together and all Y routers

are connected together. Thus process A can send its checkpoint messages to the Y

router in its own enclosure and be assured that, through a succession of other Y

routers, those messages will reach the backup process no matter where it is.

The complex interconnection of all routers into two parallel structures creates two

independent data-routing entities, each called a ServerNet fabric. The interconnection

of the X routers makes up the X fabric, and the interconnection of the Y routers makes

up the Y fabric.

Note. ServerNet routers differ in the number of ports available for routing. For example, SEBs

use 6-port routers, whereas MSEBs use 12-port router 2s. PMF CRUs in S7000, S7400,

S70000, and S72000 servers use the 6-ported routers, whereas PMF 2 CRUs in S7600,

S74000, S76000 and S86000 servers use the 12-ported router 2s. IOMF CRUs use the 6-port

routers, whereas IOMF 2 CRUs use the 12-port router 2s. Generally, this manual shows 6

ports for a router unless the illustration pertains specifically to hardware that uses the 12-port

router 2s.