[cgl_discussion] Re: Question about how to configure the load shared TIPC server.
jon.maloy at ericsson.com
Wed Apr 2 07:46:48 PST 2003
Guo, Min wrote:
> I am now working on the TIPC test suite development, and I can not clearly understand
>a requirement in the test spec you gave us.
>The req is
>4) Send a message to an existing port name. The port name is
> published from several ports on different processors as
> "load shared". Verify round robin algorithm.
> So it means that we can publish the same port name from different
>processors at the same time, Is that right? If not, why I can start two
>same ports from different machines at the same time? If so, there maybe exist many
>same port services in the same zone at the same time,is that right?
Exactly so. This is useful for load sharing and redundancy.
> And I don't know how to verify round robin algorithm on the different
>processors, so could you please tell me how to configure the topology of
>such network? I myself can not determine which machine acts as
>a service machine in the loadable shared network, so how can I verify the RR
>algorithm?So ,can you give me some suggestions?
I suggest that you publish the port name on all processors, so that the
caller by knowing
the number and addresses of these processors implicitly "know" the number of
publications. If each server then responds by indicating its processor
can easily check whether all processors are accessed, and if it follows
(There may be room for some improvement here: We could add a port name
subscription option for _individual_ publications of the same port name,
not only the binary "available/non-available" as we have now. If the
name event then even
contains the corresponding port/processor address (port identity), we
will have an event
channel with a lot higher resolution than now. I have not yet decided
whether this is a
useful feature or not. Are there any opinions about this ?)
>The content of this email message solely contains my own personal views,
>and not those of my employer.
More information about the cgl_discussion