Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[RPRWG] Ring/node availability




Hello,

From the presentation slides of the WG meeting in March I learned about
defined availability requirements for RPR networks (e.g. slides from S.
Silvus).

For given ring availability requirements it can be interesting to look
at the needed node availability performance.  For the calculation of the
availability maybe you can find the simple formulas in Section 3 useful:
http://www.lkn.ei.tum.de/~ds/papers/amcis2000.pdf
The network model would apply - as far as I understand - to an RPR with
wrapping.  The availability (or reliability) formulas provide lower
bounds, since only the one-failure cases are considered (however the
most significant) and section availabilities are assumed equal (e.g. set
to the lowest value).

Consider the availabilities of the RPR-node (L1/L2) and the internodal
links only. Take the values from S. Silvus:
* R_all = 99.999% (Network availability)
* internodal links with 40 km length
Assume for each link kilometer a mean time between failures of 1000
years and a mean time to repair of 4 hours.

Then for 8 and 32 nodes the RPR-node has to have a max. outage time of
39.38 and 9.69 seconds, respectively.  This would impose very high
reliability requirements on the node design.

Taking 99.999% as a two-terminal-availability, a max. outage time of 
approx. 2.6 minutes would be required for a node.

Best regards,
Dominic

PS: These are the maple formulas for the calc based on the paper:
> r_1 := (R_n * R_h * R_f * R_s * R_l^2)^n:
> r_2 := n * (R_n * R_f * R_h)^n * (R_s * R_l^2)^(n-1) * (R_s * (1-R_l^2)+(1-R_s)):
> r_3 := (n-2) * (R_n * R_f)^(n-1) * R_h^2 * (R_s * R_l^2)^(n-2) * (R_n * (1-R_f)+(1-R_n)):
> R_f_target_all := solve(R_all = r_1 + r_2, R_f):
> R_f_target_e2e := solve(R_all = (r_1 + r_2)/(R_h^(n-2)) + r_3 , R_f):
> R_n:=1: R_h:=1: R_l:=1: R_s := (1000*365*24 / (1000*365*24 + 4.0))^40:
> R_all:=0.99999:
> n:= 8:  (1-R_f_target_all)*365*24*60*60;
39.38
> n:= 32: (1-R_f_target_all)*365*24*60*60;
9.69
> n:= 8:  evalf((1-R_f_target_e2e)*365*24*60);
2.6252
> n:= 32: evalf((1-R_f_target_e2e)*365*24*60);
2.5612

--
Dominic Schupke 
Institute of Communication Networks, Munich University of Technology
Building 9, 1st floor, Room 1910, Arcisstr. 21, 80290 Munich, Germany
Tel.: +49 89/289-23511, Fax: +49 89/289-63511
mailto:Schupke@xxxxxxxxx, WWW: http://www.lkn.ei.tum.de/~ds