The Reason for Abandoning the UPA Draft

(UPA draft) describes one proposal that tries to announce the prefix unreachable information within the IGP domain. It explores one seldom used LSInfinity feature that is defined in for OSPF(with the value set as the 24-bit binary value of all ones:0xFFFFFF), MAX_PATH_METRIC in "IS-IS Extensions for Traffic Engineering" and MAX_V6_PATH_METRIC in "Routing IPv6 with IS-IS" , with the value set as 0xFE000000(the 32-bit binary, 2^32 - 2^25) It defines also two new flags U/UP, that are used to transfer the unreachable reason within the IGP domain, which does not fit into the role of IGP protocol. The other parts of the document describe how to control the advertisement of unreachable information on ABR, its consideration with the area partition etc, which are first and throughly described clearly and more throughly in (Founder Draft). The proposal doesn't work even in very simple scenario and lacks the key consideration of the the explicit withdraw method. After analyzing these issues, this document recommends the IETF community to abandon the forwarding of UPA draft, replace it with Founder Draft, which is one far earlier and more comprehensive solution.

UPA proposal doesn't work even in simple scenario. Such issues can be illustrated in the following example: In Figure 1, PE1 and PE2 locate in different areas, which are connected via the backbone area 0, via ABR1 and ABR2 respectively. When one prefix P0 on PE2 become unreachable, according to the UPA proposal, ABR2 will signal the UPA, flood it within area 0, with the metric of the summary LSA set to LSInfinity. When such UPA signal reaches ABR1, according to the description in section 12.4.3( "Summary-LSAs") of : "......, if the routing table cost equals or exceeds the value LSInfinity, a summary-LSA cannot be generated for this route." There will be no summary LSA generated for the received UPA signal. It is to say, UPA signal can't cross area border. PE1 which locates in area 1 can't know the P0 on PE2 is unreachable as expected. Any application switchover on PE1 will never happen.

The UPA procedure is broken in such scenario.

defines the LSInfinity feature, but doesn't give any reasonable explanation. Actually, using the LSInfinity value for one prefix to indicate the prefix is unreachable is problematic even in the simple topology that illustrated below:

In Figure 2, the value between the router is the cost of the link. Suppose the Router R4 has two prefixes, one is P0, with the metric set to (LSInfinity-10); another is P1, with the metric set to LSInfinity. R4 advertises these two prefixes within the Area1, with two different LSAs. When R3 receives the LSA for prefix P0, according to the description in RFC 2328, it will treat the prefix P0 as reachable, because the cost within the LSA for prefix P0 is lower than LSInfinity. When R3 receives the LSA for prefix P1, according to the description in RFC 2328, it will treat the prefix P1 as unreachable, because the cost within the LSA for prefix P1 is LSInfinity. But on R3, the total cost to P0 and P1 are all LSInfinity, but one is reachable, another is unreachable, results in Fist contradiction conclusion. When the above LSAs reach to ABR, ABR will take the similar treatment, that, it should treat prefix P0 as reachable, and prefix P1 as unreachable. But according to the description in RFC 2328, "if the routing table cost equals or exceeds the value LSInfinity, a summary-LSA cannot be generated for this route", then ABR can't generate the summary LSA, neither for prefix P0(reachable, but total cost exceed LSInfinity), nor for prefix P1(Unreachable). The routers within area0, R1 and R2 can't reach prefix P0 then. This is Second contradiction conclusion. From the above examples, we can know the LSInfinity feature described in is flawed. Such analysis can also apply to the MAX_PATH_METRIC in and MAX_V6_PATH_METRIC in used for the same purpose. The IETF community should erase such flawed features, instead of exploiting it to signal unreachable information. Then section 3 and section 4 of UPA draft should be removed.

The UPA document specifies that the UPA must be withdrawn when the prefix is reachable again, but gives no explicit UPA withdrawn method. The author declares that stopping advertising UPA is equal to UPA withdrawn, actually it doesn't. The following are the detail analysis: There are at least two reasons to stop advertising UPA: 1) Case A is that the UPA has been advertised for enough time to let the receiver accomplish the related application switchover 2) Case B is the specific prefix is reachable again. For Case A, when the receiver stop receiving UPA signal, it doesn't represent the prefix is reachable again. If there is no explicit UPA withdrawn method, the network can't recover to the normal state before the UPA triggered switchover. Explicit UPA Withdrawn Method is necessary.

Section 5 of UPA draft defines two new flags: U and UP to signal the reason of unreachable information. It says U flag is used to indicate "prefix is unreachable due to the unplanned reason.", the UP flag is used to indicate "the prefix is unreachable due to the planned reason". These flags are not used to influence the SPF calculation of each router, just gives some ambiguous and insufficient reasons(for example, what's the specific unplanned reason?----node failure, link failure, configuration failure? And what's the plan of planned reason") for the unreachable, which is belong to the maintenance information, not required by IGP protocol. Then such information should not be flooded within the IGP protocol. Section 5 of UPA draft should be removed.

The UPA proposal doesn't work even in simple scenario, is based on the flawed feature, lacks explicit withdraw signal, is not appropriate(U/UP Flag, which floods maintenance information within IGP protocol). Then, the IETF community should abandon the UPA draft, put forward to the Founder Draft instead.

The mechanism described in this document does not raise any new security issues for the IGP protocols.

TBD.

None.