With apologies to Jaya, and others, as Victor and I go a bit off topic.
For those without a copy of Doyle's "Routing TCP/IP Volume I" book (highly recommended), my first edition copy has on page 466:
(OSPF)
"Virtual links add a layer of complexity and troubleshooting difficulty to any internetwork. It is best to avoid the need for them by ensuring areas, particular backbone areas, are designed with redundant links to prevent partitioning. When two or more internetworks are merged, sufficient planning should take place beforehand so that no area is left without a direct link to the backbone.
If a virtual link is configured, it should be used only as a temporary fix to an unavoidable topology problem. A virtual link is a flag marking a part of the internetwork that needs to be reengineered. Permanent virtual links are virtually always a sign of a poorly designed internetwork."
I suspect, the foregoing, is what Victor has in mind with his reference Doyle.
I too mainly agree with Doyle's opinion, but I don't believe even what Doyle writes precludes usage of virtual links in all cases. I also draw an important distinction between planning and initial designing, and between making something work within existing constraints, which I believe a careful reading of the above also supports.
Where I might differ with Doyle, and likely Victor, is on the question of whether a virtual link could be used permanently. Doyle mentions "temporary fix", but one should consider the actual risk of the "temporary fix" vs. the cost of implementing a better "permanent" solution. For instance, if we really wanted to insure Paris was its own OSPF area, we could connect the 2600 directly to the area 0 backbone, avoiding the need for a virtual link, unknown how difficult or costly this would be within the existing physical topology. We also need to consider the size of this area, which considering there currently isn't local Paris routing, would infer small.
Another better question might be, do we even want a 2600 to be an ABR, and Paris its own OSPF area, regardless of direct connection or virtual link? This, to me, is even more crucial. I would lean toward just including Paris as a member of the OSPF area it's attached to, but I since don't know all the other considerations with the current OSPF design and the redistribution model, I couldn't recommend either approach. I've only suggested possible "how to's".
Victor, as to your concern about sounding pompous, perhaps only in the degree of something being "fundamental". Much of network engineering, and its applied art, is based more on "it depends" than on "always" or "never".
We might also differ where you take Doyle as "gospel of routing" (perhaps Cisco too), I'm more of a "trust but verify" person. I often find, even when the information was accurate at the time when it was published, with progression of technology, it may no longer pertain or have been superseded. Even when accurate, you need to fully understand its application; often demarcations can be nebulous. Examples of the latter: How many hosts should a subnet contain? How many routers should be in an OSPF area? So, if you say, this book or vendor or industry says NEVER have more than 254 hosts in a subnet or 30 routers in OSPF area, that might indeed sound pompous even if correct for the instance at hand. Personally, I haven't found any book, any vendor or even the whole industry omniscient.
