Valdis.Kletnieks@vt.edu wrote:
Sure, *any* good router vendor can build a router that can handle 100 million routing table entries.
Not. The empirical evidence suggests that aggregate flap rate is proportional to the number of prefixes in RIB. Now, when people talk about route update processing they tend to forget that IGP and routing table updates are easy; matching prefixes against ruting policy filters is not. I would say that a computing device capable of doing today's border routing policies at 1M updates per second is well into the realm of science fiction.
The questions are (a) can they do it for a pricetag of under $2M, and (b) how many will they sell?
The question of profitability of cheese mining on the Moon is irrelevant, because Moon isn't made from cheese. --vadim
Vadim Antonov: Wednesday, May 17, 2000 12:32 AM
Valdis.Kletnieks@vt.edu wrote:
Sure, *any* good router vendor can build a router that can handle 100 million routing table entries.
Not. The empirical evidence suggests that aggregate flap rate is proportional to the number of prefixes in RIB. Now, when people talk about route update processing they tend to forget that IGP and routing table updates are easy; matching prefixes against routing policy filters is not.
I tend to agree, the numbers I threw around earleir were strictly first-order approximations for raw sizes. Second-order would include performance issues and algorithm requirements. I see that Vadim ihas already arrived there.
I would say that a computing device capable of doing today's border routing policies at 1M updates per second is well into the realm of science fiction.
The other item usually included at this level should be bandwidth requirements. What is the size of an update, and how many Gbps load would be generated, at a rate of 1M updates per second? (I don't have the base quantity handy, would someone please provide?)
The questions are (a) can they do it for a pricetag of under $2M, and (b) how many will they sell?
Why $2M? From price ranges in the current market, I would think that they'd have to hit under $200K. Actually, I would have a difficult time convincing clients of anything over an additional $60K. This gets back to my earlier question, how many backbone routers are there (nearest order of magnitude should suffice here)?
The question of profitability of cheese mining on the Moon is irrelevant, because Moon isn't made from cheese.
Typical rough market guidelines are that development cost must be less than 1% of total market size or the project is a non-starter, business-wise. Typical costs for this sort of project are $1M to $3M, over 8 months, with COGm at about $50 (relative to a minimum Number of Goods sold [NOGs] and assuming that it is technically feasible).
On Wed, 17 May 2000 02:04:55 PDT, "Roeland Meyer (E-mail)" said:
Valdis.Kletnieks@vt.edu wrote: Sure, *any* good router vendor can build a router that can handle 100 million routing table entries.
I tend to agree, the numbers I threw around earleir were strictly first-order approximations for raw sizes. Second-order would include performance issues and algorithm requirements. I see that Vadim ihas already arrived there.
OK.. I meant the ability to *store* 100M table entries. Flapping is another story, which just aggrivates the cost problem ;)
Why $2M? From price ranges in the current market, I would think that they'd have to hit under $200K. Actually, I would have a difficult time convincing clients of anything over an additional $60K. This gets back to my earlier question, how many backbone routers are there (nearest order of magnitude should suffice here)?
Exactly. We can all *SAY* we want features X, Y, and Z, but who will actually *buy* them if they cost more? I tossed out $2M as a straw man - it just seemed like a good "you could build almost anything for under that price, but nobody would buy". If the added cost is $50K per box, that increases the number of boxes you can sell, but.. The curve for price versus number sold is probably a hyperbolic (even if my old economics texts drew it as a straight line ;) , whose exact shape will depend *very* heavily on just how much price elasticity there is. And most organizations being what they are, it probably will be pretty flabby until we get to the "You need this level of router or you're screwed" state of affairs, at which point there will be a mad rush to buy them. ;)
Typical rough market guidelines are that development cost must be less than 1% of total market size or the project is a non-starter, business-wise. Typical costs for this sort of project are $1M to $3M, over 8 months, with COGm at about $50 (relative to a minimum Number of Goods sold [NOGs] and assuming that it is technically feasible).
Hmm. 1%? Based on what I've seen for cost estimates for other high-ticket low-volume stuff (mostly mainframe-class computers, etc) I would have guessed 10%. In any case, I think the point is made that we can talk all we want about how we want <insert router vendor name here> to provide a truly high-end router that solves everything, but the reality of the cost pressure does need to be considered.... -- Valdis Kletnieks Operating Systems Analyst Virginia Tech
Valdis.Kletnieks@vt.edu: Wednesday, May 17, 2000 7:17 AM
On Wed, 17 May 2000 02:04:55 PDT, "Roeland Meyer (E-mail)" said:
Valdis.Kletnieks@vt.edu wrote: Sure, *any* good router vendor can build a router that can handle 100 million routing table entries.
I tend to agree, the numbers I threw around earleir were strictly first-order approximations for raw sizes. Second-order would include performance issues and algorithm requirements. I see that Vadim ihas already arrived there.
OK.. I meant the ability to *store* 100M table entries. Flapping is another story, which just aggrivates the cost problem ;)
Dealing with flapping is part of exception analysis, which is part of third-order approximation.<g> At second-order, all you have to do is identify it and determine a rough estimate development cost <g>, as input to the third-order analysis.
Why $2M? From price ranges in the current market, I would think that they'd have to hit under $200K. Actually, I would have a difficult time convincing clients of anything over an additional $60K.
Exactly. We can all *SAY* we want features X, Y, and Z, but who will actually *buy* them if they cost more?
This is a pure marketing issue. Those that absolutely need it, will buy immediately, Those that may need it in the near future, may buy some now, for evaluation, otherwise they will wait for the first price drop. Those that hope to need one, will wait for the price drop before even looking at one, and so on ... this is independent of cost.
This gets back to my earlier question, how many backbone routers are there (nearest order of magnitude should suffice here)?
So far, I've heard 100's and 1,000's, do I hear 10,000's?
Typical rough market guidelines are that development cost must be less than 1% of total market size or the project is a non-starter, business-wise. Typical costs for this sort of project are $1M to $3M, over 8 months, with COGm at about $50 (relative to a minimum Number of Goods sold [NOGs] and assuming that it is technically feasible).
Hmm. 1%? Based on what I've seen for cost estimates for other high-ticket low-volume stuff (mostly mainframe-class computers, etc) I would have guessed 10%.
So would I, in the old days <sigh>. Hardware margins are much thinner now and competitive risk is higher. Regardless of design and function, using known technology, any circut board can be built for about $20US COGm, add another $10 COGm for the case, and $30 for shipping. What's expensive is the amortized development cost, which is fixed, so the development cost burden is distributed over the sold volume. The problem is that it is too easy for garage-shops (HP) to come up with a competitive product, once you've proven the market, if all that you are using is known current tech. What I'm saying is that the market commoditizes too quickly these days, for the old 10% valuation to hold good. This is why hi-tech companies resist commoditization (and why IBM generally leaves a market once it gets there).
In any case, I think the point is made that we can talk all we want about how we want <insert router vendor name here> to provide a truly high-end router that solves everything, but the reality of the cost pressure does need to be considered....
That is one major point, maybe the entire point.
participants (3)
-
Roeland Meyer (E-mail)
-
Vadim Antonov
-
Valdis.Kletnieks@vt.edu