Gigapop Geeks BOF
“The forum where the Geeks can speak out”

Welcome!

Hosts:

Dan Magorian, MAX

Brent Sweeny, IU/Abilene Noc,

Standing in for Jon-Paul Herron

Tonight’s Discussions

Dave Farmer, UMN:   “Sub Light Paths”

Brent Sweeny, IU:    “How are gigapops/rons doing connecting to NLR?

Dan Magorian, MAX,:   “Networking clubs: Are they really a good idea?”

Networking Clubs: Are They a Good Idea?

Alert: Nontechnical topic!

We’re all used to the I2 “membership club” approach to high-performance networking:

Institutional I2 memberships, institutional gigapop memberships plus gigapop bandwidth fees cost-shared. With line cost, are the 3 costs.

Works well for large institutions, well able to afford. And people enjoy feeling like members.

“Change is everywhere, everything’s up in the air”. So why not examine the membership approach altogether?

But how well do they work for small folks?

Typically, smaller folks might pay US$ over $110k/yr: $47k/yr I2, $40k/yr MAX, $24k/yr 100M gige circuit. Over half of this are memberships.

MAX just had our 5^th drop of a smaller institution: Fujitsu Labs.

Is this a problem financially to us? Not at all: have high growth rate of big folks, esp Fed agencies.

But it’s becoming obvious that standard pricing model doesn’t work well for small folks

Important to remember that most of growth is in smaller folks: we already have most of the bigs.

By contrast

A commodity 100M connection might cost US $12k/yr plus $24k/yr line, 1/3 of the I2/gigapop cost. No membership, no networking club to join.

Does I2 deliver 3x the “commodity internet” value, esp now that everyone’s internet is faster, and uses similar technology? Also, “world-wide I2” peerings with R&E nets may be getting slower, so there may be if not convergence, then at least lower differential. Yes, there are still differences: no IPv4 multicast, no native IPv6.

Plus, many folks in the community would like to not have to “pay twice for” I1 and I2. This could mean many things technically.

NLR amplifies this effect

Membership is shifted to the aggregator, who then has to “resell” it. Some claim costs 50x more than I2. Aggregators end up assuming lot of debt using “upfront” vs “pay as go” funding model. Works well for Cisco and L3, does it serve the community well?

Club memberships are enforced in I2 via prefix acceptance to prevent daisychaining. But lambdas don’t have prefixes; opaque, really unenforceable. Rumor has it people are already buying lambdas for others, circumventing the club memberships.

So lot of thought going into merging membership models right now. But is anyone examining whether is good approach at all?

Current thinking is, “Who Cares?”

Adage right now is, “A large fraction of I2 members have no research agenda. A ‘big boys club’ that excludes small fry is fine”.

I disagree completely with this logic. Who says that most members have no research agenda? Small folks might have even stronger agendas than big folks. And many might have modest agendas, middleware, AAA initiatives, etc.

A lot of the perceived value of I2 comes from how many folks are attached. I claim I2’s success has come about from moderate even-handed pricing and attention to people’s needs.

IMHO, this thinking that seems to come from the NLR side is wrong-headed and basically elitist.

Why am I wasting your time talking about membership stuff that geeks don’t control?

Because I think the current logic is wrong-headed. As they rethink membership, I think they need to make it more affordable for small folks rather than less. By lowering the I1/I2 gap to 2x from 3x, more folks can afford to be a part of I2/NLR and there’s more research, not less.

Research agendas are not proportional to the size of the institution nor their ability to pay. There are lots of small folks out there with new ideas who would be good members of the club if we don’t price them out with a lot of elitist membership policies based on old “R1” model.

What do folks think?

Thanks!


	Welcome!
	Hosts:
	Dan Magorian, MAX
	Brent Sweeny, IU/Abilene Noc,
	Standing in for Jon-Paul Herron


	Dave Farmer, UMN: “Sub Light Paths”
	Brent Sweeny, IU: “How are gigapops/rons doing connecting to NLR?
	Dan Magorian, MAX,: “Networking clubs: Are they really a good idea?”


	Alert: Nontechnical topic!
	We’re all used to the I2 “membership club” approach to high-performance networking:
	Institutional I2 memberships, institutional gigapop memberships plus gigapop bandwidth fees cost-shared. With line cost, are the 3 costs.
	Works well for large institutions, well able to afford. And people enjoy feeling like members.
	“Change is everywhere, everything’s up in the air”. So why not examine the membership approach altogether?


	Typically, smaller folks might pay US$ over $110k/yr: $47k/yr I2, $40k/yr MAX, $24k/yr 100M gige circuit. Over half of this are memberships.
	MAX just had our 5^th drop of a smaller institution: Fujitsu Labs.
	Is this a problem financially to us? Not at all: have high growth rate of big folks, esp Fed agencies.
	But it’s becoming obvious that standard pricing model doesn’t work well for small folks
	Important to remember that most of growth is in smaller folks: we already have most of the bigs.


	A commodity 100M connection might cost US $12k/yr plus $24k/yr line, 1/3 of the I2/gigapop cost. No membership, no networking club to join.
	Does I2 deliver 3x the “commodity internet” value, esp now that everyone’s internet is faster, and uses similar technology? Also, “world-wide I2” peerings with R&E nets may be getting slower, so there may be if not convergence, then at least lower differential. Yes, there are still differences: no IPv4 multicast, no native IPv6.
	Plus, many folks in the community would like to not have to “pay twice for” I1 and I2. This could mean many things technically.


	Membership is shifted to the aggregator, who then has to “resell” it. Some claim costs 50x more than I2. Aggregators end up assuming lot of debt using “upfront” vs “pay as go” funding model. Works well for Cisco and L3, does it serve the community well?
	Club memberships are enforced in I2 via prefix acceptance to prevent daisychaining. But lambdas don’t have prefixes; opaque, really unenforceable. Rumor has it people are already buying lambdas for others, circumventing the club memberships.
	So lot of thought going into merging membership models right now. But is anyone examining whether is good approach at all?


	Adage right now is, “A large fraction of I2 members have no research agenda. A ‘big boys club’ that excludes small fry is fine”.
	I disagree completely with this logic. Who says that most members have no research agenda? Small folks might have even stronger agendas than big folks. And many might have modest agendas, middleware, AAA initiatives, etc.
	A lot of the perceived value of I2 comes from how many folks are attached. I claim I2’s success has come about from moderate even-handed pricing and attention to people’s needs.
	IMHO, this thinking that seems to come from the NLR side is wrong-headed and basically elitist.


	Because I think the current logic is wrong-headed. As they rethink membership, I think they need to make it more affordable for small folks rather than less. By lowering the I1/I2 gap to 2x from 3x, more folks can afford to be a part of I2/NLR and there’s more research, not less.
	Research agendas are not proportional to the size of the institution nor their ability to pay. There are lots of small folks out there with new ideas who would be good members of the club if we don’t price them out with a lot of elitist membership policies based on old “R1” model.
	What do folks think?