Please don't distribute. Outcome of the call I was on yesterday. CeC (from mobile) Begin forwarded message:
From: "Smarr, Larry" <[email protected]> Date: March 24, 2009 4:19:55 PM CDT To: Jim Dolgonas <[email protected]>, Ed Lazowska <[email protected]
, Ed Lazowska <[email protected]>, "Johnson, Kristen" <[email protected] , "David J. Hutches" <[email protected]>, Amin Vahdat <[email protected] , "Dunne, Brian" <[email protected]>, "DeFanti, Tom" <[email protected] , Tom West <[email protected]>, "Harel, Elazar" <[email protected]>, "Orcutt, John" <[email protected]>, <[email protected]>, "Arrott, Matt" <[email protected] , "Krueger, Ingolf" <[email protected]>, <[email protected]>, <[email protected] , Tim Hunkapiller <[email protected]>, Terry Gray <[email protected] , Ed Lazowska <[email protected]>, <[email protected]>, Jeff Gardner <[email protected]>, Paul Gilna <[email protected]>, Mark Ellisman <[email protected]> Subject: RE: AMZN summary
What a great outcome—thanks to all of you for making this happen!
Ed—I think the easiest use case would be to ask Bob Grossman to do h is OpenCloud terasort between the cloud and his other sites as Johns Hopkins, Chicago, and UCSD. He could compare (or not depending on what Amazon wants Hadoop and Sector) and demonstrate that the Amazon cloud works as well as standard racks in universities. Given that Bob won the Supercomputing Bandwidth Challenge last year with this, it would be a powerful demonstration. Let me know if you would like me to contact him…
Secondly I suspect we could use Amazon for a CAMERA extension of our server complex for pre-computes, which are currently constrained to our CAMERA cluster at UCSD. Ideally, Phil P. could work up a Rocks “roll over” to the cloud—Phil, what do you think?
Thirdly, we might be able to put up a small CineGrid exchange with HD and compressed 4k, which could be shown on demand from anywhere attached at 10G to the cloud. Of course, we would use content that was university produced like Donna’s thunderstorm visualizations. To m—what do you think?
Fourthly, we could do backups of DNA array data that Bing Ren is storing in the Calit2 machine room flowing from his instrument in the medical school across campus. We might also extend to our Penn State 454 genome sequencing pipeline. Paul—what do you think?
Just some specific ideas to get us rolling…
From: Jim Dolgonas [mailto:[email protected]] Sent: Monday, March 23, 2009 9:46 PM To: Ed Lazowska; Ed Lazowska; Johnson, Kristen; David J. Hutches; Smarr, Larry; Amin Vahdat; Dunne, Brian; DeFanti, Tom; Tom West; Harel, Elazar; Orcutt, John; [email protected]; Arrott, Matt; Krueger, Ingolf; [email protected]; [email protected]; Tim Hunkapiller; Terry Gray; Ed Lazowska; [email protected]; Jeff Gardner Subject: Re: AMZN summary
let me add a couple of points here, based in part on the dinner discussion with Kurt.
We were talking to the decision makers at Amazon. The guy who said "is there any reason we shouldn't do this" has the ultimate authority for deciding whether to proceed. Kurt confirmed this over dinner.
Our dinner opened up some add'l possibilities, none that would interfere with or detract from what was discussed earlier in the day. These would be more focused on Amazon services for non research purposes.
I think the meeting (s) accomplished all that we could have wanted.
Jim
At 07:45 PM 3/23/2009, Ed Lazowska wrote:
Here's a very brief summary -- please forward if I missed anyone:
We met for 2 hours at UW and 1 hour at AMZN.
Deepak Singh will send me a complete list of the AMZN participants and their roles. I'll pass this along.
Deepak is the biz dev guy for AWS. He's our AMZN point of contact. I agreed to be the point of contact for our group.
Kurt, who organized the meeting, is "Senior Manager" for AWS.
Werner Vogels is CTO.
At AMZN, we were joined by Alyssa Henry, the GM for S3. Also her counterpart who is the GM for EC2. And two networking guys who seemed extremely sharp. Again, I'll pass along all the names later.
It really couldn't have gone better. After about an hour at AMZN, one of their guys said "Look, is there any reason we wouldn't do this?" That was pretty much the end of the meeting. It's a go.
AMZN is extremely pragmatic. They are not interested in solving hypothetical problems. Whenever one of us asked a slightly abstract question, their response was "Tell us exactly what you're trying to do, and we'll tell you whether we can do it."
We agreed that there would be an initial focus on getting the connectivity working. One thing they noted is that various things other than backbone charges are lumped into their data movement prices, so even if we reduce the backbone charges to zero, the costs won't go to zero. For example, they have multiple datacenters on each coast on redundant private rings, and they need to amortize that cost. We agreed that our goals are not "zeroing the cost" but rather (a) ensuring prodigious *achievable* bandwidth and (b) reducing the cost as much as possible.
Clearly we need to do both Seattle and VA. They felt that if we and they both put some pressure on Level 3, we could probably get the VA gap closed. But getting Seattle up first makes sense -- that allows us to get going and to prove some things.
They want us to describe two "use cases." We need to get hopping on that. We agreed that one of them would be a "sexy stupid pet trick" (our famous HD video demo, but done from the cloud). The other should be some sort of large-scale science use. We need to figure out what this might be. It's going to be a while before the OOI is delivering data -- we need something "now." So the ball is in our court to describe these two "use cases."
They seem really intent on viewing this as a growth business, and as a situation where there is a window of opportunity to do something special (before scientists start using the Google/IBM facility or the Yahoo facilities or the Open Cloud; before Microsoft gets its shit together). I tried to make the case for "the business case" and for "doing it now" and they definitely bought it.
I punted dinner to head home and do some wash before flying east; maybe some dinner attendee will have more to report.
Bottom line: this is *clearly* a go. Couldn't have been a better session. At this point, it's up to the network engineers, and we need to describe the use cases.