Douglas N., The Gilderati have headed home after having spent their $2500 registration fees and the hotel bills with the bar tabs. Storwidth is the paradigm. Here is a report from someone who was there to hear it.
OPEN Q) I got the basic idea of SAN = block data. NAS= Files. The GE folks want to move blocks too. Who will be the first to make files a piece of cake on the SAN. Who will make a Great SANfiler? THe basic Q is why no files on the SAN? Thanks for any answers and or links. How about a SANappliance that can also be a file server?
By Dave Nadig (rating 3.85) on 13:20 09.15.00
Morning Sessions: The Storewidth Paradigm
This morning saw a much more subdued and distinctly tired crowd sitting at the 50 or so tables. But it was indeed packed at 8AM, because we were finally going to talk about something other than optical networking!!!
George Gilder kicked it off with a session introducing us to the Storewidth paradigm.
He asked the challenging question: when bandwidth is infinite and free, what's the critical bottleneck? The Speed of Light. Even if you put a perfect strand of fiber across the pacific, it still takes 60 milliseconds for the signal to get there. The collision of today's technology with light speed is the key issue facing Telecosm companies. Even in microchips, where light speed translates to 9 inches a nanosecond, as we approach 1/4 mile of electrical pathway per chip, the speed of light becomes the limiting factor.
The main solution to date has been to bring things closer together, particularly data. Thus was born caching. And caching moved closer and closer to the chip, because to put data on the network cost you a 100 fold penalty in speed.
But this has changed. Today, the network is actually becoming faster than the processing devices themselves. Last year was the crux point, where backplanes in computers lost the speed advantage, where you now get a 5 fold benefit in moving storage onto the net. Going forward, it's actually advantageous to "hollow out the box" and use the network, not the backplane, as the transmission system. Voila, the Storewidth paradigm. (All very compelling, but I still don't understand where the 'e' in 'storewidth' came from. Doesn't it make you think this has something to do with retail?)
And George gave the nod to Network Appliance as the company pushing this forward. It's worth noting that despite the brevity of this report, all this took about half an hour, as George revved the audience up as if he was the warm up band for the main event. Daniel Warmenhoven, NTAP's CEO, was the headliner.
As a speaker, he was pretty average. He gave a history of how functionality: application serving, routing, print serving, etc. has steadily migrated out of the box, and how it's taken both the bandwidth blowout and lots of proprietary smarts to get storage out of the box too. Popping charts to prove George's claims on speed.
But he takes all this for granted. His worry is something pretty mundane: disk drives. Disk drives are the bottleneck, and their performance has doubled only about once every 3 years. That's where their technologies come in. He then ran through their market and competition, listing the top ten competitors, of whom only themselves and EMC are focussed. All of the other competitors (SUN, Compaq, HP, etc.) are all boxmakers. His belief (which I agree with) is that the focussed competitors will win, and in the last year, only EMC and NTAP have moved up the top 10 list (NTAP is 7th by the way).
Then it was technology time. He regaled us with the failures of SCSI, RAID, and all the other storage acronyms. He did it in one slide in less than 30 seconds, and not a single person in the audience followed it (well, maybe some of the techies, but there was noticeable snickering as he ran through it). Thank god I've actually read their materials in the past (and indeed, MetaMarkets.com uses NTAP products), so I understood the basic idea: build a brand new drive access system that gets rid of SCSI and plugs into ethernet, letting a whole network access storage at the speed of the connections (1 gig ethernet).
Without taking a breath, 5 more people walked on stage to start arguing about what all this means.
Avi Freedmen from Akamai was up first (and he wasn't wearing shorts for the first time) with a presentation called "Why the Internet Sucks and the Flood Won't Come". His points were simple: networks (carriers) don't like each other; the business model is the bottleneck to free bandwidth, not the technology; and the whole network based storage thing doesn't translate into better user experiences. All of this stuff is just about LANs. He was a real bear on the bandwidth boom: the fiber doesn't matter, because the large networks have no economic incentive to peer with each other. Speed might be blistering in, say, the Worldcom network, but as soon as the user tries to get something from the Sprint network, they fall victim to these peering bottlenecks. Of course, at the end of all this, the solution is Akamai (surprise), because they put the data on each network, avoiding the bottlenecks. An incredibly refreshing point of view amongst all this cheerleading. George's comment: "Oh so that's what it's like out there in the real world!"
Avi was a tough act to follow. Drew Major from Novell made the heretical comment that there needs to be intelligence in the middle of the network. In an audience that has adopted David Isenberg's stupid network principle as dogma, he wasn't popular. Nor was he a particularly compelling speaker. He advocated a model where people surf through proxy servers between their machines and the network. The proxies watch the user's stream of requests to the net, managing, adding content, or redirecting based on rules (user selected or otherwise). He used an example of having a United Airlines ad pop up whenever someone surfed to a car rental site to remind them of their frequent flier partner program. This is progress? I'm totally unconvinced, and I don't think anyone else was either. I hope for his sake that this doesn't represent the future of Novell. I'm also not sure what the heck it had to do with the topic at hand. His applause was anemic.
Bill Miller from Storage Networks tried to rescue the enthusiasm in the room. Unfortunately he committed a cardinal sin: personal history. He regaled us with the take of his rise inside the Oil and Gas industry, and how he wrestled with the challenges of making large quantities of data available across the world, and how he and his team invented their version of the storage area network (SAN). He ended by bringing up an interesting point though: storage networks, while cool, solve pretty simple problems: Akamai gets static data to the edges, companies like NTAP get data in the LAN, but the real problem is coherency (version control) in a world where data changes constantly. Didn't give any answers, but it sure sounds like the right question.
Niel Robertson from Exodus was last, and took heat from George for being just about the youngest looking person at the conference. He heads up research for Exodus, but the consensus at our table was that he was about 18. He laid out a framework for thinking about speed: perceived network speed is the sum of Network Latency (hops) + Caching Efficiency (how well you redirect requests to cached data) + Management Efficiency (which he failed to define very well, but it seems to just be the catchall for how well the system is designed). Feeding into the latter, and the biggest speed problem, is CPUs. In the real world, most companies and users can get sufficient bandwidth for today's applications. But his customers' servers can't keep up with the peak bandwidth. This defines the problem Bill Miller brought up: as the Internet sites or any other distributed applications become more complex, system intelligence and speed is critical.
After George took a moment to have an apopleptic fit over Avi Freedman's view of the world, Steven Milunovich from Merrill Lynch kicked off the questions, first noting that the market seems to reward companies the further away they are from the customer: ISP's are the least valuable, storage firms like NTAP and STOR are the most. He asked about the development of open standards for network storage: a kind of opensourcing of Akamai. Needless to say, Avi Freedman was not amused.
He also asked about customer resistance to losing control of their data by outsourcing it to folks like STOR. Bill Miller acknowledged the emotional issues involved in sales, but discounted it as real resistance. Avi, the consumate bear, disagreed, and believes that customers have an irrational desire for control of delivery that isn't going to wane soon. Neil Robertson disagreed, believing that self aware CTO's who look at how well their own organizations deal with data integrity quickly realize that outsource providers are safer custodians.
The panel also took a few minutes to slam EMC, and collectively sing the praises of Network Appliance. The reasoning reverted back to Avi's comments: most major SAN providers just focus on moving the disks out of the computer without actually dealing with the flexibilities inherent in making it networked. Bill Miller from STOR let it be known that they had just gone from being all EMC to all NTAP.
A refreshing break from the alphabet soup of optical networking, and in the end, a compelling support for Network Appliance. Whether you believe the full dream of Storewidth, where your data might live thousands of miles away, or Avi's vision of a LAN storage based, policitally confused network, NTAP seems well positioned in any case.
Comments anyone?
More reports of other sessions at....
community.metamarkets.com |
