I loved Chuck Hollis’ (EMC) blog, and equally enjoyed both Dave Hitz and Tony Asaro’s responses. With EMC, Netapp, and another ESG’er pontificating on an issue, it must be good.
All of the aforementioned folks are smarter than I, but let me give you some different things to consider.
First, I’m an iSCSI proponent. I remember meeting an intern who worked for Andy Bechtolsheim (who’s name I can’t recall, but I do remember he looked 12 and had more brain capacity in his pocket protector than I’ve ever possessed) who was charged with figuring out how to push block data over IP. I think it took me 13 seconds to recognize the implications. It took me a bottle of wine to come back to reality – which is: 1. even if you could use IP networks for block traffic, it will be a long time before people use it to replace core data center technologies, such as fibre channel and 2. eventually it will not only replace fibre channel, but whatever other new specialized connectivity protocol/transport the world comes up with. I shall explain.
I remember within a very short time of the iSCSI initiative announcements people immediately got on one side or the other – there was not middle ground. I was in Silicon Valley when Nishan announced their IP storage router (in an awesome display of bravado, arrogance, and downright stupidity) by telling all that the days of fibre channel were over. They said this in the height of go go valley boom, when there were more than a dozen fibre channel players all getting piles of dough and driving Lamborghini’s. My timing was brilliant, albeit pure luck. My cell phone rang with abandon, and everyone from the Wall St. Journal to the Financial Times was asking what I thought about Brocade going out of business. I remember Ashok Kumar from Piper Jaffray getting on CNBC or some equivalent and discussing the end of the fibre channel market. I believe it was early 2000.
Billions of dollars of value were traded in short order. I got a call from Brocade and was asked to come in. I had never met Greg Reyes, but knew most of his team, and certainly knew of him. They wanted my take – and I gave it to them – death is not near, but it is inevitable. I haven’t changed my tune. Greg told me point blank “you can’t put block data over IP”. I suggested he not say that. I suggested he instead say “Brocade builds block storage networking equipment designed to enable sharing of high-performance, mission critical storage assets across multiple servers. In order to guarantee both performance and more importantly – reliability – we do that today using fibre channel. We don’t care about fibre channel per se, nor any other specific protocol or transport – we care about providing the most reliable block networking services available to the thousands of customers we have around the world. If IP, or any other technology advancement comes along – as things always do – Brocade will adopt them and offer them to our customer base.” In short, I suggested he tell the world this was no big deal, and if it mattered eventually, they would offer it.
He didn’t. He said “you can’t do block data over IP, you simply can’t”. He was wrong. The message was wrong, and this single statement did more to polarize the factions than anything else in this debate. (Brocade, of course, has subsequently adopted this position, rightfully so, as have most all fibre channel players. Once you remove the emotion and realize no one is throwing out the baby with the bathwater, cooler heads prevail.)
So Chuck’s observations are hard to argue with – and Tony’s right in stating that it’s really a matter of perspective. Since I prefer my perspective, and feel I’m the most qualified to share it simply because I was there on day one (with no regard for competence, mind you), here’s what is going to happen and why.
History is littered with proprietary solutions that are developed to solve very specific tactical problems that turn into markets which last long enough for some to get rich, others to die trying, and then Darwinism does its thing and some sense of reality is applied to that market for that period of time – and to the victors go the spoils. Then, eventually, the reason for that market no longer will exist – it will be replaced with the next thing, be commoditized and standardized, and the original proprietary solution will cling to life as long as possible with arguments like “if it ain’t broke, don’t fix it”. Water always finds the lowest point, and so do technologies. Remember MassBus (I know, I’m old.), SDI (then ESDI), IPI and a zillion other block interconnects and protocols which existed in the day – only to ultimately give in to SCSI for a protocol (and a transport) which won because commoditization and Darwinism do really work. How about Token Ring, DECNET, Banyan, x400, blah, blah, blah. IP won. Ethernet won.
Things don’t eventually win because they are better in the short term – they win because they are more ubiquitous. They become more ubiquitous because they take advantage of commoditization economically and VALUE can be generated because of that shift. IP killed the other networking technologies over time. Once we gave in, life didn’t end – because only those who fought IP died. Those who recognized the inevitable were able to plan accordingly and either jump on the bandwagon and fight to own the commodity market or moved their value up the food chain above the commodity fray. Those who hoped it would go back to the old days or defended the fact that they were better, slowly and painfully died. Banyan might have been better, but now they give me phone numbers on-line. (Switchboard.com).
The reason iSCSI wins long term is because the ubiquitous market elements are in place (IP is everywhere and Microsoft has added it to the O.S.), and the cost is zip. Betting against that is like being the guy at the party talking to the prettiest girl when Tom Brady and Derek Jeter walk in and still thinking you are the guy she’s going home with. Theoretically possible, but I don’t like the odds.
So, since you all now totally agree that I’m right (eventually), from a more pragmatic perspective, here’s why you shouldn’t care. 99% of the time today, iSCSI should not be used to replace fibre channel. It should be used to augment and enhance all the brilliant things fibre channel does. 90% of the servers in fibre channel SAN shops are not connected to the fibre channel SAN. If the block storage network is good enough for 10% why isn’t it good enough for the rest? iSCSI enables companies – from small to enormous – to extend the value they received by having a FC SAN to their internal mass server/connectivity market.
What surprises me is that the EMC’s/IBM’s/HDS guys who are in the core haven’t been the ones leading the push toward iSCSI. The name of the game is “filling up the box” and what better way to help a customer spend more money with you than to show them how they can take control over the data on 90% of the servers vs. 10%? To NetApp’s credit, they did figure that out a while ago. They ship a huge percentage of their Filers with iSCSI connectivity as well – though I don’t know how many use it nor what percentage of the systems are used for block vs. file. Regardless, the other guys own the core, and if they don’t use iSCSI to add connectivity and capacity to the core (which steals share from all their downstream competitors, fyi), it’s naive to think that someone else isn’t going to create a parallel iSCSI network. Once that happens, users will realize that this stuff is pretty good, and as a matter of fact, some of the “cheap” arrays I bought for my cheap iSCSI SAN happen to work way better than my more expensive core stuff. Oh oh. Do you really believe none of Equallogics’ thousands of customers are big enterprises – or better yet, big EMC enterprises?
The argument that the SAN guy doesn’t want to deal with it is also inevitably destined to fail. The backup guys still don’t like to deal with desktops and laptops – and hate the thought of having to deal with remote office data – because their world is held together with spit and chewing gum. Adding more complexity to their lives is not something they are interested in – but it doesn’t matter. Whether its mandated by law or becomes the good corporate practice it should, the market will force those attitudes to change, as it will with iSCSI.
The one point I think Chuck nailed was the services piece, but not around requiring services to move from FC to iSCSI. Rather, the services – which will ultimately lead to automation technology opportunities, lie in moving from DAS to SAN – or DAS to NAS – regardless of protocol or transport. No one seems to be arguing that networked storage and disaggregated components is not better than tightly coupled “systems” at least.
So, the good news is iSCSI is righteous, and it is happening. The bad news is that the society we live in (IT) would benefit greatly by a more rapid adoption of this, and many other commoditized/standardized technology implementations but tread slowly because they aren’t being shown the strategic potential they could gain by doing so by their big vendors – who have a natural vested interest in selling more of the same stuff they know how to make, sell, and support.
Darwinism always works, but it takes a while sometimes.
Tag:
Add to 
Bookmark Murdok:
Steve Duplessie is the author of the “Steve’s IT Rants” blog, and the founder and Sr. Analyst of the Enterprise Strategy Group.
