1.029: atmos. with, and without, the sphere
I'm just back from vacation cruising several Italian, French and Spanish ports aboard the Wind Star on the Mediterranean with my wife. It was a relaxing, multi-cultural Adventures Afloat trip arranged by her employer (Elderhostel), a not-for-profit who specialize in educational travel and learning opportunities. With a foundational belief that learning is an integral part of a healthy and fulfilling life, the organization offers its unique Adventures in Lifelong Learning to anyone who is interested - at an exceptional value! So, if you're looking for a travel programme with more than just the usual tourist trap visits, I encourage you to visit their web site and/or order their free catalog.
Oh, and don't let the name fool you: participation is quite diverse, and you'd better be in good shape or you might just get left behind.
Anyway, being on such a trip with my wife, I wisely avoided all things work for the duration.
Preserving the atmosphere, you might say.
But so much has gone on in the past couple of weeks, I thought I'd take a stab at connecting some of the key sights from my cruise with a few of the more interesting events of the past week or so.
So let's have a little fun. Shall we?
italy and napoleon
Our cruise departed from Rome (actually, the nearby port of Civitavecchia), and visited lots of really cool sites. The first port of call included a shore visit to Napoleon's exile home on the Italian island of Elba.
The whole theme of world dominance followed by crash-and-burn exile reminded me of Hu Yoshida's ill-fated attempts to establish technological independence of Hitachi's approach to storage virtualization (here and again here).
I say "ill-fated" because not only has Hu once again demonstrated his abject lack of true understanding of his competitors' products, but he has also riled up the wrath of Sir Whyte in the process. BarryW has not only rebuffed Hu's original misrepresentations of SAN vs. Storage virtualization, he has now also embarked on what looks to be a multi-episode tutorial on how SVC really works.
Good sport, that. But a word of advice, BarryW - you may indeed be embarking on a Lifelong Learning adventure of your own!
monaco and prince albert ii
On US Election Day our port of call was Monaco. And our excursions took us to visit the the Monaco Cathedral, the Prince's Palace and of course the Monte Carlo Casino ("Burke. Barry Burke." I said, as I handed them my passport).
You know, you've just gotta admire a guy who is the head of state of the world's most densely populated sovereign country. He's not flashy nor seemingly authoritarian - he just seems to quietly rule his (quite wealthy) country of 36,000 with a benevolent hand. But several of my cruise-mates were totally surprised to witness that the good Prince chose to hang out with all the American visitors and ex-pats down at the Stars'N'Bars on Election Night 2008 (and yes, apparently he was rooting for Obama).
By the way - it is interesting to note that citizens of Monaco are not even allowed to enter into the Monte Carlo Casino. Turns out that the country owns the majority stake in the Casino, and laws prohibit the conflict of interests of owners gambling against themselves.
Seems somewhat similar to the twisted knot that StorageBod has lured NetApp into with his WAFL is a Platypus discourse a couple of weeks ago.
nice is nice. benchmarketing - not so much.
One of my favorite stops was Nice, and it was actually very nice. That's Nice, as in the resort city in Provence-Alpes-Côte d'Azur (aka, the French Riviera). Even though it rained cats and dogs the whole time we were there, the city just has the feeling of a warm embrace, and most of us agreed we'd like to go back for a longer visit.
Not so nice, on the other hand, was the reaction to my previous post about benchmarketing. While most managed not to attack me personally, almost all missed my point altogether. At the core, my complaint isn't about science experiments at all - it was (and is) all about the continued use of a benchmark suite which has no bearing on reality as a comparison tool.
And not only because the configurations that vendors run them on are ludicrous. Which, indeed, they are, as one end-user noted. But even he's been misled - he runs the SPCs on "real" configurations, and I assert he still gets misleading results.
Why? Because indeed, as IBM and ESG admit, benchmarks like the SPCs do not capture the workload dynamics that result from running multiple different applications against a single array.
And whether or not those multiple applications are each running in a VM on a single host, or if they are each actually running on different physical servers truly makes no difference. The bottom line is non of the SPCs capture the true essence of multiple dynamic workloads on the same array.
The simple fact is that the SPC's don't present a workload that looks ANYTHING like what an array like Symmetrix routinely supports.
In the real world, an array like Symmetrix is rarely afforded the luxury of supporting a single application running on a single host - it happens, but it's not the mainstream normal use case. No, in the typical Symmetrix use case, I/O requests from multiple different applications running on multiple different hosts arrive on each and every front-end port, and these requests are for a very disparate group of LUNs or Volumes. This chaotic stream of requests arriving at a given front-end port is effectively randomized across all the target back-end devices, even though the net result may well be sequential requests to each LUN device. And between each I/O request to a given LUN there is typically some sort of delay while the host application processes the retrieved data; meanwhile some other application will probably post its I/O request to some other set of LUNs or Volumes.
And unlike any other array on the planet, the Symmetrix will sift through all these chaotic I/O streams and dynamically adjust its queuing priorities, pre-fetch and cache aging algorithms to deliver the fastest possible response times for each application.
The SPC benchmarks look nothing like this. There are relatively few LUNs required by the tests, and these are typically spread across as many spindles and the benchmarketing vendor can support. All I/O is generated by a single host, and the I/O requests are created in a specific, stochastic order, sent as quickly as possible down each I/O port. Each LUN receives a predictable percentage of the I/O requests, yet these are artificially randomized such that cache hits are minimized to ratios well below the real-world norm for most applications.
In essence, the SPC's measure nothing more than the efficiency the disk I/O path of a cache miss, which tells you nothing about how efficiently and effectively the storage array delivers cache hit I/Os.
And as a result, these SPC benchmarks are of no use when trying to compare how an enterprise-class array will handle real-world workloads. Maybe they're sufficient for less intelligent arrays with less advanced algorithms and limited caching capabilities (IBM's SVC and DS8000 come to mind), but these simplistic benchmarks are nothing but misleading for truly advanced storage arrays.
Does the SPC care? I think not. Nor does the competition want to do anything to change the status quo - witness the fact that nobody (and I mean NOBODY) has stepped forward in the past year (not to mention few weeks) to work with Dr. Kartik on a better benchmark.
barcelona and els quatre gats
The last stop on our cruise was the city of Barcelona - truly a city of energy and excitement. From the revolutionary architectural works of Catalan Modernisme pioneer Antoni Gaudi, to the hangout of famed artist Pablo Picaso, to the awe-inspiring Temple Expiatori de la Sagrada Família, Barcelona is truly a wonder to visit (mind your wallet as you stroll on La Rambla, though - Barcelona is the training grounds for pickpockets and La Rambla is where they get their masters degrees).
One of the more memorable stops in Barcelona was dinner at the famous (and fabulous) 4Cats Restaurant - a Bohemian-esque throwback where Picaso held his first solo art exhibition, and where he also spent many the evening wining and dining the worlds' challenges away with his contemporaries.
The proprietor of the 4Cats was also a visionary, for he managed somehow to retain a number of Picaso's sketches, many of which adorn the walls of the dining room along with other originals and prints of the famous artiste.
Visiting the city where Gaudi and Picaso made their marks was an excellent precursor to Monday's Atmos announcement. Both men were paragons of innovation; both saw new possibilities where other saw only a country devastated by years of war, tyranny and fascism. And both led the way to a new era of artistic and architectural excellence.
As undoubtedly will Atmos.
Sure - there's lots of hype around Atmos, but hey - what else do we have to get excited about this week?
Seriously, though, I am somewhat taken aback by the reaction of some to the announcement, especially those that haven't taken the time to try to understand what Atmos really means. Granted, everyone will see anything new through the filters of their own localized realities. But some of the reactions are just plain silly.
- Atmos isn't anything like a WAFS - if only because WAFS could never scale to the scope of Atmos. If it could, heck - the smart guys who have been building Atmos would have built it upon WAFS. And Google would have built their global infrastructure on WAFS. Heck, even Akami doesn't use a WAFS to get their cached objects close to the users. No - a wide-area file system simply isn't robust enough to handle the scale and scope of the world Atmos was designed for. Nor could any existing WAFS handle the multi-tenancy, secure storage domains and dynamic policies that are the foundation of Atmos.
- And speaking of policies, some seem to have missed the fact that Atmos policies are established BEFORE an object is stored, not defined afterwards. In this manner, the storing application truly has to have no knowledge about the policies - just insert the objects, and let the system figure out how the established policies apply to the object. Transparently. UPDATE: Steve Todd's post offers more Under the Hood insight into Atmos' policy engine.
- As to the assertions of "proprietary," Atmos is as open as it gets - you can use SOAP, NFS, CIFS and several other standard interfaces to insert and retrieve objects. Sure, there's a lot of unique features in Atmos that aren't present in any other storage platform today - but isn't that always the case with something that's "new"? And EMC has announced that you'll be able to buy the SW to run on your own hardware, should you choose - so where's the "lock-in"?
Interestingly, EMC faced similar complaints about Centera when it was introduced in 2002, but even still there are today over 200 applications that are integrated with Centera using one or more of its APIs. And EMC led the way to define XAM, and Centera is the first to support it. So enough with the conspiracy theories already...
Now, granted, most of the nay-saying over Atmos is coming from would-be competitors who now will struggle through the stages of denial and deferral until they finally come to grips with the value proposition of the technology and enter into acceptance and finally come around to attempt to deliver some form of competitive response. We've already seen the attacks from almost all of the expected fronts, in fact. Hopefully, the few we haven't heard from yet are taking the time to let this all sink in first before they react- a wise move, IMHO, for there is lots more to Atmos than the early nay-sayers have even begun to understand.
Still, many others see fit to attack Atmos because they just can't see any use for it - it has no recognizable place within their own storage domains. And if they don't understand the need for it, then by gosh, nobody needs it. One user even laughed that "nobody" buys storage in increments of 120TB, cloud or otherwise.
It's all a matter of perspective, I guess.
Just because not everyone needs a Google file system doesn't mean that only Google does. Atmos seeks to blaze a new approach to "cloud" storage (oh how I hate that term), to create a global storage platform that is not only cost-effective to install and grow, but extremely efficient to operate as well...set-it-and-forget-it cloud storage. And trust me, if your business thinks in petabytes or even exabytes of unstructured data, you're already looking for a totally new storage paradigm, because nothing - and I mean NOTHING - built on current commercial file systems or databases will handle that kind of storage.
Finally, perhaps the biggest laugh I got from Monday's announcement coverage was one writer's back-handed assertion that Atmos threatened EMC's core Symmetrix business. I chuckle because there's not a lot of Symmetrix going into environments to support shedloads of photographs of drunken teenagers, and honestly, that's not really the Symmetrix target market. Nor is Atmos architected for the kind of random I/O workloads typically deployed on a Symmetrix. Unlike the confusion IBM has created between XIV and the soon-to-be-retired DS8000, Symmetrix, CLARiiON, Celerra, Centera and Atmos each offer value propositions optimized for very clearly defined market segments and use cases.
I laugh again just now as I write this - a silly thought just jumped into my mind:
How the heck would you design an SPC benchmark
to measure the performance of Atmos?
I dunno. But I bet someone will try.
PS: apologies to the Storage Architect...see, I didn't do a tear-down in my coverage of Atmos like you thought I would. As soon as I start getting predictable, I'm going to try to change!