Wednesday, July 16, 2008

Specialist! Am I still needed?

It is not that often that my lead photo is of an advertisement, but recently I came across the one I have included here. The yacht described in the advertisement was called Tanami, which is the Australian aboriginal word for “never dry”– there have been others who have suggested that the yacht was called Tamami, which is yet another Australian aboriginal word meaning “dry desert” as in Western Australia, there is the Tamami desert. Oh well, who would know an N from an M …

It’s not so much the story of Tanami, and about speculation as to the true interpretation of “never dry” (a reference of course to either the quantity of drinks onboard, or to the water that flowed through the unprotected cockpit in high seas), as it is the reference made of it being the “beautiful sister-ship to Theme and Mistral! All three, fractional-rigged, 40’ sloops, came off the drawing board of Peter Cole and followed many of the lines of the innovative 12 meter America’s Cup yachts of the day – and to me, they each bore a close resemblance to the first Australian challenger, Gretel.

In 1972 the then-owner of Theme, John Hagan, asked if I would like join his crew for a season of racing on Sydney Harbor. Having had no previous experience sailing whatsoever, and with no idea what being a crew member would entail, I quickly responded “Sure!” In the years that followed, I learnt enough to become a for’ard hand, and part of a team responsible for “managing” the sail wardrobe for’ard of the mast. This meant bringing on-deck all the spinnakers, and repacking them after use, as well as changing all the headsails, mostly jibs, as well as the much larger genoas that were called upon when the wind backed off.

In time, I became the sole for’ard hand and developed the special skills this role demanded. Just as sky-divers fold and pack their own parachutes, so I would be the first onto the boat each race day to lay out all the sheets and braces and to make sure nothing would foul or otherwise disrupt our ability to change sails quickly and smoothly.

I came across the advertisement for Tanami as I was going back through previous blog postings, checking out the comments being posted, and “Google-ing” recent press releases. While the memories of my days sailing on Sydney Harbor came rushing back, I was reminded of how racing a yacht demanded similar skills and commitment as building a NonStop server. And what caught my attention, and started this trip down memory lane, were two comments posted by different anonymous readers.

Following the posting “Blades drawn at Mandalay Bay” (June 18, 2008) one reader asked, “there's been talk of the NSVA (NonStop Value Architecture) servers not providing the same levels of hardware fault tolerance as the previous S-series servers, whilst on the other hand NSAA (NonStop Advanced Architecture) - particularly with TMR - provides a greatly increased level of hardware fault tolerance. And how do the new blades servers factor into this discussion?”

In the later posting “It’s Vegas” (June 19, 2008) another reader asked something quite different but interesting all the same “what is this HP Superdomes?? Is it (more) versatile than mainframes???”

Maybe it’s just me, but I found these two questions touch on a re-occurring theme. Is HP building servers that can challenge the mainframe in size, power, and flexibility? And will these servers include NonStop, but possibly today with less “availability” than in the past? As I listen to, and then read about, conversations like this, I continue to wonder about the questions not being asked.

Do we really need to have something comparative to a legacy mainframe, to be viewed as having a “premier server” for today’s modern enterprise? When do we really need a general purpose server (typically how we view the largest server offerings from both IBM or HP), or is the future predicated on our ability to cleverly cluster “specialist” servers ourselves so as to create something more general purpose?

And then what really differentiates the specialist servers – why is one server viewed as better than the other? Is it because of cost? Choice of operating system? Versatility? Or even because of the levels of availability offered? After five decades, why is there still any interest in mainframes, and “almost a mainframe”, types of servers?

When the question of whether the HP Superdome is as flexible as a mainframe is raised, then for many users, the overwhelming answer is yes! If you add a HP NonStop into the mix, working alongside of the HP Superdome, then it becomes an even more compelling alternative. For those companies with the requisite skills in-house, integrating different specialty server / OS / DBMS environments such as Superdome and NonStop into a single homogeneous platform, brings with it considerable advantages.

When the question of whether the NSVA architecture is as available today as has been any previous Tandem, up to and including the Himalaya (both K and S) offerings, then let me be very clear on this point – the answer is an unequivocal yes! For the MIPS based servers, additional steps were taken to maintain the data integrity levels required, but I am not sure that taking this had much to do with making the systems more available. These new servers were always going to be much faster than the previous generation of NonStop servers, but ensuring five nines (999.99% availability) remained the goal.

The move to the MIPS RISC technology for Himalaya made it necessary to take additional steps to ensure the chips themselves were functioning correctly – back in the early ‘90s, this was pretty scary stuff. By comparison, more than a decade later, the need to apply a similar model to accommodate the Intel Itanium technology, proved unnecessary, as the checking and correcting logic you wanted is built into the chip itself. There would be difficulties guaranteeing that arbitration between two independently-functioning chips would provide any additional uptime. NSVA does not represent lesser availability to me but rather, just a smarter way to exploit the true characteristics of the chip.

The picture I have included here is of me alongside a very interesting prototype. It was on display to all HPTF&E attendees who visited the HP stand. There were no attempts to hide it behind a curtain or in other way preclude anyone from giving it a very good look. At the previous event held in 2007, Martin Fink talked openly about his desire to see a hybrid, or clustered specialty servers, functioning with some degree or integration, all housed in the one blade-center. And at this year’s event, there was a fully-functioning embodiment of what Martin had described. Yes, NonStop was running on blades, just as were the other OS’s in the blade-center and NonStop was no less available than any other NSVA servers!

Described at the event as an Engineering Prototype (EP) – not a product, as people were quick to point out, and still requiring considerable input from Product Management before it ever made it into production – it provided generalists with the opportunity to have a cluster of specialty servers delivered to them by HP and functioning right out of the box! All the elements of a “pocket mainframe” with support of a Window’s-based web server, NonStop front-ending transaction processing, and a HP-UX / Oracle data base representing just one possible configuration. Cool! And NonStop at the heart of it all, integrated in a way we have relied on specialists to do in the past. Way Cool, and an incredibly innovative way to exploit the power of blades!

I have to believe that production versions of the EP would be welcome at any development shop and for those users needing a test environment, mirroring their real-world configuration, but for a fraction of the price. Financial institutions that are not yet in the upper ranks with the majors could also benefit from a configuration like this as they leveraged more mature applications for a lot less than their bigger rivals.

Having special skills, whether it’s sailing or building a server, will always be a requirement to make it to the top. There are times where you can really leverage the skills of these people, and these systems, and many companies will highly value the advantage that comes from having such an “edge” despite the cost.

But for others in the community, a highly versatile “out-of-the-box” solution holds many tantalizing advantages and can go a long way to keeping them in the game! I have often written about innovation, and about disruptive technologies that can fuel radical innovation and for me, the productizing of the EP with its support of multiple operating systems can only help push our utilization of NonStop even wider across our companies.

And surely, isn’t this a better answer to the question “what is the future for premier servers?” Adding NonStop alongside of Superdome, for instance, will surely make the resultant server even more versatile than traditional legacy mainframes, and having access to all of this in a single package (utilizing the latest commodity blade technology) just has to appeal to all of us – generalist or specialist. Such a solution will allow us to better focus on the business issues we are so often asked to meet.

I have seen the future, and it’s a bunch of blades slotted into a blade center running NonStop, Unix, Linux and Windows!

1 comment:

David Finnie said...

Richard,

Excellent post, and you've touched on a couple of points I'd like to use as starting points for digressions :-)

Is (hardware) fault tolerance less important these days ? Well - kinda - because the hardware these days is so much more reliable than it used to be. Of course, that doesn't mean it's not important - there is just less focus on it because it has become taken for granted. So... yes, I think offering something that compares well in price/performance is becoming more of an issue - so long as fault tolerance doesn't also start to suffer. HP certainly seems to be doing well on both fronts !

And that leads me to the next issue - we now have the new J-series OS coming out that can run multiple processes simultaneously in a multi-core CPU set up. Bloody brilliant ! But where are the OS APIs that let us take advantage of that (and no, I'm not talking about pthreads from an OSS application - I want this stuff from a Guardian app) ? What about light-weight *Guardian* processes (i.e. threads) running in parallel in a Guardian process, using shared memory and inter-thread comms ? The pthreads library or some variation of it would probably be fine for this - it just has to be available from Guardian, and this is the perfect opportunity.

It is important that we can write applications that make use of the new hardware (and OS). If the future of hardware is multi-core (and it seems to be), then we need to be able to make good use of it without expensive IPC interactions between separate processes.

Sun is doing some amazing stuff with compilers doing analysis of code - working out where it is safe to spin off separate threads to make use of the new multi-core hardware, and automatically creating executable code that does this without any effort on the programmers behalf. Could HP's NSK compilers (in conjunction with the OS) do the same ? Why not ?

Making use of the new performance characteristics of the new chips, in conjunction with HP NonStop's well recognised fault tolerance has got to be a winning combination. Surely ?