OK, I realize this is a very old-school attitude, but I’m at least partly serious — the process of upgrading from an OK old computer to a great new computer can still be pretty eye-opening.

I recently upgraded my primary home office desktop to a screaming-fast Maingear F131 workstation, and it’s a huge improvement.  I wrote a review of it on the Maingear site.

I realize that Maingear is primarily known for high-end gaming desktops, but I’m no gamer.  Instead, I am what used to be called a “power user” – a guy who uses a lot of applications that eat up a lot of computing power.  In my case, it’s usually for some type of economic and statistical analysis on large and unwieldy datasets.  My review gets into all the specifics of the new machine and my experience with Maingear (I recommend the product and the company highly), but here I want to talk about the economics of desktop computing.

For as long as I’ve been buying computers (my first was 28 years ago), I’ve believed that computers are ridiculously cheap, and that buying the best one you can afford is pretty much of a no-brainer.  I’m amazed at how few others share this view, so let’s start with a justification based on performance and productivity.

Imagine I can choose to buy one of two computers — standard and high-end.  Let’s say high-end saves me ten minutes of lost productivity per day (more on that below).  That’s 50 minutes per week, or more than 40 hours per year.  If I replace my main computer once every three years, that’s more than 120 hours saved be choosing high-end over standard.  The value of my time (whether calculated on billing rate, take home pay, or any other reasonable measure) justifies paying a lot more than any real-world premium for a high-end computer.

In reality, this estimate is very conservative.  I hear experts, pundits, and defenders of the conventional wisdom howling that this analysis makes no sense if “all you do is email and Web browsing and a few spreadsheets” or whatever.  I submit that these experts have not done a lot of side-by-side comparison testing.  Just to pick a simple real-world example, starting Excel and opening a relatively simple one-page spreadsheet can take 2-3 seconds on a standard new computer, and takes less than one second on my new high-end computer.  Same with Word, and starting up Outlook, or opening a browser, etc.  (And yes, I’ve done the same kind of tests with a standard Mac and a high-end Mac, and I’ve also tried Thunderbird and OpenOffice on Windows; the results are comparable.)

For me, the real benefit is not just saving two seconds a few hundred times a day.  I also do some compute-intensive analysis from time to time in Excel and data-intensive work in both Access and SQL Server.  A not-especially-huge-and-complex spreadsheet I’ve used in recent economic analysis projects takes more than two minutes to open and recalculate on a decent new computer.  On my new Maingear box, that process completes in 45 seconds.  That’s a huge difference!  Yes, I can get up and make coffee while I wait two minutes, but when we’re in the throes of analytic work, we use many similar tools many times a day; even I don’t drink that much coffee.  The performance difference with Access is even more pronounced.

Yes, this post has all been about personal productivity and not about using this new system as part of a Digipede Network grid; I’ll have more to say about that another day.  But suffice it to say that a network of potent desktops like the Maingear F131 would make a very powerful grid indeed.

One of the most painful and joyful events in the life of a software company is the release of new software.  While this seems like an obvious statement, let me just say to all my friends who are NOT in the software business — you have no idea.

Many customers, prospective customers, and industry observers shrug and even smirk at a press release that “merely” announces the release of a new version of an existing product.  (Smirk away — here’s ours.)  But it’s gratifying to receive congratulations from those who actually understand this process (thanks, friends at Microsoft!).

So now that the apparently-endless cycle of build and test is over, and the last (known) snafu has been fixed (how the %^$&$! did we put an uninstallable version of our SDK out on our community site?), we can take a deep breath, step back, and discuss what this release means — to our customers.  Because, as our press release says, this is a software release entirely driven by our customers.

From the beginning, we set out to make the Digipede Network “radically easier to buy, install, learn, and use” than any other distributed computing platform.  Reviewers say we’ve done that, and customers tell us they can come up to speed quickly with our software.  Ah, but once a customer comes up to speed quickly, that customer gets ideas!  “Why does Digipede use all the cores on each compute resource? Can we reserve one or more for other uses?”  “When I try to delete thousands of jobs at once, weird things happen — are you guys just idiots or what?” “I thought you guys were supposed to be Microsoft-savvy; why can’t I host a .NET 4 application on your software?”  “When we run millions of jobs with lots of really short tasks, the Digipede database gets really big — can you fix that?”  You get the idea.

Well, to be honest, we never tested that “queue thousands of jobs while thousands of other jobs are running and then just delete the thousands that are queued” case, so yeah, weird things happened.  Should be better now.  And yeah, .NET 4 is a reasonable expectation from us — works fine now.   Yes, it’s true that there are ways to make the Digipede database grow — and while we’ve always had tools for managing that, those tools are simpler and more useful now.

That multi-core thing turned out to be the most popular one though, and it’s been one of my pet issues for a while, so let’s talk about that in more detail.  I’ve spoken at conferences, written articles, made videos, and given interview for years saying basically this:

• Mainstream developers know single-threaded object-oriented coding techniques, which take advantage of a single core.
• Meanwhile, chip makers are developing CPUs with more and more cores.
• The Digipede SDK is the simplest way for a mainstream developer to WRITE WHAT THEY KNOW (i.e., single-threaded object-oriented code) and EXECUTE that code on multiple cores on a chip, multiple chips in a box, and multiple boxes on a grid, all using the same programming paradigm.

And this has been great for us and for our customers — up to a point.  For purely compute-intensive applications, this approach scales linearly in cores and machines up to hundreds and even thousands of multi-core compute resources.  But many complex applications have a lot of I/O requirements as well, and just loading up (for example) an 8-core server (most likely, a dual quad-core box) with 8 cores worth of computation can actually slow down execution as processes wait for I/O.

So in the most recent release, we took a very simple brute-force approach to fixing this issue – we now allow users to “reserve” one or more cores per compute resource through a simple option in Digipede Control.  Early users report excellent results, with 6 or 7 cores computing away while the remaining one or two handle all other chores (including I/O).  Equally important, this approach is robust to additional increases in the number of cores per chip (which is forecast to reach several dozen within just a few years).

If you want to take the new version for a spin, ask for a free evaluation copy here.

Now, how about what’s NOT in our press release?  Well, you won’t find the word “cloud” in there…

Is it just me, or has the cloud meme really jumped the shark?  Look.  I used cloud computing before it was called that, and I’ll use it after that name has wandered off into the scrapheap of forgotten marketing buzzwords.  If a cloud is Google and a cloud is a cluster in a datacenter somewhere the user can’t see it, then a cloud is everything and nothing.  If a cloud is Amazon or GoGrid, then sure, our customers can deploy the Digipede Network there, or they can deploy it on their own infrastructure (then, if they want to, they can tell their bosses they’ve built a “private cloud” for all I care!).

The market knows Digipede as a provider of distributed computing software for the Windows platform, and as a provider of  high-productivity distributed computing tools for .NET developers.  That’s our role in the cloud and on the ground and everywhere in between.

When we started Digipede (more than 7 years ago!), my partners and I had spent more than a decade working together in the electric utility industry, and frankly were ready for something new.  While some of the ideas that eventually became Digipede had been rattling around in our heads for years, we built the Digipede Network as a general-purpose grid computing framework, not a tool for electric utility IT departments.  Indeed, while we knew grid computing was important in finance, military, biotech, and manufacturing applications, we didn’t think utilities would be particularly interested.

I guess you never know.

Over the past year, one of our hottest segments has been the electric utility industry.  We now have customers in generation, transmission, distribution, and power marketing companies, running a variety of applications from risk management to market simulation models.  We’ve had an opportunity to work with utility software giant Ventyx (recently purchased by even-more-giant ABB, the same ABB that bought our previous utility software company Energy Interactive — I think the world really is smaller than I realized…).

Our first bit of collaboration with Ventyx has involved adapting their energy planning and analytics software tool, PROMOD IV, to run on the Digipede Network.  This has been an instant hit with utility customers.  (OK, the phrase “instant hit” may not quite capture the pace of utility procurement processes, but you get the idea.)

PROMOD is a very detailed simulation model, and users often have to run thousands of scenarios — so projects can take days to complete on a single high-performance workstation.  (Indeed, my first encounter with PROMOD was in the early 1980s, on a mainframe at Portland General Electric, but that’s another “small world” story…)  Users tell us they end up walking from machine to machine starting multiple runs before going home at night — we call this behavior, which goes far beyond the utility industry, the “sneaker grid.”

Not surprisingly, “sneaker grid” users make GREAT Digipede customers, because (a) they know how inefficient and limited such manual work is, and (b) work is really piling up!  Ventyx knows this too, and actually had a grid solution through Sun a few years ago — but nobody wanted to install a Sun grid for a single application when all their other infrastructure was on Windows.  Opportunity knocks…

Now PROMOD IV users have a scalable solution that allows them to get order-of-magnitude increases in modeling throughput, using the tools and platform they already know and understand.  Ventyx and Digipede worked together on a description of this solution, which can be found here.

So now I’m a grid computing guy AND a utility guy.  Full circle.

The video of that conversation is now on Channel9; you can watch it here.

We’re offering developers who watch that video (and even those who don’t) a free copy of the Digipede Network Developer Edition — go to this page to get yours today.

Many thanks to  William and the rest of the Microsoft Evangelists for giving us this opportunity to get the word out about how Digipede and Microsoft work together to make software run faster and scale bigger!

Penny Crosman provided some good news for banks, hedge funds, and other money managers in her article today in Wall Street & Technology — good news for financial developers and IT professionals who need to access more processing power without complex application re-engineering. 

You can read the article for yourself — there are good quotes from AVM CTO Paul Algreen, a longtime Digipede customer – but from my perspective, the gist is this:

• CPUs are getting faster these days almost exclusively through putting more cores on a chip. 
• Hence, when you buy a fancy new server, performance only improves for applications that take advantage of multi-core architectures. 
• Yet most applications are single-threaded, leaving all but one core doing, umm, nothing. 
• AVM noticed this problem more than two years ago, and started using the Digipede Network to address it. 
• They’ve adapted compute-intensive legacy applications to run on a grid of multi-core boxes without expensive re-engineering, seeing huge performance gains. 
• Thanks to the intuitive programming model offered by the Digipede Framework SDK, AVM has added more and more applications to the grid since then, and they haven’t looked back.

This is quite typical of the experience many Digipede customers have had — that for most applications in financial services, multi-core and grid computing can be handled most effectively as two cases of the same general distributed computing problem.  

And yes, I’m going to plug our now-famous four-minute video on this topic again — you can watch it here.  Then you can request a free evaluation copy of the Digipede Network, and try it out on your own compute-intensive applications.  Because Intel and AMD aren’t waiting for the world to re-tool a few million enterprise developers; they’re banging out chips with more and more cores with every new generation. 

But with the right tools, you can take advantage of all that power — and that’s a welcome dose of good news for Wall Street!

We did a press release today  about the recent sale we made to the US Navy.  You can read that here, so let’s go a little deeper. 

The Navy has access to huge volumes of very accurate geodetic data — information that tells the location and elevation of every point on earth.  (I never knew much about this area until this year, but a lot of decent public information is available.  You can look at this Wikipedia article if you’re interested, and dig around from there.)  Processing geodetic data is a very compute-intensive process.  Combining that information with 2-dimensional image data is even more complex. 

One application for this data is “terrain generation,” a process of converting raw data into a format useful for visualizing terrain in flight simulations.  A group at Lockheed Martin develops specialized software for this purpose; we’ve been working with their TOPSCENE group for some time. 

That group approached us for assistance in increasing the speed of their terrain generation process.  They’ve been great to work with — they’ve put the Digipede Network to work on their problem, and have achieved great results.  Their own press release  about the Digipede-enabled version of their software documents a 20x speedup in processing for our mutual customer at the US Navy.   

We’re excited about this application, for a variety of reasons.

First, Lockheed Martin is the largest independent software vendor (ISV) we’ve worked with, and they’ve validated our premise that ISVs would find the Digipede Network (and particularly the Digipede Framework SDK) the best choice for grid-enabling complex applications.  This is an important point for us.  ISVs can work with other vendors, or build their own application-specific distributed computing solution, and Lockheed Martin certainly has the resources to pursue either path — but they chose Digipede, and have achieved great results. 

Second, we see many, many more applications in processing geodetic data.  We’ve already made other sales in that area to government agencies outside of defense (no announcements yet, but stay tuned!), and we see increasing interest from commercial customers in this area as well.   

Finally, as many financial customers struggle with market issues (some of our clients from 2007 no longer exist in recognizable form), it’s important that Digipede diversify and demonstrate growth in other markets.  Customers like the US Navy certainly help with this important goal.  While government purchases can be  slow, we’ve been quietly working this area for years, and it’s great to see results we can discuss. 

Case in point: Marc Adler’s recent post about Velocity, in which he says:

. . . It is no secret that the most prevelant use of object caches on Wall Street is with Grid Computing. How will Velocity interface with Compute Cluster? How about Digipede (if I know John Powers, he probably has support for Velocity already)?

Well, Marc, if you’re going to post on Sunday mornings, you may not get instant confirmation of your clever guesses, but now it’s Monday afternoon, so I can confirm — yes, we have a working PoC in the lab at Digipede, proving (to ourselves at least) that the Digipede Network and Velocity work great together. And yes, it provides significant performance improvements for certain types of activities important to Wall Street folk.

If anyone is still wondering why we chose .NET as the basis for our grid computing software, this is just the latest example — Microsoft just keeps giving us great free stuff on which to build. The Velocity CTP came out last week; this week, we have working code that provides real benefits.

Rob just posted his initial comments; there will be more.  And we’ll have feedback for the Velocity team.  Watch this space (and Rob’s, and Dan’s…).

In any case, Derrick Harris of Grid Today has done a great job of connecting the dots for us in his excellent article today.  Read the whole article, because he offers good insight on how important this announcement really is, but here’s his analysis of how it affects Digipede:

Whatever emerges from Velocity also should be good news to Microsoft’s technology partners — in particular Digipede, which has been delivering distributed computing to .NET apps and now might get the add-on technology it needs to compete with the big boys. Digipede has received no shortage of praise from customers and commentators alike about its relatively inexpensive and very user-friendly solution, but one of the drawbacks has been its limitation in terms of what types of jobs the Digipede Network can handle, namely CPU-intensive jobs benefitting from parallel processing. If Microsoft and Digipede can make Velocity and the Digipede Network function as a unit and keep the price down, Digipede could find itself selling to a whole new, real-time-data-loving audience. That this integration will occur is pure speculation on my part, but it seems to make sense on the surface.

I have no comment on specifics at the moment, but let’s just say — Derrick, you nailed it.

And as he reports, thanks to improvements in both Mono and the Digipede Network, the answer is — not much. We’ve got a working prototype of a Digipede Agent running under Mono on Linux that runs a Digipede job.

Digipede on Linux? Has the world turned upside down?

Hardly.

Since the beginning, Digipede has been focussed on adding value to the Microsoft platform. And customers know that. Customers also understand that Microsoft is getting better and better at making sure its products interoperate with others, even on other platforms, and Microsoft’s partners have to facilitate that. We get questions from customers pretty frequently about Mono, and lately those questions have gotten more specific, so it seems prudent to investigate any technical blockers from time to time.

So let me re-state what Robert said, and what I said above — this is an initial assessment, a technical experiment only, not a shipping product.  Rob’s post (and mine) are not a product announcement — this is a blatant “trial balloon.”  We want to hear what the market thinks of Digipede on Mono.

Why might this be interesting? Let’s back up a step and take a look at enterprise grid and HPC deployments.

Most enterprise customers have what is often called a “mixed” IT environment. That’s a euphemism for an unplanned and chaotic assortment of technologies that have piled up over the years into some type of barely-managed infrastructure. In almost every enterprise, Windows runs on most or all of the desktops. In almost every enterprise, there is some mixture of Windows and Linux servers, with maybe some Solaris and/or other UNIX flavor(s) thrown in. In almost every enterprise with an HPC infrastructure, most or all HPC nodes run Linux.

This is just reality — Windows is miles ahead in 2008 desktop market share, and Linux is miles ahead in 2008 HPC market share. Do I wish it were different? Sure — if Microsoft had a bigger share of the HPC market (and we’ve been working diligently to help make that happen), we’d have an even bigger market into which we could sell our software. And that will happen, I have no doubt. We tell all our customers “Windows HPC Server is the best option for adding power to a Digipede grid,” and that’s the truth. Go buy some now.

But the fact remains, there’s a lot of existing infrastructure — desktops, 32-bit Windows servers, Linux servers and cluster nodes, Solaris servers, and more — that enterprises are not going to throw away.  All this infrastructure represents potential grid computing power.  The Digipede Network has always run on heterogeneous Windows networks — with Agents running on 32- and 64-bit Windows desktops, 32- and 64-bit Windows servers, and cluster nodes running Windows HPC Server (formerly Compute Cluster Server). Our reluctance to include boxes that don’t run Windows has always been mostly about applications — it’s still relatively rare to find applications that are actually deployed across multiple operating systems simultaneously.

But as Mono gets better and better, we hear from enterprise customers and prospects who are getting more interested in it. They like the idea of being able to use more of their existing infrastructure more efficiently. They want to take advantage of Digipede’s great developer experience to deploy more applications — with minimal changes to that infrastructure.

So let’s get back to Robert’s closing question: “Now that we can do it, what should we do with it?” What do you think? Is the market crying out for a multi-OS .NET grid? Or is what we’re hearing just idle curiosity?  Let’s hear from all sides.

Sure, the individual steps like specifying, developing, testing, documenting, and planning support for new software features are difficult enough — but the discipline of knowing when to STOP adding features, and to focus instead on finishing a complete, polished, release-ready product is tougher than it sounds to those outside the industry.

In any software organization worthy of the name, there are more good ideas than can possibly be put into any specific product release. There are also just a stunning number of bad ideas competing for inclusion in shipping products (I am notorious within Digipede for proposing needlessly specific bad ideas. Mercifully, my partners of 20 years have honed their skills in talking me out of the worst of them.)

We decided early on at Digipede that our feature set would be guided by three principles: Performance, simplicity, and a focus on adding value to the Microsoft platform. Over the past five years, these principles have helped us make decisions on what to include (and as importantly, what to exclude) from our software.

Last month, we reached general availability of the latest release of the Digipede Network, Version 2.1. You can see what’s new in this release on our Web site, but now that our customers have had an opportunity to upgrade, let’s look at a few of the specific new features to see how we did in sticking to those principles:

Job concurrency: The improved Digipede Agent software can manage different applications running simultaneously on multiple cores of a single compute node, maximizing utilization of compute nodes on the grid. Users can set Job Concurrency values to allow the Digipede Agents to work on multiple jobs simultaneously: designate which applications can safely run with other applications, which applications can run side-by-side with themselves, and which applications are not compatible for concurrent jobs.

Performance! This one is just amazing. As new machines ship with more and more cores inside, I am continually baffled at the lack of attention from ALL the major vendors out there about how to take advantage of those cores. Sure, Intel talks about compilers and Microsoft talks about Parallel Extensions and so on — but in shipping products in 2008, there’s just incredibly little help for users and developers who want to take advantage of multi-core processors. What we shipped in Version 2.0 last September is still miles ahead of other software options in terms of both development patterns and execution modes for multi-core processing. With Version 2.1, we’ve extended that lead significantly — if you want to take advantage of dual-CPU quad-core servers and desktops TODAY, you need to take a look at how the Digipede Network handles concurrency. Watch the 4-minute video that shows how, then get an evaluation copy of the software and try it yourself!

Management APIs: New management APIs give developers programmatic ability to create, modify, and delete resource pools. (Available in Professional Edition only)

Performance (specifically, scalability), and Simplicity (of grid management). A browser-based UI for grid management is great — for small grids. As our customers deploy larger and larger grids, they need both the browser-based UI of Digipede Control and a wider range of tools for the programmatic manipulation of grid resources. It is vastly simpler to take advantage of thousands of grid nodes through simple extensions to our management API.

Risk-free sharing: “Pool Rank” permits risk-free sharing of resources: you can add your servers to the enterprise grid and ensure that they always work on your jobs first. That means that by joining the grid, you can only improve your application performance. You can donate your cycles when you are not using them without worrying that your application performance will degrade, because you are always guaranteed that your machines will work for you.

Performance and Simplicity. We’ve also referred to this feature as “Selfish sharing.” We hear from other grid vendors about how users “must” get over the practice of “server hugging.” We try not to be so arrogant; we’ve never found that scolding our customers is good business practice. If customers want to preserve unconditional priority on their own servers, we say “good for them.” So we’ve built a straighforward way to preserve absolute priority for the resource owner, even when they offer to share surplus resources. From what our customers tell us, we think this approach encourages efficient resource sharing far more than lecturing ever would.

First Grid Computing Solution Certified for Windows Server 2008: We followed the long and winding road of the Early Adopter program to become the first grid solution to obtain this important certification, so that customers can be confident that our software works not only with the Microsoft products they use today, but with all the latest improvements Microsoft is bringing to market now.

Performance, Simplicity, and Microsoft focus. By aligning with Microsoft’s technology and strategy, we help our customers create a truly dynamic IT infrastructure. Server 2008 brings many benefits in performance and manageability, and we’re confident that our customers will be upgrading quickly (more quickly than, say, to Vista); we want to be sure they can use our latest capabilities on Microsoft’s best OS platform.

Let me be candid here; these benefits do not come free to ISVs. I have considerable anxiety over extending yet further the number of versions of Microsoft products we support — for example, while I think Server 2008 is great, and Visual Studio 2008 is great, and the new SQL Server 2008 will be great, staying current means we’ll have to start enforcing our requirements by turning away requests for support of Windows 2000 and SQL Server 2000. The combinatorics for testing on multiple OS versions, .NET versions, SQL Server versions, IIS versions, and upgrade paths for our own software versions get out of hand quickly. I’ll have more on this issue another day. For now I’ll just say I’m happy with our decision to stay current — mostly.

Automatic Failover Package and Integration with NLB: Failover has long been a feature of the Digipede Network Professional Edition but with the optional Automatic Failover Package, organizations can now have complete out-of-the-box integration with Windows Server 2008 load balancing, giving “hands-free” failover to mission-critical applications.

Performance, Simplicity, and Microsoft focus — yes, even this advanced capability was guided by our goal of simplicity. While automatic failover is often considered a complex requirement, we made some basic decisions to keep it as simple as possible. First, we made automatic failover it’s own SKU, so customers without the need for high-availability configuration don’t even have to think about it. Second, we did away with a lot of the manual scripting that often slows implementation of failover solutions — you can have it running very quickly. Finally, we left as much as possible to popular existing technologies — SQL clustering and NLB — so the implementation steps will be as familiar as possible.

Reports Package: Assembles critical information about the use and optimization of the grid, with easy-to-understand charts and graphs, flagging of critical information, and drill-down capability, giving enterprises fully integrated optimization of grid performance, with tracking of who contributes to and who benefits from grid resources.

Performance, Simplicity, and Microsoft focus — In larger systems, simple and informative visual tools are essential for wringing the most performance possible from a grid. Users and administrators become far more productive in their routine monitoring functions and troubleshooting activities with this new package, which plugs directly into Digipede Control (our admin UI). And by building on SQL Reporting, we’ve created a framework for future extensions.

Overall, I’m pleased with the extent to which we have driven the improvement of our product by staying focused on the three principles described above. To be a little less self-congratulatory, I wish we had stopped adding features at least two months earlier and brought most of these capabilities to market sooner, rather than piling quite so much into a single release (and there’s certainly more than I’ve had a chance to discuss here).  Perhaps another day, I’ll have a chance to discuss some of the things we (purposely) left out!  Now that V2.1 is in the market (and getting rave reviews from our customers), I’m eager to see what great new applications our imaginative customers create and deploy on our latest platform.