Upstart To Breathe New Life Into Venerable OpenVMS
The many thousands of organizations that are still happily using the OpenVMS operating system to run mission-critical applications have heaved a collective sigh of relief now that Hewlett-Packard has decided to license the source code to OpenVMS to a third party company that has promised to enhance it, and more importantly, port it from Itanium to X86 processors.
OpenVMS is, of course, the latest in a line of operating systems that stretch back to the VAX minicomputers from Digital Equipment Corporation back in the late 1970s. VMS, and then OpenVMS after it got some Unix features, were widely used in scientific and technical computing as well as in commercial settings. OpenVMS systems are used in radar systems, nuclear power plants, nuclear submarines, and chemical factories but have clearly lost their cache in technical computing. But, the VAX Cluster and OpenVMS Cluster software that allowed for multiple VAX, Alpha, and Itanium machines to be ganged up to share data and work is not only legendary, but still useful. In fact, this code was licensed to Oracle to be used as the basis of its Oracle Real Application Clusters way back when Compaq owned DEC and was short on cash. DEC had previously sold off the Rdb relational database, which ran on both OpenVMS and Tru64 Unix, to Oracle in 1994 when it needed some cash.
A few years back, in an effort to cut costs as sales of OpenVMS-based Integrity systems started to wane, HP offshored most of the development for OpenVMS to facilities in India. At the moment, OpenVMS is only certified to run on four-core "Tukwila" Itanium 9300 processors in the Integrity i2 systems, which are pretty long in the tooth. HP had, as it turns out, done a lot of the work to port OpenVMS to the eight-core "Poulson" Itanium 9500 processors, which launched in early 2013.
Now VMS Software Inc., a newly created company formed by a company with 30 years of experience supporting VAX, Alpha, and Itanium machines running VMS and OpenVMS, is going to be putting the OpenVMS development team back together in the old DEC haunts around Boston, Massachusetts. The plan, says Duane Harris, CEO at VMS Software, is to hire software engineers and get to work enhancing OpenVMS. The first task will be to finish the OpenVMS port to the Poulson Itanium processors, which Harris says is mainly working on some drivers and doing quality assurance and testing on the code to make sure it runs correctly on HP's Integrity i4 servers. The hope is for this OpenVMS release to be available in early 2015, but it all depends on how the testing goes.
Under the deal inked between HP and VMS Software, the latter company has been granted the rights to the source code for the OpenVMS stack and it is allowed to make derivative works based on this code. But, explains Randy Meyer, general manager of the Mission Critical Systems division at HP, VMS Software does not have the right to open source any of the code and HP has no intention of doing so, either. HP will continue to sell the existing OpenVMS 8.4, which Meyer says has several thousand customers who have support contracts. It is hard to say how many OpenVMS machines might be out in the field, but Harris says the last numbers anyone saw estimated the base at between 100,000 and 400,000 machines about a decade ago. The base is very likely quite a bit larger than what HP sees, given how reliable the OpenVMS operating system is and that many organizations have lost the source code to their applications and cannot easily port them to Windows or Linux. (This sounds ridiculous, but Harris says this is one reason why OpenVMS persists.)
HP is not planning job cuts among its staff of OpenVMS software engineers and Meyer says that they will continue to provide full engineering support for OpenVMS through at least 2020. Extended product support, which means that HP will not create new bug fixes but will help customers find existing patches and workarounds as they discover issues, will continue on Integrity i2 machines running OpenVMS 8.4 through at least 2025. HP will also resell the OpenVMS release that come from VMS Software and will, if customers want, provide Level 1 and Level 2 support for it if they want one throat to choke.
VMS Software will also be licensing its OpenVMS releases and selling support services for it. The future releases include a port of OpenVMS to the future "Kittson" Itanium processor from Intel, which looks less like a radically different processor as Intel had planned and more like a tweak to the existing Poulson chip. Given this, a port of OpenVMS to the Kittson chip, perhaps coming late this year or sometime next year, is not a big deal. But VMS Software has more ambitious plans, including a port of OpenVMS to X86 server chips.
Harris has located VMS Software in Bolton, Massachusetts, on the Interstate 495 corridor not too far from DEC's old stomping grounds. (Route 128, where a lot of DEC offices were scattered, is "too busy" Harris said with a laugh.) When HP offshored OpenVMS development to India, there were a lot of DEC coders in Massachusetts and New Hampshire with deep experience with OpenVMS, and Harris says he will hire somewhere between 70 and 100 of them to do the OpenVMS port to X86 iron.
"We have no illusions that we are going to take the market back from Linux any time soon," says Harris with a laugh. "But frankly, we don't want to limit ourselves to X86 chips, either. We have a marquee operating system that is widely used in embedded environments, and it is time to broaden its scope." That could mean OpenVMS ports to ARM, MIPS, and other processor architectures.
The interesting bit as far as EnterpriseTech is concerned is OpenVMS Clustering, one of the earliest and best scale-out architectures for running both scientific and commercial applications. "The clustering capabilities of OpenVMS are still unmatched these days."
In its latest iterations, the software can be used to cluster up to 96 machines together to share work, and depending on the nature of the underlying network, the work can be relatively tightly or loosely coupled across those nodes. The genius of VMS and OpenVMS was that this clustering was relatively easy to install and use as well as scalable. That cannot be said for most of the commercial scale-out clustering available on the market today.
Here is the plan that VMS Software has put together to enhance OpenVMS:
Harris doesn't know how long it will take to do the port of OpenVMS to X86 machines because he has not put together the software development team, which will happen over the coming weeks and months. But as you can see from the roadmap, VMS Software is not only promising a port to X86 with its v.Next release but also performance and scalability enhancements that will allow it to span more cores, processors, and nodes. Interestingly, the future v.Next release will also include a dynamic and static translator that will be able to convert applications compiled for VAX, Alpha, and Itanium processors so they can run on X86 iron. (That is not shown in the roadmap above, but is in this document on the VMS Software site.)
Each of the VMS Software releases will be supported for a minimum of five years under a standard support contract; OpenVMS 8.4 and earlier releases are supported under the existing HP schedule.
VMS Software is a spinout of Nemonix Engineering, which is based in Northborough, Massachusetts, and which has been providing services and support for VAX, Alpha, and Integrity machines for the past 35 years. Harris was appointed CEO of Nemonix in February 2013, and approached HP about a year ago to try to work out a deal to keep OpenVMS from slipping into oblivion. VMS Software has a number of investors backing it, notably Johan Magnusson Gedda, one of the co-founders of Rocket Software. Gedda has engineering degrees from MIT and an MBA from Harvard. Rocket is based in Waltham, a techy suburb of Boston, and has expertise in providing tools for legacy IBM minicomputer and mainframe environments, among other things.
Meyer said that HP has gotten all kinds of pitches concerning what to do with the HP-UX, OpenVMS, and NonStop environments, which were all ported from their respective PA-RISC, Alpha, and MIPS platforms to the Itanium processors in the wake of HP's acquisition of Compaq in 2001. HP decided last fall to do its own port of the NonStop fault tolerant database cluster from Itanium to Xeon processors. In the past several years, HP has at least twice put together a plan to port HP-UX to X86 chips (including both Xeons and Opterons) but could not convince the bean counters to fund the projects. The ports of NonStop and OpenVMS are more practical, given that the software stacks are more self-contained and customers largely code their own applications for these platforms. HP-UX usually runs an Oracle database and any one of thousands of third party applications, which would all have to make the jump from Itanium to X86 as they made the jump from PA-RISC to Itanium more than a decade ago. The odds of that happening are very small, considering how much the HP-UX base has shrank in the past several years. HP has concluded that its best option here is to help make Windows and Linux more resilient on mission-critical HP Xeon machines and support HP-UX customers on Itanium machines far out into the future. And further down the road, HP is focused on bringing The Machine, a cluster of microservers and memristor memory linked by photonics that it revealed as a massive research effort back in June, to market.