GIS Software Techniques and Implementation

A newsletter by Daniel Elroi

In this issue I shall address both technical and organizational issues related to migrating an ARC/INFO installation to the NT platform. In future issues I shall continue to address both technical and implementation topics such as Preparing for ARC/INFO 8 and ODE and 3D Analyst: Treat or Trick? I invite you to read my newsletter and to pass it on to friends and colleagues. You may send me your comments at the e-mail address at the bottom of the newsletter.

Migrating ARC/INFO from UNIX to NT: Is this the right time?

Here is the situation you are probably in, more or less: ESRI has been pushing the NT platform for ARC/INFO for a couple of years now, and you certainly have some NT machines on which you run ArcView, Access, Word, PowerPoint, and Project. But NT hasn't come close to matching your SUN's or HP's performance, at least as far as you can tell. You have investment to recoup in your UNIX machines. Plus you've heard that the security is either uncertain or complicated in NT. So you've let things stay the way they are: ARC/INFO lives on UNIX. The question is: Is this the right time to switch to ARC/INFO on NT? If so, how many of your licenses? How quickly? And at what cost?

To answer these questions we should briefly list the pros and cons of migrating some or all of your ARC/INFO licenses to NT. Then we'll go over some ways for balancing the pros against the cons.

Pros and cons of migrating ARC/INFO licenses to NT

Comparing performance between UNIX and NT platforms

So how do you decide on a plan of action? First, you will probably want to prove to yourself that NT can hold its own in terms of performance. For that, you will need to perform a benchmark test. Of course, you will have to get a copy of ARC/INFO for NT, an NT machine, and migrate data to the NT platform if you are contemplating a complete migration. Then you will want to design a test that will compare processing, throughput, and display performance. You might obtain results such as those presented below, or you might see significant variations.

The following results were obtained on a Digital Alpha 3000/800 hardware, Digital UNIX 4 operating system, and ARC/INFO 7.1.1; a Silicon Graphics Indy hardware, IRIX 5 operating system, and ARC/INFO 7.1.1; a Micron Pentium Pro 200MHz hardware, Microsoft NT 4.0(3) operating system, and ARC/INFO 7.1.2., and a Micron Pentium II 400MHz hardware, NT 4.0(3), 7.2.1.

There are certain aspects that might greatly influence the results that you obtain. For example, in some cases accessing data on UNIX drives from NT machines, even using a good NFS software on the NT side, and a decent network in between, can see performance degradation of up to 1000% over NT-to-NT data access. This is hugely significant in the strategy you adopt for migrating your data, not just your software and applications. Another critical aspect to using ARC/INFO in NT is that there are several performance-crushing flaws (bugs?) in the current release. For example, the sort command in tables seems to have trouble releasing scratch files and memory, which means that the first command used after an eight-minute sort, can itself take five minutes. Even the quit command! I am fairly certain that such problems are memory management problems, and hope that they will soon be rectified by ESRI. The point is that performance should be measured against your specific uses of the software, preferably in pilot project conditions.

When is performance good enough to warrant considering a migration? When it, plus all other benefits outweigh the problems. So you may have to accept some performance degradation in certain situations. But the benefits of having only one computer on a user's desk, the lower cost of the hardware, etcetera, may outweigh that performance hit.

Impact on hardware and software purchases

Choosing to migrate to NT has obvious impacts on hardware choices and costs. But it also has impact on software choices and costs. On the hardware side, you will be able to purchase a single hardware platform per employee, as opposed to providing them with a UNIX box for GIS and a PC platform for the rest of their computing needs. The unit costs will be lower, as well as your ability to take delivery of hardware faster, and to compare prices on identical configurations. One of the most difficult aspects of shopping for a UNIX platform is leveling the playing field in terms of specifications. This is pretty much academic when comparing NT machines.

Not only is the base price of a hardware unit lower, but adding such things as additional memory or hard drives, new monitors, a faster CD reader or a CD writer, are easy, fast, and cheap with PCs. At the same time, your choice of software packages may change as well. For example, you may choose ArcPress when you migrate to NT, because you will be able to share it with ArcView users at no additional cost. You may abandon your SoftPC, SoftWindows, and other PC emulators on your UNIX platform, and begin to use Windows software at their true native speed. And you may choose entirely different software for such tasks as image manipulation and analysis, animation, and so on.

So how can you decide on whether to, and when to, switch to NT based on hardware and software configuration? Consider the life expectancy of your existing hardware. A fairly typical lifespan of end-user hardware these days is about 2-3 years. Where in that cycle are your machines? If you retire them early, will someone else be able to make use of them while they still have "life" in them? Can they be traded in or re-sold at this point? Can they be better utilized as servers somewhere in the organization? Can new PCs be purchased from the operating expenses budget rather than from the capital expenses budget? In some cases the savings resulting from terminating an expensive UNIX maintenance contract alone can pay for a zippy new NT machine!

Impacts on personnel

What does it mean to an organization when it no longer requires UNIX gurus to maintain its hardware and network? When it can make use of its pool of general systems administration staff to manage its GIS system, rather than having to maintain specialized staff for that purpose? When the exchange of knowledge between the UNIX and NT staff is increased, thereby making the UNIX staff less critical to the smooth running of the operation, and therefore making the organization less likely to come to a screeching halt when they leave unexpectedly?

Moving to an NT platform can bring a GIS group back into the mainstream of an organization's information technology infrastructure. This can have positive economic and political ramifications for the group. The GIS group can cease to operate as a "special needs" group, and begin to operate as a regular unit of the organization. This also means that the staff hired to support the system can be less expensive and difficult to obtain, and that other staff can step in to fill in for a vacant position.

As a matter of fact, the migration to NT in general, and in the case of ARC/INFO in particular, can also bring the programming requirements of a GIS group into the mainstream of an organization. One of the main purposes of migrating to a Microsoft-based operating environment, after all, is to standardize on common tools. This includes programming tools, such as Power Builder, Delphi, Visual C++, Visual Basic, and Java. And this in turn means that programming GIS software can to a large extent be handed over to... programmers! Not to diminish from the considerable achievements of ARC/INFO programmers in the past, but the burden of finding personnel who are both knowledgable in the use of GIS software, spatial analysis, and programming, is difficult for an organization. By enabling more of the programming to be performed by personnel trained in programming, rather than in GIS, an organization attains greater freedom in its hiring procedures, which in turn can have positive economic benefits. Should existing GIS programmers be threatened by this? Not really. An AML programmer who learns Visual Basic in order to take advantage of ARC/INFO's Open Development Environment (ODE), will at once make himself or herself more marketable outside of the GIS arena, and also give them skills that can be used with other software such as ArcView and Map Objects.

Conclusions

There are many issues impacting the decision to migrate ARC/INFO licenses from UNIX to Windows NT. Some of the issues which I have not touched on include license management, data sharing, specific hardware specifications, Alpha versus Intel chip implementations, and so on. Please refer to ESRI's workshop notes on the topic from the 1998 User Conference (see below). However, whereas ESRI addresses the more technical aspects of the transfer, I am more concerned with the people side. Will the users be happier with an NT implementation? Will the managers? The bean counters? Will your mother approve?

My conclusion is that there are a large number of you out there in ARC/INFO Land who ought to take a serious look at transferring at least some of your licenses to NT right now. The sites that I have helped migrate have not been sorry they had!

References and Further Reading

The technical notes from the Adding Windows NT to Your Existing ARC/INFO Environment workshop at the 1998 ESRI User Conference in San Diego, can be downloaded in PDF format from ftp.esri.com/pub/userconf/techwork (the file name is tw508.pdf).