My recent “Virtual List on Steroids” presentation at DIG-FM, and the preparation of a revised version to give at dotFMP next week, has prompted me to reconsider certain assumptions re: optimizing performance both locally, and across a LAN and/or WAN.
Specifically, there are two things we’re going to look at today. First, a way to dramatically speed up sorting on related data. Then, having incorporated that trick into the Fast Summary approach, we will compare Fast Summaries vs. Multi-Finds under various scenarios.
If you do complex data analysis, then from time to time you probably need to group, summarize, and parse data into variables and/or fields. There are various ways to accomplish this, including the Fast Summary technique, which regular readers of this blog may be familiar with, as it has made a number of appearances here over the years.
Well today we’re going to look at a couple alternatives to Fast Summaries, with the help of some demo files, which you can use to do your own benchmarking.
Or, if you’d rather not do your own benchmarking, scroll down to see results from my tests. Bottom line: each of these three techniques can be fastest, depending on circumstances.
(Also, if you’ve been putting off exploring Perform Script On Server, a.k.a. PSOS, host the demo files on FileMaker Server 13 or later. The reporting routine optionally uses PSOS, so you can dive in painlessly and see what you’ve been missing.) Continue reading →
Picking up where we left off in part 1, today we’re going to take a look at examples 2 through 6 in the Virtual List Charts demo file (the demo has been updated since part 1, so I recommend downloading a fresh copy).
We covered example 1 and most of the general concepts last time, so today we’re mainly going to touch on specific points of interest, but to briefly recap… Continue reading →
Today we’re going to look at applying the virtual list technique to FileMaker charting with the goal of producing a reusable chart “object”, or rather, a series of chart objects. We’ll need more than one because while certain attributes (e.g., chart title) can be set programmatically, others, including type (e.g., column or line), must be hard-coded into the chart object.
We’ve already explored Bruce Robertson’s virtual list on this site a number of times, but briefly, you create a utility table in your solution to facilitate non-standard viewing, reporting, etc., and pre-populate it with “more records than you’ll ever need”. The records in this table will derive their data “virtually”, by parsing it from an array — typically one or more $$variables.
Well it turns out the technique can be applied to charting as well, and today we have a demo file, Virtual List Charts, that contains six examples: three for Web Visits…
In part 1, we looked at a new way to count unique entries in a sub-summary report based on an arbitrary found set of sales data. To avoid redundant explanation, this article will assume the reader is familiar with part 1, but in a nutshell…
1. we used a pair of FM 13 summary list fields
2. to create found-set-aware multiline keys
3. which we related to special purpose table occurrences
4. which provided the basis for related value lists of salespeople
5. the ValueListItems function allowed us to grab lists of unique salespeople
6. and the ValueCount function allowed us to count them
The demo in part 1 was based on a single “flat table” of data. Today we’re going to unflatten that table, and see how the technique can be applied to a properly normalized system, i.e., with separate tables for Sales, People and Zones, related like so…
Several recent postings on this blog have offered variations on a theme: using the new-in-13 summary list field type as the basis for a multiline relational key. The net effect is to produce a relationship that is found-set-aware. (If you aren’t familiar with summary lists, you can read about them here.)
Last week we looked at basing a value list on one of these “summary list relationships”, to facilitate counting unique values within the current found set…
…and today we have a demo file, FM-13-Count-Unique-v1, that extends the concept to sub-summary reporting.
Last time we looked at several summary reporting tricks, including a conditional subsummary (when an item’s Status is “Scheduled” it will have a value in the Substatus field — otherwise Substatus will be blank). The challenge was to generate a summary report showing Substatus only where appropriate, without seeing any annoying empty gray rows beneath Pending, Cancelled or Completed. And last week’s report worked fine… in preview mode.
One of the best ways to learn about a particular FileMaker feature or behavior is to build a demo. You might build one in response to a client request, or to try to answer a question somone has asked, or just to see what happens. At any rate, today we’re going to look at three demos, each of which explores some aspect of summary reporting. For reasons of backward compatibility today’s demos are in .fp7 format, but you can convert them to .fmp12 format if you are so inclined.
What these three demos have in common is multiple sub-summary parts on reporting layouts… so what you see on the report depends on how you sort it.
Last October I began a series on Outer Joins, which explored a number of different ways to display summarized information in a grid, and at the time I concluded that the “fastest” method was to leverage the FileMaker relational model. And it was plenty fast, locally… and not too bad on a LAN… and technically, it was the fastest method on a WAN but only because the other methods we looked at were even dog-slower than it was.
I like to test solutions on a WAN, even if they’re only going to be deployed on a LAN, because it’s a great way to uncover performance bottlenecks. But recently I needed to deploy a summarized grid on a WAN, and was incentivised to come up with something faster… and after a bit (well, okay, a lot) of trial and error settled on the approach we’re going to look at today. To cut to the chase, with a million records in the test file, the previous best grid rendering time of 11 seconds on a WAN has been reduced by a factor of 10, to just over one second.