Review of Performance Gallery
Shawn M. Gordon
System performance is something that affects all of us at one time or another, but not everyone is equipped to make well informed decisions about how to manager it. Long ago there were very few tools to help you look at and manager performance, now the field has started to become very crowded. The folks at Lund Performance Solutions (LPS), have been providing various tools to measure system performance for a number of years now in the form of the SOS product (Which I will review next month). Their is however a less nuts and bolts type product that gives graphical overviews of response over time for various aspects of your computer system. This product is known as Performance Gallery, and it allows you to collect data on your HP over time, download it to the PC, and then generate various types of graphs and charts so you can quickly zero in on your problem areas.
I tested Performance Gallery on my HP-3000 series 37 with 2 meg of memory and 150 meg of disk space. My PC is a 25Mhz 486 with 12 meg of RAM running MS Windows 3.1 and Novell DOS 7. I am using a direct serial connection at 9600 baud from the PC to the 3000.
The screen snapshots of the PC based graphs are done in the monochrome mode, so you will see all sorts of strange line drawing to distinguish between different values in the charts. The color charts are very nice and easy to look at, but you will have to get a copy of the software to see them. Also, some of the dates may seem funny on the snapshots. I had accidentally had my system date set back for testing another application, and I neglected to change it. I hadn’t noticed until actually doing this review, and I didn’t want to spend another two weeks collecting data just to fix the date.
With any product that is concerned with showing you meaningful information over time, it takes some time to collect the data. So I first had to set up and stream the performance collector job. There are actually a number of versions of the collector job, which can be a bit confusing when you are first getting started. It depends on if you want the job to run for 8 hours, or over a 24 hour period. It also depends on if you also have the SOS performance product from LPS. So this gives you 4 different jobs to choose from. The only real difference between the 8 and 24 hour job, is that the 24 hour version of the job will re-stream itself every day at midnight. It seems that we want to have a performance log file for each work period. If you are a regular 8 hour day kind of shop (rare these days) then use the 8 hour job.
Once you have selected the appropriate job, then you may want to make some changes to the job in terms of it’s sample rate time. The default sample rate is 600 seconds (10 minutes), you may want to set it tighter to get more samples. Keep in mind, the shorter the interval the more impact you will have on the CPU, I wouldn’t get to concerned unless you set it down to 5 or 10 seconds, so it is almost constantly sampling, this could kill a system potentially.
Especially on the Spectrum computers where performance tools typically use the HP supplied Measurement Interface AIF, which seems to have a very large overhead. On my Classic, I didn’t really notice a problem.
Another configuration option is the ability to define ‘Workloads’. A Workload can be a logical grouping of some type that you want to be able to track more specifically on some of the graphs. There are several graphs that allow you to break it out by workload. By default you will have JOB (all batch processes), SESSION (all online users), and SYSTEM (which is system processes). You can further define this by modifying a file called SOSWKDEF.PUB.LPS. A work group can be defined by specifying JOB, SESSION,
or BOTH, program names, user names, logical devices (this is only marginally helpful if you are using virtual devices). Here is a sample SOSWKDEF file from the Performance Gallery manual.
ACCOUNTREC !NAME of group (max 10 characters) JOB !GROUP TYPE (JOB/SESSION/BOTH) USER=@.ACCTNG,AR !USER SPECIFICATIONS !At least 1 blank line (required) ACCOUNTPAY !NAME of group (max 10 characters) SESSION !GROUP TYPE (JOB/SESSION/BOTH) USER=@.ACCTNG,AP !USER SPECIFICATIONS PROG=@.EXEC.ACCTNG !PROGRAMS TO INCLUDE !At least 1 blank line (required) COMPILING !Let's track the programmerrs BOTH PROG=COB@.PUB.SYS !Track any COBOL PROG=CC@.PUB.SYS !Track any C
As you can see, this will give you the ability to really zero in on people if you need to. The only problem here is that you have to set it up before hand. There is no way to be able to drill down
through a graph after the fact to trace a particular process or user. I would have liked to have seen an exclusion parameter so you could use a wild card to include everything, and then exclude just a few objects.
Ok, so you found the job, modified the sample rate, set it up to stream, and have collected data for a week or month, or any time frame that you want to work with. Now you are set to Combine, Extract, and Download the performance data.
Figure 1 shows an example of the online application and how simple it is to gather and download your data. After you have collected enough system data, you will want to go through this step.
Figure 2 is an example of the menu that is available once you are inside SYSLOGX. This gives you access to various features and capabilities. To gather log information into a downloadable file, you would enter the Performance Gallery submenu.
Figure 3 shows an example of the log information that is displayed in the SYSLOGX program when you first start it up. As you can see, I set my snapshot timer to just 2 minutes. This also gives you an overview of the response time without even having to download the collected
performance data. Of course this isn’t going to be near as good or comprehensive as pulling it down to the PC.
Using your favorite terminal emulator, you will now download the combined and parsed log file to your PC with a file extension of PFG. The next step is to run Performance Gallery for Windows and open your log file that you just opened.
Selecting “GRAPH” from the menu bar, and then “NEW” will display a window as seen in Figure 4. As you can see their is a HUGE list of chart types to look at. Any chart that says “by Workload” means it can be broken down by the Workloads that you defined (as described earlier). If no Workloads are defined then you will see the chart broken down by Job, Session, and SYSPROCS (system processes). Even this minimal breakout can be pretty informative. Figures 6 and 7 show a couple of different types of Workload breakdown graphs.
Figures 5 and 8 and included as examples of some of the different graphs. It isn’t practical or necessary for me to include a sample of each possible graph that is included. You really do need to see the color graphs, they look outstanding.
There are various options for editing and saving graphs, as well as printing and modifying colors from the defaults. The program is highly usable and configurable. What’s really nice is the inclusion of a standard Hypertext Windows Help facility. This makes it almost completely unnecessary to even look at the printed documentation. I have learned more products just by using the online help than I care to mention. The inclusion of a robust help facility is always greatly appreciated.
Usability and Installation
The installation on the PC is very straight forward, as it follows the standard MS Windows installation procedures. This takes under 5 minutes to get setup and loaded. The HP part of the installation follows the standard that most vendors do, load a tape, restore a job, stream the job, remount the tape, and after a few minutes you are done. I like the fact that they clean up their installation jobs after they are done.
The HP based software consists of making sure a background job is running, and then pulling down the collected data. The host based program for pulling data together is menu driven, and very easy to use. I didn’t even look at the documentation, and I got through it quite easily.
The PC based software is very easy to use. However it does take a little bit to get used to the interface, as any new program does. It looks like they spent a lot of time using Borland Windows products, judging by the look of their interface. This is fine since I happen to like Borland products a lot.
The software is very precise and easy to use. The HP based software always worked, and I had no problems with the PC based software. The reliability of Performance Gallery rates pretty darn high.
Your only real concern with performance is if the data collector is making a significant hit on your CPU when it get’s it’s data. Since I had a copy of SOS anyway, I used that to see what the hit was from the data collector. It was marginal at most, but it does cause a little spike when it wakes up and gathers data. The other portion is the host based program that gathers the log files and parses them out to your download file. On my system this process was kind of slow, but it didn’t really cause much of a hit on the CPU.
The PC based software zips right along. Graphs are generated very quickly and what appeared to be accurately. Since the PC hardware requirements for Performance Gallery are so meager, I don’t think anyone will have a problem with this.
Support and Documentation
The documentation is well written for the most part, but there is one kind of odd chapter called ‘Getting Started’ (Chapter 2), that really isn’t a getting started sort of section in my opinion. It is more of an overview of the PC based part of the product, and sort of walks you through it. The problem is, until you have collected data, it doesn’t do you a lot of good. Since it doesn’t supply a specific sample file for you to work on, it is only marginal as an intro walk through guide. I would think it would be better as a separate document that was a true ‘Getting Started’, with a specific sample data file that walked you through the options available. Other than Chapter 2, the manual flows in the right direction.
The appendix at the end of the manual gives you a comprehensive list of each data item that can be collected. It is sorted by both the data item name, and the data item description. This is a good quick way to know what data is available to you. They also include a glossary of terms as well showing keyboard shortcuts and a Quick Reference Guide. All of this makes for an easy time in finding any information quickly and easily.
The support from Lund Performance Solutions is very good. Everyone is very friendly and helpful, and seem to have quick answers. The only real questions I had were with some of the set up, the product is pretty darn straight forward, so support isn’t a real big issue. Maybe that is why they charge so little for support.
This is just a downright fun product to use. The simplicity of generating graphs and stepping through time frames with the scroll bar, make it so simple. Being able to quickly step through
various types of graphs makes it quick to scan your system in a complete fashion. The Print facility generates a very attractive representation of the graph that you see on the screen as well.
Performance Gallery is a nice, easy to use product. I have tried other products like this, and they tended to be very cumbersome to use, not to mention more expensive. The price of the product is
modest enough, and the information helpful enough, that it should be able to pay for itself. Once you know where some of the pulse points are on your system, you can start doing some tuning.
Sometimes it’s as simple as tuning the dispatcher queues, sometimes it’s changing how files are distributed across the disk drives. Sometimes it’s time to buy new hardware, but at least you can make an informed decision about it, and back it up with charts and graphs that the money people can understand.
So I like Performance Gallery, if you need to look at your performance then I would recommend that you check it out. It would be neat if you had something like this in a Client/Server real time performance monitoring tool with drill down capability. Maybe we will see something like that in a few years.
Performance Gallery version A.12
Lund Performance Solutions
440 First Ave. S.E., Suite 3
Albany, OR 97321
FAX : 503-926-7723
Call, write or FAX to receive a demo of the software. Includes PC software on both 3.5 and 5.25 inch diskettes, HP-3000 software on any tape format you require, one indexed 119 page manual. Software price is $1995, support is $325 first year, $175 per year after, price and support is for any size machine.