I work as a software developer (BS in Computer Engineering) for a wire saw company. Several of the wire saws for years had a random problem of wire running off the reels, but it was considered a minor problem. A new saw model that was being built was bigger and had multiple cutting yokes. According to the schedule, we had a month; until sales told the customer we could deliver the saw in two weeks (the senior project engineer said we needed two months). The mechanical/hardware people handed over the saw a few days before the shipping date to software where all we had to do was load our usual software (modified), test, and ship to a happy customer.
Then the fun began. Each cutting yoke had unique problems and all of them had run-off problems. Mechanical and hardware assemblers said it was obviously software. Being the only embedded software engineer on this saw, all their eyes fixed on me as the bottle neck in their finely tuned organization. During the meeting, it was decided that everyone had done their job using the same mechanical/hardware as other saws before and did not have these new problems, so it was time to own my job. In a few hours of testing, I came to the conclusion that the same software running on several parallel yokes but displaying different problems was indicative of hardware malfunctions or miss-wires (with several hundred wire screws, this had happened before).
My logic was not accepted, and it was seen as trying to blame others for my buggy software (again, everyone did their job and I was the FNG who replaced an uber engineer, who over the years could do backflips on water while juggling anvils). Over the next few days, I was called to frequent meetings to give updates. After a few rejections by the technicians of my idea to check the wiring, I grabbed a screw driver and after a couple of eyeball scans looking for differences between each yoke's hardware, found several miss-wires in each yoke. Now the software was running uniformly on all the yokes. Upper and middle managers left their meeting room and asked me on the shop floor what I had fixed in my software; I said it was a just a few mis-wires. They looked at each other with even bigger scrunchy faces and ran back into their meeting room (no atta boys).
This still left the reel run off problem that now plagued the saw. Again, after several meetings, I still needed to fix my software problems and ship that saw. By this time management called back the uber engineer and another engineer to get their HMI/data-logging program to work "even better" (no, not buggy at all) and I contributed a few minor "improvements" they missed while interfacing the hardware, but I digress. I was under the microscope to fix the reel runoff problem that I had obviously made worse with my new software and my manager made the casual comment that younger engineers work harder. I was bouncing between the shop floor and my small office/file room making changes and not progressing.
On my office desk I cobbled together (out of replaced parts) a test saw to test my software (management promised engineering their own production saw to develop SW on, but it never transpired). At first, software changes took 20 minutes to load and verify (walking to the shop with an eeprom). After a few months of begging for a debug eprom and an IDE, I was now able to make software changes in seconds. In my office the software worked correctly on the test saw but randomly failed on the production saw. During even more meetings I showed that the problem resided in a single line of code that looked at the ADC to detect wire speeds, but this was questioned by sales--why did this LOC now fail when it has been working fine for five years in all the other saw models? Running the IDE on the production saw would prove my point. Earlier, I had requested a better laptop to run the IDE on the production saw (engineering had an older Pentium laptop while management/sales had P4/Core). Alas, I used the old laptop which resulted in a lot of crashes due to IDE lag times. This impressed everyone when crashes now spewed wire everywhere. The technician (who wired the saw, dropped an oscilloscope, and later broke a laptop) kept moving my laptop while running tests every time it was in his way. Eventually, the IDE showed that the ADCs, even though receiving the correct stable voltages, had a sizable ripple that prevented zero speed from being detected by the LOC I predicted was the one causing the wire to runoff the reel.
Because the test saw did not have this problem, it was a simple process of ceteris paribus that indicated the 18' lengths of display cables caused the ADC's levels to ripple. I shortened the display cables to 6'' and the saw functioned correctly. Upon showing the technician the problem, he commented that customers were not going to walk around and open the cabinet to use the saw displays and it did not look professional. My ennui kicked in and remembered the story he told months earlier of another difficult saw project he worked on that bankrupted this company thus leading to the present management team. After consulting the embedded board manufacturer, another engineer and I designed a display buffer that fixed the problem. The saw shipped, my job description no longer existed in the company, and when parting, my manager said I needed to "own my job." They then hired a son’s young friend to do a "totally different position." He didn't last. Ironically years later, most of those special people were purged. The remaining engineer wrote a recommend stating my HW/SW solutions were still being used in their current saws.
Well, I've worked with enough boneheaded software types who didn't know anything about hardware to say that you are the exception, Tim. Embedded engineers who know both are worth their weight in gold, even if boneheaded managers don't realize it. I totally agree with zeeglen; forget the finger-pointing and just figure out what's not working. Sometimes it's both H/W and S/W!
This reminds me of a time when a watchdog timer would sporadically time out when a processor took too long to execute a command/response sequence on a control bus. This was a new product in both HW and SW, and the late night sessions in the labs were not burdened by "it's a HW fault / no it's a SW fault". We just admitted that nobody knew yet where the fault was and we had to work together to find it.
Finally, using an analog scope (digital scopes had not been invented yet) the SW guy and myself saw an event whiz by that the timeout occurred within the 1 second allocated time of the hardware. I took another look at the hardware, a long-chain ripple counter and realized that the guy who designed this had done the stage count based on a complete cycle at the final stage. He forgot that the timeout actually occurred on the rising edge HALFWAY through the cycle.
Solution: knife and green wire.
Lesson learned: Never assume HW or SW. Test, test, test....
Interesting that in a new system -- new hardware and new software -- what ultimately turned out to be a signal integrity problem was blamed on software. "Mechanical and hardware assemblers said it was obviously software." And management simply took their word for it, without any data to back up that claim?
I've written and debugged firmware for many years and I can really relate to this story. Solving problems like those described in the story take time and a very good engineer. Unfortunately there's no way to measure the difficulty of a problem so there's no way to measure the value of the solution. Managers usually just look at how long you took and draw their own conclusions.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.