4 ms·

Thanks for sharing. 9 hours to run payroll at its best. Wow.

Jtsummers 5y ago
Depending on the server or mainframe hardware it was being run on, databases, network connectivity, interactions with financial institutions, and potential physical activities (printing checks and such), 9 hours may not have been unreasonable in the 1990s. Digging around on C2: ChryslerPayroll (Essentially a batch app) db size: 2.5 - 4 GB server: Sun Enterprise 2000 4GB RAM interactive users: 5 So not exactly a beast by modern standards.
- WJW 5y ago
  Still, according to wikipedia it paid around 9000 people. This would make it run at about 1k/people per hour or about 3.6 seconds per person. Now it's true that the complexity of payroll calculations is often underestimated but this seems slow even for 1993.
  - Jtsummers 5y ago
    Given disk speeds at the time, it seems the DB was residing on the box, I wouldn't be shocked by a speed like that on a single or even dual CPU machine. And that's assuming all necessary data was in residence on the machine when the batch job started, and no network connectivity was needed (like retrieving that period's time sheets from legacy COBOL systems or pushing out the results of a paycheck calculation to a legacy printing system). More on its design: The project is essentially a batch-processing system. When it is time to pay, it cycles over all employees, computing the pay due, and exporting results. There are about 15 main external export receivers and a half- dozen different input sources. There are around a hundred different input transactions. http://wiki.c2.com/?ChryslerPayroll http://wiki.c2.com/?ChryslerPayroll EDIT: http://wiki.c2.com/?WhyIsPayrollHard http://wiki.c2.com/?WhyIsPayrollHard Talks more about the scope of what was being done. 3000 calculations per employee, formatted and transmitted to various COBOL systems from the legacy payroll. 3.6 seconds isn't exactly stellar, but on hardware from the mid-1990s it's not obscene either. It's really hard to judge systems based just on a performance number without knowing details about the actual hardware and network configuration and the systems it communicates with. If I tell you I can sort 1000 item lists in 10 seconds, that sounds bad. But if I then tell you it's an 8-bit microcontroller running at 1kHz and getting data over a 300 baud modem, it suddenly sounds reasonable.
  - flerchin 5y ago
    1997
- flerchin 5y ago
  The wikipedia article says that it paid 9k people. So 1k paychecks per hour. 3.6 seconds per paycheck. Agree that this would make sense if the limiting factor was the printer, but even so, parallelize the print job to multiple printers.

Depending on the server or mainframe hardware it was being run on, databases, network connectivity, interactions with financial institutions, and potential physical activities (printing checks and such), 9 hours may not have been unreasonable in the 1990s. Digging around on C2: ChryslerPayroll (Essentially a batch app) db size: 2.5 - 4 GB server: Sun Enterprise 2000 4GB RAM interactive users: 5 So not exactly a beast by modern standards.

WJW 5y ago
Still, according to wikipedia it paid around 9000 people. This would make it run at about 1k/people per hour or about 3.6 seconds per person. Now it's true that the complexity of payroll calculations is often underestimated but this seems slow even for 1993.
- Jtsummers 5y ago
  Given disk speeds at the time, it seems the DB was residing on the box, I wouldn't be shocked by a speed like that on a single or even dual CPU machine. And that's assuming all necessary data was in residence on the machine when the batch job started, and no network connectivity was needed (like retrieving that period's time sheets from legacy COBOL systems or pushing out the results of a paycheck calculation to a legacy printing system). More on its design: The project is essentially a batch-processing system. When it is time to pay, it cycles over all employees, computing the pay due, and exporting results. There are about 15 main external export receivers and a half- dozen different input sources. There are around a hundred different input transactions. http://wiki.c2.com/?ChryslerPayroll http://wiki.c2.com/?ChryslerPayroll EDIT: http://wiki.c2.com/?WhyIsPayrollHard http://wiki.c2.com/?WhyIsPayrollHard Talks more about the scope of what was being done. 3000 calculations per employee, formatted and transmitted to various COBOL systems from the legacy payroll. 3.6 seconds isn't exactly stellar, but on hardware from the mid-1990s it's not obscene either. It's really hard to judge systems based just on a performance number without knowing details about the actual hardware and network configuration and the systems it communicates with. If I tell you I can sort 1000 item lists in 10 seconds, that sounds bad. But if I then tell you it's an 8-bit microcontroller running at 1kHz and getting data over a 300 baud modem, it suddenly sounds reasonable.
- flerchin 5y ago
  1997
flerchin 5y ago
The wikipedia article says that it paid 9k people. So 1k paychecks per hour. 3.6 seconds per paycheck. Agree that this would make sense if the limiting factor was the printer, but even so, parallelize the print job to multiple printers.