An Example Scenario for Web Print Scaling
Last modified on 11 August 2020 10:46 PM
In version 16.2 of PaperCut NG and PaperCut MF, we introduced the ability to run multiple Web Print servers at the same time, improving throughput for Web Print jobs, as well as increasing resiliency. Our user manual describes the recommended process for deciding how many Web Print servers to deploy. This Knowledge Base article expands upon the manual with an example scenario, showing how the process might look in action.
West Face University has been using Web Print in Sandbox mode with a single Web Print server for some time. This Web Print server is a virtual machine, and supports Microsoft Office and Adobe Reader. The IT team have noticed that students and other staff sometimes complain that their Web Print jobs take longer than expected to print, particularly in the afternoons. Not only that, but they once experienced an outage of their Web Print server, which prevented any Web Print jobs until the outage was resolved.
Initially, they deploy one additional Web Print server virtual machine. Microsoft Office is not installed on this machine, as the IT team currently do not have any available licenses, and would like to be sure that more licenses are needed before submitting a purchase order. A monitoring tool is set up to track the number of pending Web Print jobs, so that further adjustments can be scoped out.
The monitoring tool reveals that the number of pending jobs is often below five in the early hours of the morning, only intermittently jumping up above this level. However, the number climbs sharply in the early afternoon, reaching as high as 20, and remaining above 10 until evening. They sent some small test print jobs during this peak period over a number of different days. These test jobs seem to take on average of around 10 minutes to finish processing, which immediately strikes the IT team as excessive, given that they know the students often walk a fair distance between rooms to collect their prints before returning to their areas of work. If their prints are not waiting for them when they get to the printer, this would be quite frustrating for them, and would explain many of the complaints they have been receiving.
They also ran a few “Print logs” reports to determine the total number of Web Print jobs processed during the peak periods. The average number of jobs submitted during this peak each day is about 60. With two Web Print servers handling 60 jobs, they now know that each server is likely taking around 30 jobs during this period. With a suspected average wait time of 10 minutes per job, it could be estimated that 300 minutes (or five hours) is how long this peak may last, which roughly fits with what has been observed.
They would like to decrease the average wait times during this peak to five minutes. 60 jobs across five Web Print servers seems like a good bet; each server would have 12 jobs to process during the afternoon, and a 2.5 multiple increase in available Web Print servers should bring average wait times down by at least somewhere around half.
But before introducing more Web Print servers, it is also considered that only one of the current Web Print servers has Microsoft Office installed in order to handle Office document printing. More “Print logs” reports covering the peak printing periods are generated, this time with distinct reports being created for common Web Print filetypes, such as “.pdf” and “.doc”. Upon inspection, it would appear that close to 60% of all jobs commonly submitted during the peak printing period are Microsoft Word documents. As three-fifths of these jobs require a Web Print server capable of printing Office documents, two additional licenses are purchased, so that Office can be installed on to three out of the five planned Web Print servers.
The three additional Web Print server virtual machines are now deployed. Initially, average wait times only drop to around eight minutes, but when the extra licenses are used install Office to two more Web Print servers as planned, average wait times decrease to just under five minutes, meeting the expected target.
However, the IT team know that a large number of complaints about the Web Print service were received during the exam period last semester. Wanting to know if they can expect average wait times to remain stable, more reports are generated, comparing the average number of daily Web Print jobs during past exam periods to the numbers now being observed. They discover that the number of PDF documents being printed per day skyrockets during each exam period.
With the number of jobs increasing by 50% during exam time, and these additional jobs being almost exclusively PDFs, the IT team plans to temporarily provision an additional two Web Print servers at the start of the next exam period. They do not need additional Microsoft Office licenses, and they will then shutdown these additional servers at the end of exams. This way, the extra resources will be used only when needed. They will monitor, track, and alarm the Web Print pending jobs statistic, to anticipate growing needs, and detect any sudden Web Print issues. They will also periodically generate reports on the number of Web Print jobs, knowing that around 60 Web Print jobs with five Web Print servers can be expected to have average wait times of five minutes.