1992 LISA Time Expenditure Survey Rob Kolstad Jeff Polk and I administered the second time expenditure survey at 1992 LISA. Its primary goal was to learn where system administrators spend their time. The 1991 survey contained too few categories; the 1992 survey contained 54 categories and that seems like enough (since few people wrote in other places their time was spent). Of course, one must be careful about interpreting results of these surveys. The respondents were self-selected. They tended to be members of larger shops. With a dozen exceptions (e-mail responses), the respondents were among those who attended the LISA VI workshop. As administered, the broad categories of the survey included: S/W [14 items] Network H/W [4 items] Non-Network H/W [5 items] Users [2 items] Filesystem Management [4 items] Technical Ongoing [16 items] Less Technical Ongoing [9 items] The survey also attempted to collect site demographics. Results were tabulated in a `reasonable' way. Decimal answers rounded to two places to right of the decimal point (i.e., granularity of 0.01 hours). Most people rounded their replies to the nearest quarter hour. I ignored responses like `0.001 hours/week' The responses were `average hours per week spent on tasks in this category'. Now, this is very difficult to compute in reality. Consider the LISA conference at 40 hours/year -- an average of ~0.8 hours/week. Remembering how much time was spent configuring a recalcitrant printer last March is a tough task. Here's the survey as it was administered: ------------------------------------------------------------------------------ 1992 System Administration Time Expenditure Survey Please return to Rob Kolstad (kolstad@bsdi.com) Please estimate the average number of hours per week that you spend on the tasks below (please record hours or hours and tenths). Be sure to add (at the end) items on which you spend time but that this survey overlooks. Note that if you spend 1 week per year at conferences, that's just under 1 hour per week, on average. S/W: ____a Install/Configure Application S/W ____b User assistance: Application S/W ____c Install/Configure System S/W ____d DB maintenance ____e User assistance: System S/W ____f Testing: S/W ____g Sendmail & other mailer problems ____h BIND & Name maintenance ____i Passwd file/YP/NIS maint. ____j News system maintenance ____k UUCP maintenance ____l S/W Development Network H/W: ____m Install/Configure Net H/W ____n Troubleshoot/Repair Net H/W ____o User Assistance: Net H/W ____p Testing: Net H/W Non-Network H/W: ____r Printer install & maintenance ____s Install/Configure Non-net H/W ____t Troubleshoot/Repair Non-net H/W ____u User Assistance: Non-net H/W ____v Testing: Non-net H/W Users: ____w Dealing with Users (questions, quick fixes) ____x Training ____y Backups ____z Recovers ____a Disk/storage management ____b FTP'ing files Technical Ongoing: ____c Writing Documentation ____d Reading & Writing E-Mail ____e System & Network Monitoring ____f Performance management ____g Ordering stuff ____h Reading books/trade mags/technical lit. ____i Writing books/trade mags/technical lit. ____j Conference Travel ____k Other Personal Growth ____l Planning for the future ____m Policy issues (write, implement, police) ____n Proposal writing ____o Reading/writing net-news ____p Attending to Security issues Less Technical Ongoing: ____q Dealing with Vendors (& Prod. Evaluation) ____r Demonstrations ____s Management of subordinates ____t Management by superiors ____u Attending Meetings ____v Public Relations ____w Corporate Travel ____x Miscellaneous Other things: ____ ____________________ ____ ____________________ ____ ____________________ ____ ____________________ ____ ____________________ ____ ____________________ ____ ____________________ For the section below: ``At your site'' means in your immediate purvey. Please use a consistent definition of ``your site'' throughout the answers. How many full-time administrators are at your site? a _______ How many full-time-equivalent users are at your site? b ______ [A 2 hour per day user is 0.25 full-time-equivalent users] How many servers/mainframes are at your site? c ______ How many dataless workstations are at your site? d ______ How many diskless workstations are at your site? e ______ How many data-ful workstations are at your site? f ______ How many X terminals are at your site? g _____ How many PC's are at your site? h ______ How many MAC's are at your site? i ______ How many total GB's of disk are on your servers? j ______ How many total GB's of disk are on your workstations/PCs/MACs? k ______ ---------------------------------------------------------------------- Data Entry & Validation of Responses All responses were entered into files and converted to standard format. Responses culled to include full time system admins with `reasonable' averages. Responses which stated >70 hours/week (average work week!) were removed; I don't believe very many people average >70 hours/week. Several people clearly reported percentages rather than hours. I also removed part-time administrators, who reported <30 hours/week. Validation software attempted to verify that entries were entered correctly; few errors were detected (and those were corrected). Ultimately, 152 usable responses resulted. Tabulation The tabulation proceeded as with the 1991 survey, but the results were not nearly so interesting in certain ways. For instance, hardly any correlations were found between one category and another (-0.2 < r < 0.2 for just about every pair of categories). Means of the various categories are reported below. Software is included that will create nice histograms of the way the data spreads. No variances/standard deviations were calculated (oops). There were three correlations of note (i.e., r>0.5): 0.54 for Full-time administrators vs. Data-ful workstations 0.72 for Data-ful workstations vs. GB's disk on clients 0.80 for # administrators vs. user/admin ratio It's hard to draw conclusions about the first and the third, though the second is really no surprise!. Time Expenditure Re-categorization Because I thought vendors and managers perceived certain ways admins spent their time, I organized classifications for time expenditure into: Directly helping people System Maintenance Personal development Installation/Configuration Ensuring/implementing the future Management S/W Development Miscellaneous Backup & Restore Of course, you might choose a different grouping! Results (main category, then hours for subcategories) ---------- Directly helping people [17.8 hours, 37%] 4.7 Reading & Writing E-Mail 4.6 Dealing with Users (questions, quick fixes) 2.5 Attending Meetings 2.1 User assistance: Application S/W 1.9 User assistance: System S/W 0.8 Training 0.7 User Assistance: Net H/W 0.5 User Assistance: Non-net H/W System Maintenance [11.2 hours, 24%] 1.6 System & Network Monitoring 1.1 Disk/storage management 0.9 Passwd file/YP/NIS maint. 0.8 Writing Documentation 0.8 Troubleshoot/Repair Net H/W 0.7 Sendmail & other mailer problems 0.7 Testing: S/W 0.7 Printer install & maintenance 0.6 Attending to Security issues 0.6 Performance management 0.6 Troubleshoot/Repair Non-net H/W 0.4 BIND & Name maintenance 0.4 FTP'ing files 0.4 News system maintenance 0.3 DB maintenance 0.2 Testing: Net H/W 0.2 UUCP maintenance 0.2 Testing: Non-net H/W Personal development [5.2 hours; 11%] 1.7 Reading books/trade mags/technical lit. 1.6 Reading/writing net-news 1.3 Conference Travel 0.5 Other Personal Growth 0.1 Writing books/trade mags/technical lit. Installation/Configuration [4.5 hours; 9.5%] 1.8 Install/Configure System S/W 1.8 Install/Configure Application S/W 0.5 Install/Configure Non-net H/W 0.4 Install/Configure Net H/W Ensuring/implementing the future [3.5 hours; 7%] 1.0 Dealing with Vendors (& Prod. Evaluation) 0.9 Planning for the future 0.7 Ordering stuff 0.6 Policy issues (write, implement, police) 0.3 Proposal writing S/W Development [2.3 hours; 5%] 2.3 S/W Development Management [1.7 hours; 4%] 1.1 Management of subordinates 0.6 Management by superiors Miscellaneous [1.3 hours; 3%] 0.7 Miscellaneous 0.3 Corporate Travel 0.2 Demonstrations 0.1 Public Relations Backup & Restore [1.2 hours; 3%] 0.9 Backups 0.3 Recovers Summary of Results Directly helping people [17.8 hours total, 37%] System Maintenance [11.2 hours, 24%] Personal development [5.2 hours; 11%] Installation/Configuration [4.5 hours; 9.5%] Ensuring/implementing the future [3.5 hours; 7%] Management [1.7 hours; 4%] S/W Development [2.3 hours; 5%] Miscellaneous [1.3 hours; 3%] Backup & Restore [1.2 hours; 3%] AVERAGE TOTAL REPORTED HOURS: 47.5/week My Personal Conclusions Sysadmins spend over one third% of their time in directly answering questions and performing quick fixes. Maintenance (real ``system administration'') only requires 25% of the total time! Sysadmins spend an incredible 11% of their time in personal development (if you count news reading as personal development, which I did). Note that the 5.2 hours/week average represents 2/3 of the time that administrators exceed the `standard 40 hour wrok week'. Installation and configuration is a significant expenditure but not overwhelming at 9.5%. Planning for the future is not nearly so bad as one might imagine at 7% (neither too high nor too low in my opinion). Management is a relatively low 4%. Backup & restore, at 3%, is simply not the `big problem' at the sites surveyed. Total time is 47.5 hours/week (in my selective sampling) -- certainly a good return for employers. System administrators divide their time -- every week -- among an astounding variety of tasks. I can't imagine other careers that require so much context switching. These general tasks account for 97% of admin's time Some Comparisons With 1991 Survey 1991 Survey averaged 42.6 hours total -- 4 less than 1992 -- but that included part timers. 1991 admins reported 25% time dealing with users; 1992 report is up to 36%. Backup and restore is down from 1.8 to 1.2 hours/week; automation is surely responsible. Site Statistics These averages seem to portray rather large sites: 4.6 Full-time administrators 368.8 Full-time-equivalent Users 15.8 Servers/mainframes 90.6 Dataless workstations 10.2 Diskless workstations 83.0 Data-ful workstations 21.2 X terminals 393.3 PC's 72.6 MAC's 143.3 GB's disk on servers 58.4 GB's disk on clients Note the incredible amount of disk storage! This is really quite amazing. Likewise, there's little doubt that PC's and MAC's comprise a tremendous amount of computing power out in the world. Full-time equivalent users was calculated as total contact hours/40. Regretfully, we should have asked for both contact hours and for the number of people. This would have enabled better calculations of number of admins per real user. Number of Admins per User I calculated the user/admin ratio. Then I removed 9 ratios for readability in the graph below:225, 233.3, 250, 250, 250, 266.7, 375, 400, 1000. A total 128 valid values remained. The Mean was 82 users/admin; here's the distribution: * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * 18| 19| 23| 15| 9| 11| 6| 8| 2| 4| 2| 1| 3| 0| 8| 0/ 13/ 27/ 40/ 53/ 67/ 80/ 93/106/120/133/146/160/173/186/200 The mean has declined since 1991 (was 89 -- but 1991 removed data over 100). It's easy to see that different sites have different ratios. I reckon the ratio depends on services provided -- more services means fewer users/admin. Note that it's difficult to point to a number and say, ``oh yes, 83 more users means you get another admin''. I recommend that sites deduce their requirements and tasks and then hire more administrators rather than try to fix the ratio at some magic constant. Summary & Recommendation This survey appears to present the most accurate (so far) set of data about what admins do with their time. But -- it's quite skewed with the respect to the self-selected sample. Furthermore, the numbers presented are recollections and estimates. I suggest that we, the system administration community, should hold a national `Sysadmin Time Expenditure Week' during which hundreds (thousands?) of sysadmins across the country (world?) track their time usage accurately for exactly one week. Then, we collect results via e-mail, quickly tabulate them, and share them with everyone! I propose February of 1994 for a good date to begin discussion.