Usage patterns before introduction of Green IT
The background information gives us broad bounds on the possible energy usage were we to attempt to adopt an initial position of shutting down machines out of hours. In practice the lower bound would not be achievable. There are a number of factors that affect this, most notably
- Users who leave computations running on their desktop regularly
- Users who rarely logout (i.e. they wish to preserve the desktop context for when they return, i.e. applications open etc)
Random checks show that at present 30-50% of the departmental desktops have users who rarely log out for various reasons. If this is because they are running a computation then the main alternative open to them is to run such processes on the department's computational servers. At present many people using the compute servers may remotely connect from a desktop machine and stay logged into that desktop too in order to monitor the running computation. For these users we can document and encourage the use of screen which will enable them to disconnect and reconnect to the compute server and still access the previous session and hence check on the computation as required whilst still logging out of the desktop at night. Whilst this broadly shifts the electrical usage it can provide a saving as a compute server will be more efficient that a collection of individual machines. For those users that simply stay logged in for convenience (e.g. saves time reopening applications) there are options we can explore, e.g. the machine can be put into different states of hibernation or suspension. These use differing lower amounts of energy and have different waiting times before the PC is ready for use again after pressing a button. Before using such an approach tests would need to be done to show that hibernate mode does not negatively affect the use of other tools e.g. version control and locking with files stored on a shared network drive.
In the first instance a reasonable aim would be to impact on people's accepted usage pattern as little as possible (if at all) by simply targeting a setup where machines would automatically come on each morning (say at 8am before most people arrive) and then go off if not in use at 6pm. Waking the machines up each day before most people require them would also allow time for background update processes to run without impacting on the user, e.g. automated security updates, virus checks, or package upgrades.
With this reasonable initial goal we might hope to have at least half of the desktops off out of hours and thus reduce our energy usage by desktops by about 25%.
Beyond the initial target
Once a system is in place and operating well we could explore ways that might save further energy, e.g.
- Power machines up each day at 8am but if not in use by 11am shutdown at that point. This modest extra step for a machine not in use for a week would reduce the energy further from 5.6kWh per week to 2.2kWh per week. Such a change could thus make a difference and yet still have limited impact on the user, e.g if they did arrive later they would only need to wait a minute for the machine to power up.
- Actively encourage people to log out at the end of the day and hence allow more machines to save power.
- Suspend or hibernate machine if someone is logged in but not active. Doing this automatically could be more difficult though. Allowing the user to hibernate a machine and encouraging this could be a more acceptable alternative to the previous option for some and hence still reap savings. However, it would introduce new problems if a user hibernated a machine that was also being used remotely by someone else.
A sensible set of staged aims would thus be
- Implement the initial default on period of 8am--6pm which should have almost no impact on users but will save energy and raise user awareness of green computing issues. Do this first with a test pool (e.g. one building or even just one floor within a building, e.g. DH 3rd floor) and then extend to all desktops.
- Review the percentage of PCs that are in constant use and encourage all users in general to log off each day if possible. Get feedback from those users that regularly stayed logged in to understand their particular motivation/needs.
- It would seem unlikely that step 1 is going to produce any significant negative feedback. As such after a suitable period introduce changes to the default on period from 8am-6pm, perhaps in several stages, to move towards a default on period of say only 8am-11am.
- Step 3 is likely to produce more feedback and awareness and thus some modification to the approach in some situations may be needed at this point.
- Depending on the feedback from step 2 and general acceptance consider shrinking the default on period or even removing it depending on user and system requirements (e.g. may want some default on period for system updates).
- Review all the feedback in more detail to consider further steps to allow more PCs to be shutdown, e.g. providing more/better compute services so long computation are not run on desktops, enable suspend or hibernation function so users could start a machine up again each day and carry on with the applications all open in the same state as the night before.
Whilst it could be possible to start immediately at say step 5, such a major change in operating practice is likely to produce some unnecessary negative response whereas a staged approach should produce steady acceptance at each step and thus achieve the same end technical result but with general acceptance rather than dissatisfaction.