Soho Shangdu is an office building located in Guomao, Beijing China. The flickr link is http://www.flickr.com/search/?w=all&q=soho+shangdu&m=tags
(This was written in 2009 - things may have changed)
The elevators in the south tower of Soho Shangdu are horrible. Dealing with them is 15-30 minute ordeal every day. They are annoying because:
There are 4 elevators in the south tower. Two are on the south side, two on the north side of the lobby. Not all elevators serve all floors. The north elevators serve B4, B3, B2, and 1, 5 to 30. The Southeast elevator serves B4-B1, 1-32. The Southwest elevator serves B4-B1, and 1-31. The Southwest elevator is the "service" elevator, and the walls are covered with scratches. Incomprehensibly, even during elevator rush hours it is used by service staff to carry bags of garbage and water bottles.
The elevator panels contain buttons for each floor, and door open and close buttons. Since each elevator serves different floors, the panels are all different - this means that every time you get in, you have to look closely to see the button to press. The elevators are also extremely dim - there bulb looks to be about 20 watts. This means you can't see the buttons easily.
What is the most common destination floor? Do you think that the floor which represents about 50% of all destinations should be easier to find than buttons which represent approximately 2-3% of all destinations? The '1' button is mixed in with the other buttons - it's not any larger, not marked as different in any way, and is very low on the panel, with the basement floors below it.
For some reason, the floor buttons are also different in each bank of elevators. This is due to the missing floors.
This prevents habits from developing - every time you get in, you have to look. Because of the dimness and moving buttons, residents frequently push the wrong button - about 5% of the time. This generates even more useless stops.
What is the most frequently pressed button in an elevator in China? The 'Close Door' button is pressed almost every time someone enters an elevator. This button is located in the most inconvenient position of all buttons - on the very bottom, next to the open door button, which looks almost identical.
The elevator software does not read this button at all while the doors are opening - it begins to read it only about 0.5s after the doors fully open. There is also a buffer - presses within about 1s of an unsuccessful press are ignored. This does not stop people from incessantly pushing the button, actually slowing themselves down.
The first stupid thing about the elevators is the call buttons; there are independent call buttons for each pair of elevators (north pair, south pair). On the ground floor, this isn't such a problem - in order to get an elevator on the ground floor, it is necessary to push both "up" buttons. This is usually acceptable, because even though you call two elevators, one will come first and get you, and by the time the other one arrives, there will probably be more people waiting. However, on upper floors this causes a major problem. On the 20th floor, to get an elevator down people typically push both down buttons. One north and one south elevator will eventually arrive to meet this call. The caller takes the first elevator to arrive, and the second one usually arrives soon after; soon enough that it is almost never needed. Whether it is empty or not, it is a waste of time.
An elevator moving up or down will stop at a given floor if there is an unanswered call on that floor for an elevator going that direction; so as you are going down from the 20th floor, your elevator will stop at the floors you pass on the way down if someone has pushed "down". However, if they have pushed up, it won't stop.
When an elevator is deciding where to go, it will first try to serve requests which are going in the same direction it would need to be going to get there. For example, an elevator on the ground floor which can see a request to go up from the 10th floor and a request to go down from any floor would go to meet the request to go up. If there are no requests to go the direction it would need to go to arrive at the call, an elevator does something interesting - it serves the farthest request to go the "wrong" direction. This means if an empty elevator on the ground floor sees requests to go down from the 30th and 20th floors, it will go to the 30th floor first.
Think what happens when everyone is leaving the building at 6pm: There are down requests from almost every floor, and elevators deciding what to do have all just emptied on the ground floor. Naturally, they all go up to the top floor they serve! Once they get there, they head down, stopping at every floor. Because of the disconnected elevator buttons, they will literally stop at every floor, so at about half of their stops, no-one gets on.
You can imagine how annoying it is to be going down and have the elevator stop at every floor, even though half of the time nobody gets on or off!
A partial solution to this is...
When elevators are full, they zoom down to the ground floor, not making any stops. This is really nice. The trigger for an elevator being "full" is a minimum weight - but that weight is about 2-3 people less than the weight of a full elevator. So before elevators are genuinely full, they zoom down.
This feature has been noticed; the elevator measures the weight about 1 second after the doors close. By a coordinated slight jump of occupants, timed to all fall at the right moment, the elevator can be tricked into thinking it weighs more than it really does, and zooming down.
It is possible to overfill an elevator - in that case, it begins a horrible loud beeping. People need to get off to get it well below the beep weight - but after getting off, you can go right up to the beep weight again. So usually when it beeps, two people get off, and then one sneaks back on and we go.
Since elevators start their downward run from the 30th floor, stopping at every other floor, and picking up people every other stop, eventually they fill up. This usually happens somewhere between the 20th and 25th floor. But since there are usually two elevators serving each request (because of the split north/south), they fill up half as fast. You spend a lot of time going down floors one by one, never getting quite enough people to zoom down. Across the elevator lobby, you can see your partner elevator that's stealing your weight.
Once full, they zoom down, completely skipping all intermediate floors. This is frustrating for the people in the intermediate floors.
Someone in one of the zoom-past areas once had a bright idea - even if you are going downwards, you can steal the elevators on their way up if you push "up"! Then this fresh, empty elevator stops on your floor and you can get a ride no matter what. Although it's still locked into answering a down call from a high floor, at least it won't zoom past you on the way down. Better yet, if you can fill it up completely, you can instantly zoom to the bottom, saving about 10 or 20 minutes of waiting.
Applied building-wide, this strategy results in every floor having all buttons pressed at all times. People will go out and push up and down on both sides. When an elevator on the way up stops, it picks them up, and continues up to the uppermost down call. On the way back down if it hasn't filled up, it will stop at the same floor.
But this strategies issues 4 calls for every ride - generating a large number of extraneous calls and empty stops.
This strategy also demonstrates the first of many incredibly stupid, unthinking behaviors of building residents. In the elevator lobby on each floor the current location of each elevator is displayed. For example, all elevators are frequently above floor 25; anyone on the 20th floor can see they are all above, and are slowly moving down filling up or just stopping to meet a call that isn't useful anymore. Even in this situation, stupid people still push all 4 buttons; even though there aren't even any elevators yet to satisfy those "up" calls - it is known that all elevators are occupied on "down" calls, and are also above the current floor anyway. So it is provable that pushing the down buttons is sufficient to maximize the chance of getting an elevator; there is absolutely no use to pushing "up" while all elevators are above you; and given that every elevator has one person who will bring the elevator a sufficient distance below your floor to give you time to see that it has passed you (a very likely state of affairs), you will always have time to push up and attempt to steal the elevator later, in case none of the 4 above you make it to your floor with empty slots.
So this selfish strategy, applied incorrectly, results in even more useless elevator stops; even with maximum selfishness there is not any benefit to pushing up while all elevators are above you - unless you count the labor you save by not having to watch.
Note that the stealing strategy, applied correctly, can actually improve elevator efficiency in some cases. For example, from the 20th floor, if all 4 elevators are ascending but are below you, it is likely they are all going to the 30th or 31st floor. Without action from the 20th floor, it is likely they will all stop at the same floor or within one floor (30/31, since the maximum floor for north/south elevators is not the same). When arriving, north/south will split occupants and neither will fill up. So the initial stop will be 32,31 on the south side, 31,30 on north. South 32 will descend to 30 since 31 already has an elevator, South 31 will go down to 29. On the North , 31 will go down to 29, 30 will go to 28. Each side's elevators will descend by 2 floors at a time pacing each other, leapfrogging floors, and delaying any of them filling up.
If a steal at an earlier floor manages to fill up and zoom an elevator down, the 3 remaining elevators at higher floors will fill up faster; overall human time will be saved from the increased number of zooming elevators. Also, delaying only one of the elevators on the way up will allow the double down pushes to have a higher chance of picking someone up.
The organizational strategy this demonstrates is that when you are designing an infrustructure, you have to make sure that selfish strategies which hurt the average wait time do not exist.
The elevator situation would not be so bad, except for the basement of the building. The building has 4 basements. Users in the basement are car-drivers or support staff; either high or low class. In both cases, even less likely to be considerate.
The ground floor is the most frequent floor for elevators to stop at. Because of the splitting north/south, typically two high elevators will be called to the same floors on the way down. Since they are splitting riders equally, they will tend to fill up at the same time and zoom down, arriving at 1 at about the same time. Since during the long time above, someone in the basement has almost always pushed both up buttons, both elevators will head downwards. The elevators will stop at 1 to let people off, but will still show "down" to pick up the basement guys.
If there was a 3 or 4 splitting going on (or any splitting involving two elevators on the same side), two elevators on the same side will arrive at the bottom at about the same time. Here is a typical scenario: Two elevators are heading down on the same side, one is at 4 and one is at 3. The one at 3 arrives first, but since it's the first to arrive down, it displays it's current direction as "down" when it arrives, to pick up the basement guys who won't even walk up one flight of steps. Even if the ground floor is full of people waiting (and sometimes there are 40-50 people downstairs waiting for elevators), they are reluctant to get on this down elevator because it's marked "down" and another is arriving in a second. As it is emptying out, the other elevator arrives and claims to be going up, slowly emptying out. The first one arrives at the basement; usually just one person gets on. He pushes up. While he was getting on, the other ground floor elevator was filling up. The "up call" button on the ground floor turned off since an up elevator was there already. The elevator from the basement closes and starts to move up. Since it detects an elevator already present on the ground floor, it does not stop there; although everyone in the lobby can see it passing by, they can't push the call button, because it shuts off instantly because the other elevator is already present, although mostly full. Usually the one open is undergoing an overload/get off cycle; or, people are finally being allowed to get out after having to push through people who insisted on getting on/blocking them before anyone was allowed to get off. The elevator with just 1 person in it heads up to the destination, even though many people were waiting on the ground floor.
This happens frequently; about every other day as I arrive in the morning, elevators pass from the basement upwards without stopping.
Longer-term residents of the building know this behavior, and often get on downwards moving elevators, anyway.
if two elevators on the same side arrive at almost the same time, the first one usually decides to go down and the other to go up. However, if the first one's doors can be kept open long enough for the second one to start moving, the second one will reconsider its options and realize there is an up call from the basement; sometimes it will change its direction and head downwards, and the people in the first one will find themselves zooming upwards, with only a few people in the elevator.
Some theorems on elevators, presented without proof:
Considering the low cost of energy related to the salary/rent in offices in this building, human time is important. i.e. the total building cost of running elevators is <1 kuai/m^2/month, and rent is 150 kuai/m^2 or more.
Considering the expense of the building and the cheapness of electricity, it is surprising that the elevators do not implement some obvious improvements to human time at the expense of some energy: for example, moving empty elevators to useful areas. (Proof of the environmental acceptability of trading energy for human time: The building is lit up at night, for no reason. There are many other obviously useless uses of energy around the building. So arguing that elevators should save energy at the expense of human time, while ignoring energy uses with no appreciable benefit cannot be correct.)
In general, if an elevator is empty and has no calls, the most probable floor to need it is the ground floor. To improve average response time, it makes sense to move there. For banks of 4 elevators, there are even better solutions. Perhaps keeping 2 empty on the ground floor, and 2 others at intermediate floors.
Including time optimizations would allow even greater improvements; during the morning hours, empty elevators should default to return to the ground floor. After 5 pm, there is not as much need to keep empty elevators on the ground floor; starting from a higher position would on average save human time.
There are two forms human time optimization can take. First, we should try to minimize the average time it takes to get where you are going. A second important factor is having a predictable wait time; if it takes 1 minute on average to get an elevator, but one out of every 10 times it takes 5 minutes, it isn't as good as if it always took exactly one minute (or even 1:30). In fact, variation in service time (in IT literature, this is known as "Jitter" - the variance in response time) can sometimes be more important.
For example, emergency services - would you rather have an ambulance service that takes on average 30 minutes to get someone to the hospital, or one which takes on average 29 minutes, but where 1 out of 100 ambulance calls takes 2 days? It is possible to have ambulance performance sets that meet those characteristics, and depending on the connection of mortality rates to hospital arrival times, the 1/100 chance of not arriving for 2 days may cause more deaths than would be saved by arriving one minute earlier on average. In addition, predictability in arrival times would allow for efficient stocking of supplies in ambulances; with zero jitter, they would never need more than 30 minutes worth of every supply. This would save storage, weight, gas, usage costs, improve scheduling problems, etc.
So clearly, average time is not the only value - if variance gets high enough, it can be worse than increases in average response time.
In the past, the building managers have tried various strategies to solve the problem.
The two sides were apparently linked in the past, but after a week, the problem was even worse, so they reverted. I think what probably happened was that the basement-skip problem became even worse; since the ground floor could only have one active call at any time, and elevators don't return to the ground floor by default, one elevator would be returning to the ground floor and the other 3 sitting doing nothing, even if 3 elevators worth of people were waiting on the ground floor.
South elevators go to floor 21+, north to 20+. This is good, except that there is just one unified line; constant line-jumping, and the attendants allow people to leave the line and enter the lobby and mill around well before the elevator arrives; this contributes to an atmosphere of lawlessness & mistrust.
Each elevator also has an attendant who pushes the buttons and prevents the elevators from going outside of the zone. Unfortunately, riders frequently get in the 2-20 elevator, and then beg the attendant to take them above 20. The attendant usually gives in.
Assigned each elevator as either odd or even, and restrict it to only serve those floors
This would increase the value of each call - the average number of people entering/exiting on every call would double. This would lower the useless stop problem. Proof: As of now if you are going to an odd floor, you can go in any of 4 elevators. if it were like this, you could only get into two elevators. The proportion of odd floor people in that elevator would be doubled; therefore, for a given odd floor, the average number of people exiting on that floor would be double.
This would also improve the worst case scenario, when a nearly full elevator stops at every floor on the way down, but all the calls are dead.
There's no reason that the elevators on both sides should go to the basement; the basement is doubly served now, even though more than one person almost never gets on at one time there. Most elevators visits to the basement are visits on both sides; since only one person gets on, cutting out half would not hurt much. So, one side of the elevators should completely stop going to the basement.
At least try to make people feel bad about wasting calls. Although it would have low percentage effectiveness, truly rational people who are just not introspective can learn things that they wouldn't have though of themselves. Allowing people who can appreciate correctness, but who are not curious to become more moral is a good thing.