Support Sailonline

If you haven't already - join the SAILONLINE YACHT CLUB!

Please also consider making a donation - all amounts are greatly appreciated!

Board » Technical Discussion » Performance loss

Page: First Previous 7 8 9 10 11 12 13 Next

It has been talked about for years, but I think the time for an update to the 'performance' model has finally arrived. The following is my summary and the beginnings of a proposal.

I have just re-read this thread (Performance loss) as well as the original 2008 Wrong Speed VMG topic. For anyone interested in the topic I would recommend reading both in full. The original 2008 topic gave some interesting insights in to the thinking behind the original PL approach adopted, but both then and now there are some major frustrations amongst both new and old SOLers with the model and how it works.

The original purpose was apparently to stop certain behaviours that were (presumably) causing undue load on the server or other problems. Back in 2009 jacob wrote the following:
We introduced this performance loss not to punish normal sailing maneouvers but to prevent boats to tack e.g. once every minute during hours just to ride on a tws that optimizes vmg. The prformance loss implementation succeeded in preventing that behavior and is as stated above not a problem for normal maneouvers.:-)))

BUT, with modern yachts - some of which can do in excess of 40kts - this model has clearly outlived its useful life and become a liability. It also penalizes newbies unnecessarily for simple learning mistakes.

Now, there have been a lot of good ideas proposed including momentum-based models, etc but I really think we need to keep it simple for a couple of reasons. Firstly, the model needs to be intuitive. It should make sense to IRL sailors as well as those new to SOL and to sailing in general. Secondly, it needs to be fair. It should not require intimate knowledge of the internals of the model - which I know and refuse to use - to sail a reasonable race. And thirdly, it should be easy to implement and relatively light on the server end. This one probably won't be so obvious to those without a computer science or engineering background, but it turns out that while linear responses are easy to graph and to understand, exponential responses - the way a lot of the natural world works - are actually really easy to do in 'discrete simulations' (eg. computer simulations such as SOL where everything moves in steps of say 15 seconds).

The basics of my proposal are as follows:

1. Scrap the per-degree PL penalty entirely. There is no basis in reality and I think no value in SOL. Also, newbies need to be able to play with direction changes. Of course, those effectively using autopilots (ie. large numbers of DCs) are another matter entirely and should simply declare their hand for all to see.

2. The 93% limit is a joke and needs to be scrapped. There is no basis for it and it is badly implemented. For instance, today I got down to 80% whereas if I had 'gamed it' then I would have never gone below 93%. This moves the focus from sailing to the gaming engine in which case EVERYONE loses.

Ok. That's two of the pillars of the current model discarded, so what's left?

Tacks and gybes clearly have performance impacts IRL and these need to be reflected in the SOL gaming engine. A yacht can almost stop in a tack but recovery is relatively quick whereas there is often little speed loss during a gybe but recovering any lost speed can take a very long time.

Looking at this from an engineering perspective, apparent wind strength (AWS) seems to be a very important factor here and it may be something that can be used to drive the 'recovery model'. Consider that before, during and after a tack the AWS is higher than it is for a gybe. What's more, in the seconds and minutes after a tack the AWS increases as the boat speed picks up thus increasing the driving/accelerating force, whereas, in the case of a gybe, as boat speed increases the AWS goes down and the accelerating force decreases meaning that it will take longer and longer to reach that theoretical max downwind speed for a given angle.

I propose (arbitrary) penalties of 25% for a tack and 10% for a gybe.

'Recovery' is then a different a matter. It should probably apply to ANY course change. At the time of any course change we know (a) the current boat speed which may of course be less than optimal (b) the theoretical boat speed on the new course according to the polar of the yacht and (c) the apparent wind speed (AWS). Given the appropriate combination of all three of these I think we should be able to develop a model for 'recovery' (or should we call it acceleration?) that makes sense to everyone. BTW, a change to a point of sail on the polar that indicates a lower theoretical speed than current should be immediate. The 'recovery model' should only be applied when a positive change in speed is indicated.

Should 'recovery' etc depend on the weight of the boat etc? Perhaps. Do we have this information today? No. Could we get this information? Yes. Where? I have some thoughts on this that I would be happy to share.

If we can get some agreement on these ideas then I will happily look at coding it ... with the agreement and assistance of hmmm and the SOL management team, of course.
Very quick reply because I'm running late:

Your insight that AWS is more important than TWS for performance loss is wonderful!

We have thrown some ideas around (via mails) and I think that, let's call it, "mechanical performance loss" is not the main part of the distance/time lost during a gybe or tack or anything else. It's moving sails and other weight around, many gybes/tack takes an impact on the crew.

A penalty for non tack-changing manoeuvres should stay. This mainly simulates changing sails, but it's also necessary to prevent "polar hopping" every minute along a given path where the TWS gives the best VMC.

I will read the older thread, I didn't know about it...
Thanks kroppyer. I truly think that apparent wind (AWS) is a relevant driver for any meaningful performance model, particularly for the 'recovery' phase.

It think it's clear that we disagree on the question of penalties for non-tacking course-change manouevres. In reality, most sail changes occur when we go around marks where we change from up-wind to down-wind or vice-versa. Other than that, sail changes usually occur due to changes in the sailing environment - wind strength or direction changes - rather than as a change directed by the navigator (or SOLer, in our case). Of course, these can't be easily incorporated in to SOL because (a) our polars aren't sail-combo-specific and (b) that would be way too complex for most folk anyway!

You mention "polar hopping" as an issue but I don't quite get it. As far as I can see the current system not only supports but encourages small automated changes as these incur virtually no penalty. Systems that do this have many names but I simply call them 'autopilots'. And while I like playing with such technologies, I don't really want to race against them.

As a sailor I want something close to real-life. Realism is good, but this is a game after all so I'm happy to compromise a bit.

As an engineer I just want to create a model that makes sense ... to me, to sailors, and to novices alike.

As a programmer I want something I can implement cleanly and easily that will not suck the life out of the server platform through unnecessary calculations.

The 'performance model' (loss and recovery) has clearly been an issue for many years now. I personally think the time has come to tackle it and resolve it once-and-for-all.

There are two main forum threads dealing with this issue. If there are related thoughts/ideas/etc that have been shared only in emails then perhaps they should be shared here as well.

BTW. I am likely to go ferral on this issue if we don't get some traction ...
You mention "polar hopping" as an issue but I don't quite get it. As far as I can see the current system not only supports but encourages small automated changes as these incur virtually no penalty. Systems that do this have many names but I simply call them 'autopilots'. And while I like playing with such technologies, I don't really want to race against them.
- End quote

I don't know about "automated changes", as I would understand that "automatic" would mean something that would somehow react to something as opposed to "precalculated" and "preprogrammed". I do use relatively large quantities of DCs (up to 200 per wx), which I calculate once the wx is available and I've done my routing exercises, but in my opinion there is no more automation involved in than that setting ten DCs. To an outside observer that might appear as "automated changes".

I don't know where you are going with this discussion, but if the idea would be to somehow penalise setting a lot of DCs, I fail to see the idea behind that. Would you limit the number of course changes allowed in IRL sailing?

--- Last Edited by karriv at 2014-11-13 14:47:01 ---

--- Last Edited by karriv at 2014-11-13 14:47:54 ---
My opinions may have changed, but not the fact that I'm right.
I think I'm starting to understand what dtayls talks about. Correct me if I'm wrong.
A course change of 5 degrees results in more performance loss than 5 course changes of 1 degree (over a longer period of time).

My reaction to that: that difference is negligible compared to the loss as a result of not sailing the fastest course. Those bigger jumps in HDG are only advantageous when polar hopping. Since most dents on the polar are caused by sailchanges, a performance penalty for such a change is not so weird.

I will get back to this and expand tonight (UTC+1)
Kroppyer, you are right, but that is not why I use a lot of DCs. The reason is to be as much as possible on the optimal course, and that by "rounding" the track you save distance compared to doing sharp angles. This has much more to do with implementation of optimal route as I described it in the router-thread than performance loss.

I actually tested doing a turn in multiple steps to reduce the perf loss, in a situation where you had to bear away after a mark about 30 degrees. Doesn't work, you loose much more in running a suboptimal route than what you gain in perf loss. If you are talking about course changes of some degrees, perf loss is totally insignificant even in the AC72.
My opinions may have changed, but not the fact that I'm right.
Why is performance loss necessary for non-tack-changing manoeuvres?
Sometimes you want to sail along a path of better wind pressure, when there's no shift to play with. Often, you can just sail along this path. Sometimes, the wind angle is such that you would be sailing in a dent of the polar if you followed the path. Most common is the upwind dent, or downwind dent. Solution to that is to do quick tacks of gybes along the path following VMG angles. That's not realistic, no one does 100+ tacks/gybes an hour. So performance loss is calculated for each tack/gybe, making it faster to leave more time between the tacks/gybes.

A bit less common is that it's not the upwind or downwind dent, but a dent between, say, code 0 and gennaker (I know we don't really have sailchanges in sol, but the polars show the dents anyway). In this case sailing in the dent of the polar is slower than sailing with the code 0 for about 30 seconds, then with the gennaker back across the centre of the path, and continue "polar hopping" along the path. This is just as unrealistic as 100+ tacks/gybes an hour, and this is the reason we need performance loss for non-tack-changing manoeuvres.

So, performance loss actually hurts any automatic course changes to follow a path (hard to set so many DCs and still stay in the centre of the path).

The problem with the current performance model is that it has a learning curve that is far too steep. I want a performance model that is understandable/predictable for people with only IRL sailing knowledge: a realistic model. Everyone must be able to make somewhat accurate guesses for how much distance they will loose after a given course change. That's (I think) what makes strategy games good, when predictions can be done, so that you can think ahead.

My personal conclusion (subject to change) is that the (what I call) "mechanical performance loss" (boat changing direction: kinetic energy or momentum, small duration of flapping sails, etc) is too small and quick that it's not useful to make it very realistic. Boats are updated once every 10-20 seconds. If you are able to give an accurate formula to calculate performance loss and recovery after a tack/gybe/whatever, that formula will need to be chopped down into 10-20 second pieces anyway, so all those realistic details will be removed.

Other than that, the mechanical performance loss is almost negligible compared to the performance loss due to moving weight around, and crew fatigue. Offwatch crew needs to wake up to move to the bunk on the other side of the boat. You can't to that too often or your crew starts hallucinating from lack of sleep.

I am working, with some others, on designing an alternative model. Once we have something we think is good, I hope to share it with you, including our considerations that lead to the new model. Others are free to do the same. Of course this can result in some double work, but I wasn't really expecting many people to jump up and design a new model.
karriv: I love the technology of routers. As an electrical engineer and a computer scientist I just can't help myself. My ideal today would be using Expedition plugged in to SOL via BrainAid's NMEA plugin, then taking the steering data from Expedition and automatically feeding that back in to SOL.

The problem I have with what you are doing with (say) 200 DCs every 6 hours (ie. one DC every 1.8 mins) is that it either involves an autopilot or a very patient helmsman. In my experience, giving a helmsman a tiny change every couple of minutes is a good way of getting turned into a man-overboard dummy. And from that moment he/she is is going to follow your last instruction to the letter, so don't expect to be recovered from the water any time soon!

To be honest, I think it's great that people like you, kroppyer and others are willing to put in so much effort to developing and learning how to use advanced tools. The ultimate gift is then sharing this knowledge with others.
kroppyer: I think it is fair to say that the whole SOL community is indebted to you for your indepth investigations and analyses, not to mentionion the fact that you share this knowledge so openly. Your analysis of the whole performance model question is invaluable.

I can see what you are saying re 'dents' in the polars. The upwind and downwind ones are obviously the most obvious but there are others and yes these often relate to sail changes. I think it is clear that tacks and gybes should incur a performance penalty, but I personally think SOLers should be allowed to play with the other bumps and lumps to their hearts' content. As far as I can see any gains in actively playing these would be negligible so we'd might as well take the opportunity to simplify the SOL model by removing the dTWA dependency entirely.

If anything, maybe turning across a 'dent' should drop your speed to (say) half the depth of the dent. Thereafter it would be up to my generic exponential recovery model for boat speed to return to that indicated by the polar. This would work well for tacks and gybes too delivering a a 50% drop during a tack! During a gybe things would be a little different. A square-rigger, for instance, would see almost no penalty. A VO70 on the other hand would see perhaps a 15% hit, but they can immediately turn to a wind angle that gives them high AWS thus helping them recover any lost speed very quickly before settling at their optimum TWA.

Interestingly, while my model is very simple, it potentially encourages IRL sailing skills such as those involved with efficient tacking. Let's say our optimal upwind TWA is 45 degrees. During a tack we lose a bit of speed so we generally lay off a little by going past the optimal angle to say 50 degrees or maybe a little more. As we pick up speed we harden up to the optimal 45 degrees again.
If we want to limit the number of tack/gybes per hour, why don't we do just that?

By storing one additional variable for each boat, called e.g. `crew_fatigue', this could be implemented easily. A tack/gybe increases the variable by some value, a course change increases it by a smaller value and a server jump decreases it by an appropriate amount.

The performance loss would then depend on this value with a larger loss when the crew is tired.

The downside to this would be the introduction of a new variable I suppose. To me it seems like an intuitive variable, but it probably has to be displayed in the client for transparency.

Page: First Previous 7 8 9 10 11 12 13 Next

Please login to post a reply.


Next Race: 00d 00h 00m

Current Races:

Boston to Newport TIMED Race 2021
Last raced as part of the 2018 40 footer series, we return to Boston for another race to Newport. However, instead of choosing a boat for a given start time, you get to choose a start time for the given boat choice of the speedy R/P 66! How quickly can you complete this  132nm run? 
RE-REGISTER HERE to race again after finishing a run. 
Race # 1430
INFO by brainaid.de 
NAM_AWIP WX Updates:
0245 / 0845 / 1445 / 2045
RACE CLOSE: Sunday, 31 January at 23:00 UTC
Race starts: Jan 18th 12:00 Registration will open soon
Old Flash Client GO TO RACE

Adriano's Brazil Run 2021
Adriano, who sails the SOLship batatabh, together with all his Brazilian friends bids you welcome to this amazing 300nm race in SOL's Albin 79 along their country's beautiful coastline. The race starts in Guanabara Bay, mistakenly considered a river mouth by its first explorers in January 1502, resulting in the settlement on its shore being named January's River (Rio de Janeiro). After passing Rio's iconic Sugar Loaf, the course takes racers along the north coast of the beautiful Ilha Grande to a turn for home at Ilhabela (another 'beautiful island') and a finish off the famous Copacabana beach!
Race #1416
INFO by brainaid.de
WX Updates:
0430 / 1030 / 1630 / 2230
January 22 at 2300 UTC.
Race starts: Jan 13th 16:00 Registration Open!
Old Flash Client GO TO RACE

Foiling Wellington to Ushant 2021
Time to don full weather gear and put the foot down as we race Sailonline's new foiling Imocas over the 11200nm from Wellington NZ to Ushant, France! We will cross two oceans, round Cape Horn, and try to stay on the foils through the notorious highs of the south and north Atlantic Oceans.
Race #1420
INFOby brainaid.de
WX Updates:
0430 / 1030 / 1630 / 2230
Ranking: SYC
Race starts: Jan 03rd 21:00 Registration Closed
Old Flash Client GO TO RACE

Virtual Cape2Rio Race 2021

Welcome to South Africa and RCYC's classic trans-Atlantic Cape2Rio Race from Cape Town, South Africa, to Rio de Janeiro, Brazil. First run in 1971 the next race in reality will be in January 2023, but in the meantime you can test yourself on-course in this, the Virtual Cape2Rio Race 2021, which introduces a new 74ft speedster, the stunning C2R74.
PRIZES: SMPFand See Intro Blog
Race #1411
INFO by brainaid.de
WX Updates:
0430 / 1030 / 1630 / 2230
RACE CLOSE: Wednesday, 20 January at 2300utc
Race starts: Jan 02nd 12:00 Registration Closed
Old Flash Client GO TO RACE

Go to race archive

SYC Ranking

  1. Sailonline Yacht Club Member WRmirekd
  2. Sailonline Yacht Club Member bonknhoot
  3. Sailonline Yacht Club Member NagaJolokia
  4. Sailonline Yacht Club Member rumskib
  5. Sailonline Yacht Club Member rafa
  6. Sailonline Yacht Club Member Zorba777
  7. Sailonline Yacht Club Member Kipper1258
  8. Sailonline Yacht Club Member calmxy
  9. Sailonline Yacht Club Member FreyjaUSA
  10. Sailonline Yacht Club Member Sax747

View full list


Mobile Client

SYC members have the benefit of access to our mobile/lightweight web client!

The mobile client