Rybka online........ A HORRIBLE idea!!!!!

[Gian-Carlo Pascutto]

The moment rybka is no longer number 1 this model will fail.

True, but that isn't limited to renting. If Rybka is no longer the clear nr.1, all sales models are going to have problems, so it makes no sense to say that it's a problem of renting itself.

[Gian-Carlo Pascutto]

What a strange question. It is blindingly obvious that we simply have a different concept of what constitutes a worthwhile competition.

Wait. The question is to not how to make correspondence chess fair.

The question is how engine authors can maximize their income.

[K I Hyams]

What a strange question. It is blindingly obvious that we simply have a different concept of what constitutes a worthwhile competition.

Wait. The question is to not how to make correspondence chess fair.

The question is how engine authors can maximize their income.

No. The question to which I was responding was very specifically about the effect of online access on correspondence chess. You can see the sequence below.

I just now found out, to my astonishment, that this is about to be implemented, and at great costs for usage (maybe even 750 euro per year).
It will GREATLY ruin my interest in Rybka 4, and greatly ruin the whole hobby.
I can think of many terrible things this will cause.
I really hope it will be cancelled.

Please tell us more about the terrible things it can cause.

Without knowing the details, it is possible to imagine a situation in which it may have an undesirable effect on top-level correspondence chess.

How? Right now if you're very rich you can buy hardware which is >>>> regular player's hardware which gives you an advantage. Is this any different between being rich and having access to an engine which gives you an advantage? I see no difference whatsoever.

Most people can now afford a quad. Octals are very expensive. It is possible to compensate for the difference in output between a quad and an octal running similar software. However, if he gives people the option of a super-strong program, not available on the open market and running on a very large cluster, then those who can't afford to access it may be at a disadvantage of 200 or more points. Top correspondence players are often playing 20 or more games at a time. It could make the hobby very expensive indeed if one is to remain competitive.

This makes 0 sense. You can afford a quad. I can afford an octal. durrr can afford any hardware money can buy. Right now, money can buy advantage for me and much more money can buy much more advantage. What exactly are you complaining about?

If he does so, you don't have to buy your own cluster any more but can hire his, with all the advantages that you don't have to maintain it. It's very likely that because of this you will get better worth for your money. And, maybe he hopes some professional players/federations will use it, because now none of them uses hardware/engine parameter tuning because, well, they need someone who would maintain it.

What a strange question. It is blindingly obvious that we simply have a different concept of what constitutes a worthwhile competition.

[K I Hyams]

I just now found out, to my astonishment, that this is about to be implemented, and at great costs for usage (maybe even 750 euro per year).
It will GREATLY ruin my interest in Rybka 4, and greatly ruin the whole hobby.
I can think of many terrible things this will cause.
I really hope it will be cancelled.

Please tell us more about the terrible things it can cause.

Without knowing the details, it is possible to imagine a situation in which it may have an undesirable effect on top-level correspondence chess.

How? Right now if you're very rich you can buy hardware which is >>>> regular player's hardware which gives you an advantage. Is this any different between being rich and having access to an engine which gives you an advantage? I see no difference whatsoever.

Well, I expected that response. I wouldn't expect you to be able to see a difference without knowing the business models that he has in mind.

Most people can now afford a quad. Octals are very expensive. It is possible to compensate for the difference in output between a quad and an octal running similar software. However, if he gives people the option of a super-strong program, not available on the open market and running on a very large cluster, then those who can't afford to access it may be at a disadvantage of 200 or more points. Top correspondence players are often playing 20 or more games at a time. It could make the hobby very expensive indeed if one is to remain competitive.

I think you are not thinking this through. What kind of cluster do you think they will offer? Say a 128 node box as we have? Half-million dollars worth of hardware? But what happens when you have 128 people using it at the _same_ time? You get no more power than what you would get from a single 8-core box, which is quite cheap today.

The idea of a big system sounds attractive, until you look at it realistically and try to estimate what fraction of the thing you will get. How many simultaneous users are realistic? I know correspondence players that let a computer run for 24 hours at a time on one game. Will 128 people access this thing at once? You get an octal-core equivalent, nothing special. 256? You get a quad-core. Really normal hardware today. 512 at once? Now you get a dual core and anybody can buy hardware faster than that.

Sounds good until you think about what it _really_ means.

Yes. In fact, I rejected the idea of using one extremely high powered machine as being prohibitively expensive for correspondence use for all but the very rich. My thinking had gone down the pathway that Rolf had taken, different powers being offered at different tariffs, similar to the model that my ISP uses. As you suspected, I hadn’t taken into account the enormous power drain that each user took from the system, probably because of the ISP parallel. I had considered the effect on the user, not the system.

[playjunior]

What a strange question. It is blindingly obvious that we simply have a different concept of what constitutes a worthwhile competition.

Wait. The question is to not how to make correspondence chess fair.

The question is how engine authors can maximize their income.

No. The question to which I was responding was very specifically about the effect of online access on correspondence chess. You can see the sequence below.

I just now found out, to my astonishment, that this is about to be implemented, and at great costs for usage (maybe even 750 euro per year).
It will GREATLY ruin my interest in Rybka 4, and greatly ruin the whole hobby.
I can think of many terrible things this will cause.
I really hope it will be cancelled.

Please tell us more about the terrible things it can cause.

Without knowing the details, it is possible to imagine a situation in which it may have an undesirable effect on top-level correspondence chess.

How? Right now if you're very rich you can buy hardware which is >>>> regular player's hardware which gives you an advantage. Is this any different between being rich and having access to an engine which gives you an advantage? I see no difference whatsoever.

Most people can now afford a quad. Octals are very expensive. It is possible to compensate for the difference in output between a quad and an octal running similar software. However, if he gives people the option of a super-strong program, not available on the open market and running on a very large cluster, then those who can't afford to access it may be at a disadvantage of 200 or more points. Top correspondence players are often playing 20 or more games at a time. It could make the hobby very expensive indeed if one is to remain competitive.

This makes 0 sense. You can afford a quad. I can afford an octal. durrr can afford any hardware money can buy. Right now, money can buy advantage for me and much more money can buy much more advantage. What exactly are you complaining about?

If he does so, you don't have to buy your own cluster any more but can hire his, with all the advantages that you don't have to maintain it. It's very likely that because of this you will get better worth for your money. And, maybe he hopes some professional players/federations will use it, because now none of them uses hardware/engine parameter tuning because, well, they need someone who would maintain it.

What a strange question. It is blindingly obvious that we simply have a different concept of what constitutes a worthwhile competition.

You didn't demonstrate what CHANGES when this model gets introduced. Now, money can buy advantage. With the new model, money will buy an advantage. (I'm not arguing whether the fact that money can buy things is fair or not. )

[Nimzovik]

I will state categorically that I will never do the online thing with Rybka. Just my humble opinion. So? How to compete with those that do? We all note engines have certain blind spots. Ergo I should think one should study opening systems that exploit these (as much as possible ....How? Rare openings for starters, closed systems for another that accentuate the knight value perhaps etc. ) I should think that the 'poor man's' approach would be to use one's Rybka 4 (or what ever) in the more esoteric and hopefully sound systems to gain the advantage. The online people will, I suggest, will end up rather tunneled visioned by hammering out all the intricacies of systems that can be avoided. I must say a certain amount of laziness has been incurred by the writer by using computers to just such an extreme extent. Chess is not 'solved'. There is still room for novelties and styles of play. Just ask Pablo There are only so many good moves in a position no matter what the strength of the engine may be.

[bob]

No way to have the numbers. How many correspondence players would use this? You do realize that a 128 node cluster is a half-million dollars plus? They are not going to make enough money to buy that kind of hardware, that's dreaming. Time to come back down to planet earth.

128 quad-nodes * 500 EURO/node (includes interconnect) = 64 000 EUR, or about 1/10th of what you claim.

You overlook just a bit. 128 quad nodes will need 30+ Tons of Air conditioning. Or do you want the machine room to run at 135 degrees F or higher?

Our 128 dual-node cluster sits in a room with 70 tons of A/C. When I crank up a test on this machine, the air directly behind it (all the nodes exhaust rearward, taking air in from the front) exceeds 135 degrees. So price in the A/C needed and the price climbs.

What about connections? We use infiniband. 128 nodes requires a bunch of switches. Plus the bandwidth coming in to make this usable. More expense.

And then there is the non-negligible power requirements. That continue month after month.

A half million dollars is a pretty good estimate, when you consider _everything_.

Note that if you actually *have* 128 people using it simultaneously, the economics work in favor of just buying more hardware. Really, hardware can't be the problem there. If you run out of it, it's because you've got "too much" customers. How exactly is that a problem? I'd say it's the goal!

Are you going to charge _each_ customer enough to buy them a dedicated node? This is not something that can just increase in size on a whim. We've been through two A/C upgrades in the past 4 years to keep up with our clusters as we have added hardware. I'd bet you need at least 10 users per node just to break even on the cost, and if they are heavy users, they are going to be unhappy when they get 10% of one node.

If you believe that usage will be very light, it might work, although you are still stuck with that initial half-million dollar investment to buy the hardware and environmental controls.

[Dr.Wael Deeb]

No way to have the numbers. How many correspondence players would use this? You do realize that a 128 node cluster is a half-million dollars plus? They are not going to make enough money to buy that kind of hardware, that's dreaming. Time to come back down to planet earth.

128 quad-nodes * 500 EURO/node (includes interconnect) = 64 000 EUR, or about 1/10th of what you claim.

You overlook just a bit. 128 quad nodes will need 30+ Tons of Air conditioning. Or do you want the machine room to run at 135 degrees F or higher?

Our 128 dual-node cluster sits in a room with 70 tons of A/C. When I crank up a test on this machine, the air directly behind it (all the nodes exhaust rearward, taking air in from the front) exceeds 135 degrees. So price in the A/C needed and the price climbs.

What about connections? We use infiniband. 128 nodes requires a bunch of switches. Plus the bandwidth coming in to make this usable. More expense.

And then there is the non-negligible power requirements. That continue month after month.

A half million dollars is a pretty good estimate, when you consider _everything_.

Note that if you actually *have* 128 people using it simultaneously, the economics work in favor of just buying more hardware. Really, hardware can't be the problem there. If you run out of it, it's because you've got "too much" customers. How exactly is that a problem? I'd say it's the goal!

Are you going to charge _each_ customer enough to buy them a dedicated node? This is not something that can just increase in size on a whim. We've been through two A/C upgrades in the past 4 years to keep up with our clusters as we have added hardware. I'd bet you need at least 10 users per node just to break even on the cost, and if they are heavy users, they are going to be unhappy when they get 10% of one node.

If you believe that usage will be very light, it might work, although you are still stuck with that initial half-million dollar investment to buy the hardware and environmental controls.

Thanks Bob,an extremely interesting reading....actualy it adds to my opinion that this online rental system will go down in flames....
Dr.D

[Gian-Carlo Pascutto]

What about connections? We use infiniband. 128 nodes requires a bunch of switches. Plus the bandwidth coming in to make this usable. More expense.

Already factored in, as pointed out in the original post. No need for silly expensive things like infiniband.

If you believe that usage will be very light, it might work, although you are still stuck with that initial half-million dollar investment to buy the hardware and environmental controls.

Nobody says you have to start with 128 quad-cores. That's just a number you came up with. But anyway, let me take you up on it:

There are hosters offering Core i7 920 hosting in Germany for 50 Euro per month, or 600 Euro per year.

Running cost of 128 of those boxes: 77k Euro per year.
Investment cost: 20k Euro setup fees.

[towforce]

There are ideas that are good, and there are ideas that make no sense. This is one of the latter. You put together a "super-cluster". Say even 64 nodes of octal-based machines. But the minute you have 64 people using this thing at the same time, each has no more power than a single octal. If you have 128 people using it at once, even less. How many let something run for a day or two? So this "cluster approach could actually give you highly _inferior_ hardware when you factor in other users.

It makes sense to stop reverse engineering. But it is not economically viable, I'll bet. free-market economics will dictate whether users are willing to spend the money to get something that is considerably slower than what it would be on their own dedicated hardware.

Does anyone know for sure that the search power of a large number of processors is what will give this system its strength? There are two other obvious possibilities:

* a mathematical library is being used - and he doesn't want rivals to find out which libraries he is using (still less which functions are being called in those libraries). If it was a commercial library (quite likely - at the present time, many commercial libraries are better written and produce results an order of magnitude faster (this is especially true in the field of linear programming, where it takes an enormous amount of time to work out the fastest way to collapse large sparsely-populated matrices of various patterns)) then he also would need to provide a license to every customer he sold it to

* one or more huge databases of knowledge being stored in-memory for fast access. This could be shared between any number of processors via a high-speed client-server service