Zezo forum:

Post by pollsall | 2014-10-12 | 22:38:16

Anyone experiencing this phenomenon tonight ( Sunday night 2000 gmt). Has been working fine until now.

16 Comments

Post by pollsall | 2014-10-12 | 22:41:05

Experiencing this on VOR

Post by Fair_Winds | 2014-10-12 | 23:13:43

Same here, in VorG is the longest track duration possible 3 days since this afternoon. It�s kind of a trouble now when we soon are passing Gibraltar and want to try some different longterm tactics down towards FdN...
Could you please look inte this Cvetan? Thanks!

Post by zezo | 2014-10-12 | 23:33:53

I applied the restriction as a stopgap measure because the server became extremely overloaded tonight. Sunday evening is a peak moment, but I did not expect such big load.

I will be looking into adding a server or two more tomorrow.

Post by pollsall | 2014-10-13 | 00:11:27

Thanks for rapid response Cvetan. Will await your solution, thanks again for a great piece of software and its supporting hardware.

Post by Fair_Winds | 2014-10-12 | 23:59:04

Would be perfect C (Gibraltar is coming closer in strong winds (passing early morning, monday, and by then it would be nice to have some different choices to look at, and I am really interested to look at a different (and also "classic" route -> to go weast/northwest and try to aproach the doldrums from the west instead of east (as probably 99% af the fleet will...)
- I will send you some more money via PayPal, you are worth every penny! ;-)

Post by zezo | 2014-10-13 | 08:38:08

8 day forecast will be enabled today an hour or so after the wind change.

So far the 3-day forecast gives exactly the same routing, and the only option is to go to the Canaries and continue along the African coast. After passing the low -pressure system ahead of us, that is.

Post by SYSeaSick | 2014-10-13 | 08:43:22

Maybe you could turn 8d on again after the Monday morning 08.00 CEST rush hour, shouldn't be that much load on the server then...???

Post by SYSeaSick | 2014-10-13 | 08:44:43

Ooops, sorry, missed your latest post Cvetan...

Post by xcentricdave | 2014-10-14 | 10:07:39

Still restricted to 3 days? Or is it just me?

I've made a contribution to the Zezo processing power fund!

thanks Cvetan for an excellent system; as others have said, worth every penny...

Post by zezo | 2014-10-14 | 11:15:24

Thanks, David.

The site will be running off the cloud tonight ;-)

Your current problem appears to be that you are using the depreciated /vvor/ address. Now the main chart is at /vor/chart.pl (which was also used for the prologue, hence the temporary vvor name)

Post by paradolf | 2014-10-21 | 00:37:05

Turning point Fernando de Noronha is coming closer and closer. Time to start makeing calculations where to go after turning point.
I gues atm. the choice to show track to Cape town is not possible or I have some kind of problem on my computer. Whatever I do the route is only to Fernando de noronha.
If I set different destination it is still working but that is not a big help atm.
Cvetan is that planed because of heavy traffic on your server?

Post by zezo | 2014-10-21 | 14:27:48

I did this mostly to prevent people who have not discovered the right mouse button functions from missing the gate and then complaining in the forum ;-)

It works for the server load too. Shorter tracks are faster.

But even with a 7-day track, direct routing to Cape Town may not give you optimum strategy.

There is tactical decision to be made, how deep South to dive before turning left. Sometimes you go to Rio, sometimes you go straight from the mark. It all depends on the positions of the low pressure systems and the st. Helena High.

And then nobody prevents you from placing the target few miles short of the Cape. It won't affect the routing at this stage, but will avoid the rewrite.

Post by paradolf | 2014-10-21 | 23:09:00

Thanks Cvetan. For now it looks that the south is better, but we will see in a few days. Fair winds to all.

Post by headless_monkeyman | 2014-10-21 | 02:41:09

I see we are of course getting the full 8 days, but I thought I would check on your server stress level.

What is your preferred iron for serving this. Can you scale by load balancing, or better yet going with regional servers to take up the load? Since we can't see one another on your application I would think it would be pretty easy to do.

I've got three right now that have just been re-imaged and not put into production. Singapore, and SF. It is DigitalOcean so I think the cost is the same if I'm in NY or Amsterdam as well if they have the space available. I've just been doing science projects on Docker with them for a product I am designing.

Kurt

Post by zezo | 2014-10-21 | 14:39:02

I'm currently offloading only the route calculation. It's the slowest part by 2 orders of magnitude and very simple to implement.

Duplicating the entire front/back end structure would be a bit harder with some library and custom tool dependencies.

The router itself can run on any Linux box with modern 64-bit Intel CPU, no external lib dependencies and no root access required.

Right now the load is split between the main sever (4-core i5), 2-core E8500 (my previous server) and a similar machine in some data center in the US, provided by a friend.

And then I bring up google cloud instances during the peaks after the wind updates. It's more cost-effective than dedicated server because the peak load is less than a hour/day.

Post by headless_monkeyman | 2014-10-22 | 23:20:06

Come to think of it you could do calculations using the free AWS "micro" instances, spinning them up and maxing them out, then disposing of them as soon as the demand drops. Amazon penalizes you for an extended load level of any more than a couple of users, but you can do huge bursts of calculations with them with no objections (and they are free). Many a young blog programmer have beaten their heads in trying to figure out just why the "training" servers work so backwards.

I'd love to know a bit more about how you set up Google and how it is working for you. My current project is more worried about regional access speeds to central processing and data sources than scale. (But in the past I've had to do most systems with the biggest concern being the inability to calculate growth of the audience. Would a game launch to tens or hundreds of thousands? Hundreds next week or _millions_? The hazards of doing entertainment for Disney & the like. Most plans back in those days involved a lot of grey boxes with CISCO on the front and wads of money).

Let me know if you want another 2 cores with an SSD back covering any particular geo area. I've got SF, NY, Amsterdam, Singapore and I think Australia available at the moment. Have you considered putting the route calculator into a Docker container? In fact you can put anything you want into one, dependencies and all and it will run as expected anywhere you can get time for it, isolated and with protected resources. This is a poster child for Docker. If I was doing this permanently I'd take those containers, manage their development with git, and have Dokku (the words most useful 100 lines of Bash code) distribute them to where they run directly from the Master branch.

With apologies, I forget that a lot of people come here to play a game and get away from computer talk and the like, I'll send you my e-mail via contact in case we ever want to chat about this stuff later.

Cheers, and fair winds,
Kurt