25 replies to this topic

[rubberducky]

Thankfully we got tickets this year, but only because a mate helped us out.

I live in Sweden, and had two laptops going on the same local network. I had a VPN to a work computer (so different IP address) and a second VPN to a UK computer.
I got the queuing page at 8.50 - but didn't see any other connection until midday.

This is really rubbish, and much worse than previous years. We have two years to plan for next time...

I work in networking, but this is not my area.

See Tickets must be using a load balancer :
http://en.wikipedia....ing_(computing)

Then I found this :
http://www.zeus.com/...ntre-stage.html

I guess it is this traffic manager :
http://www.zeus.com/...ager/index.html


When you first get a connection, the load balancer just drops the connection. This gives you the "interrupted" message.

The load balancer must maintain a session ID, so once you are in they keep the link open - otherwise you couldn't complete the booking process. When there is a space,you get through to the booking servers, and it keeps that connection open - thats how we got tickets as once your are in you stay  in. We are lucky the booking servers don't close the session to the load balancer, if they did you would have to start all over again. They could implement this in future.

New is a "in the queue". I think here, the load balancer is talking to you (to give the queue page) and will connect you when it can.

What I don't get is that when some people changed connection (say 3g) they just got through - meaning there is no real queue.

What we need to know is how the load balancer maintains session ID - is it putting a cookie on your machine, or is it done by IP address state.

Two laptops on the same local network are definitely treated different, when one eventually got through all browser windows went through at the same time (either in firefox tabs, or different browser windows). The other laptop never got through.

If this is true, there is no point having more than one window on a machine, but more machines helps.

Anybody know anything more?

/Mike

[windy_miller]

rubberducky, on 04 October 2010 - 08:47 AM, said:

Thankfully we got tickets this year, but only because a mate helped us out.

I live in Sweden, and had two laptops going on the same local network. I had a VPN to a work computer (so different IP address) and a second VPN to a UK computer.
I got the queuing page at 8.50 - but didn't see any other connection until midday.

This is really rubbish, and much worse than previous years. We have two years to plan for next time...

I work in networking, but this is not my area.

See Tickets must be using a load balancer :
http://en.wikipedia....ing_(computing)

Then I found this :
http://www.zeus.com/...ntre-stage.html

I guess it is this traffic manager :
http://www.zeus.com/...ager/index.html


When you first get a connection, the load balancer just drops the connection. This gives you the "interrupted" message.

The load balancer must maintain a session ID, so once you are in they keep the link open - otherwise you couldn't complete the booking process. When there is a space,you get through to the booking servers, and it keeps that connection open - thats how we got tickets as once your are in you stay  in. We are lucky the booking servers don't close the session to the load balancer, if they did you would have to start all over again. They could implement this in future.

New is a "in the queue". I think here, the load balancer is talking to you (to give the queue page) and will connect you when it can.

What I don't get is that when some people changed connection (say 3g) they just got through - meaning there is no real queue.

What we need to know is how the load balancer maintains session ID - is it putting a cookie on your machine, or is it done by IP address state.

Two laptops on the same local network are definitely treated different, when one eventually got through all browser windows went through at the same time (either in firefox tabs, or different browser windows). The other laptop never got through.

If this is true, there is no point having more than one window on a machine, but more machines helps.

Anybody know anything more?

/Mike

I noticed that too.  Though only in the same browser.  So if I got one chome tab to go through, all my other chrome tabs would go through too.  My firefox tabs were still locked out though.  I thought it was probably chrome just caching the page.

I don't believe that "once you're in you're in" though.  I could get through to the !You are in a queue" page, but within a minute I would be back to the "the site is not responding" window and would have to start all over again.  Also, I managed to get a booking form up, but once I had entered my details, I could not get the next page up.

[rubberducky]

windy_miller, on 04 October 2010 - 09:22 AM, said:

I noticed that too.  Though only in the same browser.  So if I got one chome tab to go through, all my other chrome tabs would go through too.  My firefox tabs were still locked out though.  I thought it was probably chrome just caching the page.

I don't believe that "once you're in you're in" though.  I could get through to the !You are in a queue" page, but within a minute I would be back to the "the site is not responding" window and would have to start all over again.  Also, I managed to get a booking form up, but once I had entered my details, I could not get the next page up.

I didn't think to try different browser types. May be a good idea if they came across as different sessions, although I don't know why two different firefox browsers would be treated differently to a firefox & chrome window.

We need to figure out if cookies were used. Did anyone have any luck if cookies were disabled?

[parsonjack]

of course see have load balancing...they have to so that the load can be simultaneously spread across a number of servers.  you can do it in a number of ways, by least connections, round robin etc but i'd guess they use round robin (the default on our cisco LB's) to share the load.  you also need the session to stay alive until completion...so there will be session stickiness set, again in a number of ways (by source ip, cookie etc) but i'd guess that source IP is what they use here.  there is also likely to be a session idle timeout set, but as its not easy to predict how long it will take to complete the transaction under these circumstances i'd guess its pretty long...which is why you can simply go back to your booking page and do it over and over....

[ethereal]

What was the deal with using Opera in turbo mode? Not a sniff here with IE9 or Firefox but straight through to queue and quickly through to check-out with Opera.

(Didn't do me any good mind, no amount of redoing the order pages would give me a confirmation)

[rubberducky]

ethereal, on 04 October 2010 - 09:38 AM, said:

What was the deal with using Opera in turbo mode? Not a sniff here with IE9 or Firefox but straight through to queue and quickly through to check-out with Opera.

(Didn't do me any good mind, no amount of redoing the order pages would give me a confirmation)

I talked to some of the people in the office here who know about this stuff.
Most home modems do not do IP address translation, so each computer on your network will have a different IP address as seen from See Tickets - this explains why one would get through and one not.

They also think it is unlikely that the load balancer would have enough information to distinguish between different sessions on one IP address (browsers open multiple connections) so it wouldn't matter if you have 1 window or 100 open on a single computer.

edit : As somebody else pointed out, if you have several computers behind a translating router (maybe at work) it is the router IP which gets through. Then, all computers behind that router should get through at the same time. This doesn't improve your chances of getting though in the first place though.

????
/Mike

Edited by rubberducky, 04 October 2010 - 10:26 AM.

[t0paz]

When you are at home you all sit behind one IP, the home router doesn't do IP address translation it does something called port address translation. IP addresses are getting in short supply and so more and more expensive. Your ISP does not give you one for each device for exactly that reason.

If I go to www.seetickets.com it connects on port 80 (http) at their end but can come from any port at my end (within a set range) so let's say 32004 for example. If another PC on my network goes to www.seetickets.com my router will send that out via a different port, let's say 32005 for example. Seetickets see's that as 2 different people even though we have the same IP. Our router sorts out who get's what traffic. That's why people could get through on one PC but not the other.

[rubberducky]

t0paz, on 04 October 2010 - 10:31 AM, said:

When you are at home you all sit behind one IP, the home router doesn't do IP address translation it does something called port address translation. IP addresses are getting in short supply and so more and more expensive. Your ISP does not give you one for each device for exactly that reason.

If I go to www.seetickets.com it connects on port 80 (http) at their end but can come from any port at my end (within a set range) so let's say 32004 for example. If another PC on my network goes to www.seetickets.com my router will send that out via a different port, let's say 32005 for example. Seetickets see's that as 2 different people even though we have the same IP. Our router sorts out who get's what traffic. That's why people could get through on one PC but not the other.

I was going to try later with two computers to see if they had different IP addresses.
Port address translation makes a lot more sense. But, is the source port maintained through all the routers on the network to Seetickets? I guess it must be otherwise the home router would not be able to sort route the packet back to the original machine.

In this case, the load balancer could create a session based on IP and port number?

Do multiple connections from one machine (different browsers for example) all have the same source port?
/Mike

[t0paz]

My computer wants to speak to see's computer on port 80. How it gets there doesn't matter at all and is up to each hop along the way how they do it. If I send a letter to Australia I don't need to know how it gets there. The same is true here.

Each tab in your browser is like a new copy of that browser, and so will have a different port. Think of ports like telephone lines - 2 people can't use the same one at once (ok, everyone connects to the same one at See's end (http/80) but they use some giggery pokery for that)

As all browsers got in at once then they couldn't have done an IP/port session. Web sessions like that aren't really my area though so I don't know how they would have done it.

I can tell you that queue was definitely not a queue though! That was only there to discourage people from hammering F5 even more and making matters a lot worse than they already were.

[Paul ™]

I thought I was technical but Mike you've started a right Geek thread mate    (I hope you got tickets sorted!)

[rubberducky]

Paul ™, on 04 October 2010 - 11:08 AM, said:

I thought I was technical but Mike you've started a right Geek thread mate    (I hope you got tickets sorted!)

yeah, sorry about that - but it's an arms race this, too close this time
Tickets shorted with some help, look forward to spending a few days drinking too much and talking rubbish again.

I did some tests. The switch is using port address translation so both computers have the same IP address.
Every time I do a refresh, the source port id changes. One computer is always in the range 2*** and the other in the range 3***.
Using tabs or different browsers made no difference.

So, I think they load balance on a combination of source port range and IP address.

Using multiple browsers/windows/tabs shouldn't make any difference over hitting refresh on the one window lots and lots.
Using more than one computer would help...

Mike

[wweerr208]

highly unlikely they will just use load balances, more likely use something that is designed for serious security and so happens to load balence

I've work with these beasts, no need for session cookies with some of these

http://www.f5.com/pr...are/big-ip.html

and for more reading

http://en.wikipedia....n-level_gateway

Edited by wweerr208, 04 October 2010 - 09:35 PM.

[AnnaGrant]

Can I lower the level of this impressively technical discussion?

I didn't get further than the queue (that wasn't a queue-how gullible I was to believe that) so if/when resales open I want to improve my chances of success.

Could anyone answer these questions-preferably quite simply if possible-

1) Is f5 the refresh button?  I just wanted to check I'd got that right!
2) Can you only refresh/f5 once the 'in a queue' screen comes up?
3) How can you monitor multiple open tabs to check if one of them has got through to the booking screen?
4) We used 2 computers on a single wireless home network.  Would that have given us 2 chances to get through, or would seetickets have seen this as one chance?
5) Someone said hitting f5 too quickly wouldn't help.  Is there a best frequency to f5 or refresh?

Finally, can anyone offer their best strategies for getting through?

Thankyou from non-techie AnnaGrant

[Pinhead]

From experience this year, I believe your best best is to spread your open browsers over multiple networks. For example, this year - nothing on home xDSL - straight in on 3G phone every time.

Load balancers - yes, they likely use a dedicated balancer such as the F5 Big-IP, Foundry's solutions etc. rather than Microsoft windows software solutions (WNLB), as I expect they use all Linux / UNIX boxes anyway behind it, and WNLB would really struggle with a task like this. As has been said, many are also security devices now. However, the primary function of a balancer remains to balance out incoming connections to the target server with least work to do in any given cluster. Whether See chose to limit those incoming connections to a single IP or not, I don't know, but it would make sense from the perspective of what they know will be the likely shape of the client traffic (multiple browsers open proxying off one public IP etc.) It is also likely that the balancer has been configured to terminate sessions when no servers within the cluster respond to it that they are able to serve any further connections.

Once through the balancer, it will be the web server that holds your session open whilst you are in the queue and whilst you buy all your 100's of tickets for all your facebook mates. From the balancers perspective, the connection is still open so will be held - it doesn't need to do anything more until you close your browser. You have got its 'attention' so to speak.

IP wise, although I can't speak for everyone's peraticular implementation, most ISP's give you a router set up to give all the machines behind it an non-routable IP belonging to one of the reserved ranges - typically 192.168.0.1/24. This is then NAT'ed by your router to the 'public' IP given to it (via DHCP or similar) from the ISP's ranges at their end. TBH although these public IP's were scarce a few years ago, they are now typically then bound to IPv6 addresses on the ISP's core network (of which there are billions available). The final IP address then presented to the load balancer at See could then likely be completely different from your routers 'public' IP, as it may have been further translated to another IP from the ISP's pool of available true public addresses.

Thus, the upshot of all this is that to get the load balancers attention a multiple number of times, you need multiple public IP addresses, preferably from different ISP's (i.e different networks). Thus, next time have a browser open at work, one at home, and one on your phone. There were certainly more 3G network providers (mobile ISP's essentially) connections available at See in my experience - perhaps a different load balancer was covering these on a different cluster that was less stressed . . .

[t0paz]

I had 4 tabs open in the same browser (IE8). I then refreshed these every few seconds. F5 does refresh, but holding CTRL while you press F5 makes it reload the whole page. That's that I did. Not sure if it helped or not.

Other than that I think it was just luck.

Also make sure you go straight to http://www2.seeticke...&method=DEPOSIT rather than www.seetickets.com/2011

Good luck in the resale Keep an checking here as unannounced resales do happen randomly in the days/weeks coming up. Probably as duplicate orders are weeded out or card payments don't go through.

[al_coholic]

What I think a lot of people are missing here is a very simple point.

Seetickets servers/load balancers and all their equipment are perfectly capable of dealing with demand and load for 364 days of the year. They only have a problem on 1 day a year when Glastonbury tickets go on sale.

They arent going to spend hundreds of thousands of pounds to improve these systems, when they work perfectly well 99.9% of the time.

They would probably argue that they work well enough 100% of the time as they managed to sell 130000 tickets in 4 hours.

I know there is a bunch of people on here who feel hard done by but really its just supply and demand

[Pinhead]

Regrettably true. What we do disagree upon however is the fairness of the system. We believe it could have been made fairer, indeed it used to be fairer after 2007 when they removed the session recycle 'loop hole' within the system after many complaints. It seems that See care less about fairness today and only about shifting tickets, which I can understand from a commercial perspective. I wonder however whether Michael would think it fair that the majority of his punters in 2011 are only there because they all 'know' each other on facebook. A facebook conditioned ticket sale is not the sort of narrow audience I would have expected him to want to foster.

Edited by Pinhead, 05 October 2010 - 10:12 AM.

[t0paz]

Very true.

If it were left to See I don't think they'd care if they had 1 server and the ticket scramble was drawn out over months. I think we have to thank Mr Eavis that it's in the state that it is. There is zero profit in it selling out quickly as far as See goes, and they are a cut throat business after all. A kind of legal tout if you will.

It's the best we can expect.

[Paul ™]

Pinhead, on 05 October 2010 - 10:10 AM, said:

I wonder however whether Michael would think it fair that the majority of his punters in 2011 are only there because they all 'know' each other on facebook. A facebook conditioned ticket sale is not the sort of narrow audience I would have expected him to want to foster.

Where are you getting the basis that "the majority of his punters in 2011 are only there because they all 'know' each other on facebook"?

How many are included in your basis?

[AnnaGrant]

t0paz, on 05 October 2010 - 09:38 AM, said:

I had 4 tabs open in the same browser (IE8). I then refreshed these every few seconds. F5 does refresh, but holding CTRL while you press F5 makes it reload the whole page. That's that I did. Not sure if it helped or not.

Other than that I think it was just luck.

Also make sure you go straight to http://www2.seeticke...&method=DEPOSIT rather than www.seetickets.com/2011

Good luck in the resale Keep an checking here as unannounced resales do happen randomly in the days/weeks coming up. Probably as duplicate orders are weeded out or card payments don't go through.

Sorry to be dense- but what difference does it make refreshing f5, or refreshing whole page on CTRL f5?