Ok! All Network Savy Professionals to this Post now and explain this Login Problem!

*Comes back..sits down and re-thinks..oh I have a great idea for another Song called "I Wuv You Kuruna Beer.., I mean dear i mean..thinks again..*takes another swig.* Uh, nah lets scratch that for now..hmm..how about "If I had a Second Life". OMG! IT'S A HIT!

No, it's a database problem. The problem is caused by the load, not by the client version. The solution is to address the load, ideally by seperating out data to put as little load on the login process as possible.

The 1.6 release was fooked up. All of the little 1.6.x's that have come since are still fooked. There's no stable grid, and until LL finds a way to make it stable, we're fooked as well.

To use a quote completely out of context, "What we have here is a failure to communicate."

No, what we have here is 20% growth rates being unsustainable by a company of 35.

They actually re-wrote the login server, from scratch, for 1.6. Unfortunately, so many more people joined our beautiful world that it negated any gains from the vastly more efficient server code.

Ergo, we're back where we started, althought he login server is actually in better shape.

LF

To expand on "buy a bigger bottle with a bigger neck"

I'm gonna make an assumption here, and because LL is so closed mouth about their network hardware setup, it's a big assumption. But I'm assuming they've got one log on server.

Log files show invalid circuit 69.25.104.148:12036 (LL's login server) right before the game bumps you back to login screen. Yes, most likely a timeout message on the login.

Pings to this IP address return very quickly (averaging 40-45 ms here in CO). Trace Routes also proved very quick.

So far I've only confirmed what we know, the problem isnt LL's net connection.

So following through on this assumption, lets take that handy dandy 8 million (assuming Phillip hasnt already blown it on house in the Bahamas) and setup some clustered servers for the login area! I'd say 3-4 Spark servers from Sun running with 8 Xeon processors each, attached to multiple clustered database servers (God help us, MySql???? What was wrong with Oracle LL? or at least even SQL Server 2000?) No I'm not slamming MySQL, anyway I digress.

As login requests come in they're handed out sequentially to each login server in order.

But anyway, they say they have stuff planned for 'expanding the capable user base of SL' so maybe this already in the works, but I'm not holding my breath.

It's probably not even a network problem, it could VERY easily be a system problem. As in a system(s) are overloaded and can't handle everything they're being asked to do. And yes, this can cause errors that APPEAR to be network problems--I've seen it many times.

I just hope LL didn't make cross into the forbidden land and start using Linux servers, if that's what's going on, get ready for lots of long term problems guys. Unless if course they're gonna spend hundreds of thousands of dollars on new computer systems, that MIGHT help I guess...

Are you shitting me, they're using MySQL? Impressive it works as good as it does! How many users has MySQL been tested with?

They're expecting a big shipment of interocitors by monday night, they'll have everything ship shape by Tuesday.

MySQL is cool also.

Yeah, interocitors, that's the ticket!

What I said, but in a more technical fashion

But why SPARK? Solaris is having issues of its own.

If possible, rather then have 1 login server have 3 and each would have its own ip address

1. Primary first time server
2. secondary 1
3. secondary 2

Ok heres how it would run, when someone logs into SL for the first time (or first time after system is setup) they would log into the primary first time server which would then assign thier username to one of the secondary server and would include a key to point future logins towards the ip address of the secondary. Right there you have cut login bottlenecks in half, the only problem would probably be writing a key that would then make the end user login to the correct server.

Just my thoughts

Is CrystalShard Velma from Scooby Doo? That sounded just like what she would say!!!!!

John, that is what a server cluster does that is setup correctly for load balancing. You connect to one IP, that IP routes you to the least loaded server to process your login. Putting up more Login servers in sperate clusters would not really speed up the system.. it might actually slow it down by the clusters now having to talk to each other cluster, plus to main account cluster to continiously be synced with each other.

Just need to improve the resources on the existing cluster and maybe tweak load balancing issues.

Perhaps thier not doing proper load balancing to begin with?

By the way, found an auto mouse mover/clicker so i just setup the logon screen, run the program and read a book till i get logged on.

Its called Auto Mouse and found it on www.download.com

Well after reading Philips post, I believe it is more that their current resources are unable to support the load they currently have. It also seems they are having db problems that seem to stem from bad scripting, overuse of available resources, and probably their sftw being ineffecient.

Hopefully 1.6.5 helps with alot of these issues.

I say, lets get everyone to donate some lindens, cash them in for cash, then hire someone from blizzard to go to work for them.

Or at least buy them a coupla new top of line servers.

lol if the world was only so nice.

LMAOROFLSPRDL

That's SPARC. Scalable Processor ARChitecture.

SPARC servers use SPARC processors. SPARC is a 64-bit RISC architecture that Sun designed on its own and has been using for years. It is not related to any architecture created by Intel.

The login server(s) they are using now are fast enough. The problem is that the DB server is getting slammed with email queries, which are not done in a very efficient way. MySQL tends to be faster at reading tables than any of those other databases, so going to Oracle or MSSQL could actually make things slower rather than faster.

It amuses me that people slam LL for using MySQL when they have only dim clues as to what is going on behind the scenes, and probably haven't got a clear picture of what MySQL is capable of today.

well seems to be working since the last patch, but i havent seen anymore then 1500 people on when im on so i dont know how it might be during peak. anyone know how its working during peak?

You may be entirely correct, but from what I've always seen: "You get what you pay for!" And unless there's a commercial version of MySQL out, I'd be skeptical that it's actually better than a commercial DB. (Although not skeptical that it uses a fraction of the disk space...)

I'm paying negative money to use SL, what does that tell you?

BTW, it is very hard for one thing to be "better than" another thing, when both things are as elaborate as a relational database management system. There are many things to be taken into consideration for each application and it is ridiculous to paint one of them as being "better than" another. If MySQL was so remarkably incompetent at delivering what people needed from a RDBMS, Postgres (which is also free) would have killed it off long ago.