Hello, I have a few questions if it isn't too much, I have already searched the forums, and Ask Mr. DNS.

I have an idea how to setup load balancing on the HTTP requests. But, I have a few questions. Example:

Setup NS1 to go to x.x.x.100 and NS2 to go to x.x.x.101 and NS3 to go to x.x.x.102
Then on the main server (server1) I would place multiple A records. Example:

domain.com A 1000 x.x.x.100
domain.com A 1000 x.x.x.102
domain.com A 1000 x.x.x.103

So then the load would be distributed among three servers. And when the main server (server1) goes down, the nameservers (NS2)
would kick in and go to the second server (server2) and then I would have multiple A records setup on the main server:

domain.com A 1000 x.x.x.102
domain.com A 1000 x.x.x.103

So that the second server does not become over loaded, and distributes the load between the second (server2) and third (server3)
server.

Now, I have realized two flaws.

1) What about dynamic content, such as forums, news scripts, bloggers etc. If the person was sent to the second (server2) server and
post a thread, and if another person was redirected to the third (server3) or main (server1) server, then the content would not be
seen and only be seen if a user was sent to the second (server2) server. Rsync wouldn't work because the second (server2) may have
content that the main (server1) or second (server2) doesn't/visa-versa. So it would over right, which isn't good.
2) What about email, how could I get that setup to work properly. If the main (server1) went down or is load balancing the MX
record's, then mail could would redirected to the second (server2) or third (server3) server, but what if the mail was stored on the
second(server2) or third (server3) server, then they would never get it, and rsync won't work, due to it may over right email on
either server that the backup or main server doesn't have.

How can I make these two flaws work?

Thank you all very much, and have a wonderful day/evening!!!

Are you talking about multiple servers at the same location or multiple servers at multiple locations? Your options differ depending on the results.

Assuming that you are talking about multiple servers at a single location, basically you need to have multiple servers with write access to your data and databases. This can either be accomplished via shared connection (shared SCSI, Fibre Channel, etc.), client/server model (NFS, etc.), or over a parrallel type of file system. You may want to look at http://www.lcic.org and check out their section on file systems.

Actually, I am refering to server in different NOC's so, I can have 100% server and network uptime or like 99.99999%.

Thanks!

Even the biggest ISP's always have a single point of failure.

What you are doing requires a more distributed design, and probably isn't cost feasible between different noc's unless you happen to own them or something.

Ultimately, you can build clustered environments in a colo, but if some dummy pulls the power on the load balancer or the switch , anything dumb like that can take you out.

You should have 99.9999% uptime if your server is stable, and the facility is well manned.

Then again some of the biggest colo's in the nation were out for several hours last saturday morning, and all the redundancy in the world didn't save their servers (some of which were probably million a year plus hosting clients).

rigor,

Well, if I were to do it inside of the same datacenter. NAC I would be using, what is the best and cheapest way, we would have dedicated servers, and I am not to knowledgable on this subject, I am not even too knowledgable on the Linux subject but, I can find out things :) I hope!

Thanks!

read the lcic link above. High availablity, clustering and load balancing. It's not for the timid thats for sure ;)

There's really only one way to do it right, load balancer(s), private network, modular servers, network file storage..

I'm not sure if anyone does this plug-n-play for you but the cost is a bit more $$$ than a couple single servers.

Would be kinda cool if someone came out with a blade setup like this for cheap. Maybe some places do this already? if not might be a good business opportunity to design one.

lol, I bet it would be, as long as I can figure out what you are talking about :D

The thing I can't understand is, even with Hardware load balancing, how does the content on one server get to the next one? Is it possible with the current solution I mentioned above, how can I get dynamic content to work?

Thanks!

Also, the LVS seems like a good solution, I think:confused:

Can you only use it for internal datacenter use? How does it work exactly?

Thanks again!

Why store the content on either webserver, thats were NAS (network attached storage) comes in handy!

:confused: :confused: :confused: :confused:

uhh. Could you tell me how that works??

Thanks!

I believe that lcic discusses some of the NAS (network attached storage) options. It should also discuss some of the SAN (Storage Area Network) options. One of the things that it doesn't mention is Oracle's new CFS (cluster file system) that allows multiple systems access to the same physical media.

As rigor indicated, none of this is for the timid. I would recommend working with an established VAR (Value Added Reseller) who has experience in this. You will pay extra, but you will get a working environment. The tradeoffs in designing NFS systems and/or SAN systems are non-trivial. (For example, if you don't know the answer to the question "Under what circumstances is Raid-5 inappropriate?" - then this is something you should have somebody else do.)

I would also suggest that you reexamine your goals. For example 99.999% uptime is 5 minutes of downtime a year. You have indicated that you were after 99.99999% uptime. If you really have that sort of uptime requirement you should get your own datacenter! Most businesses can easily support 99.7% (one day of outage a year). Determine what an outage really costs you and THEN determine you uptime requirements. (There is no reason to spend millions of dollars on a solution if a full day of outage only costs you $50,000).

yah datacenters are not cheap and even with your own datacenter with generators, liebert backups, controlled environments, your upstream provider could simply null route your ip and its game over.

I've had that happen before. BGP4 is not infallible by any means.

Perhaps you should focus on more sale-able items, and live with 99.7% uptime.

Good customer service, great product, etc can all make up for the occasional server downtime. It's going to happen to the biggest isp's out there.

If you don't believe that, then look what happened last friday night at some of the biggest hosting centers in the us.

Like everything there is a "diminishing" return on what you want to offer. SLA's usually cost dearly.

What are you going to sell your 99.9999% service for? That average user isn't probably going to pay for the added costs honestly. The market is super-saturated with hosting setups that are very reliable as it is, and everyone knows that servers are gonna have issues now and then.

Personally i know plenty of people that pay ridicolous hosting fee's and get alot less than 99.9% uptime, but they are paying for other value added features, and can deal with the interruptions.

Market is tight right now, sometimes the best intentions won't pay off eh?

Ahh. I see, interesting thoughts, I was just looking to provide a very high server uptime garuntee and go the extra mile, from other hosts to try to have my customers have very good uptime.

Thank you all very much! I would still like some feedback, on a simplier and cheaper system, if there is one.