Scalability

[msmedley]

I'm trying to do a little bit of thinking head towards the future, and I'm interested in what I can do for scalability, for both the web server and the database server.

First, the web server. Is there some sort of load balancing feature for Apache? For instance, I would like a user to login to www.mydomain.com, and have a load balancer redirect the user to www2.mydomain.com, www3.mydomain.com, www4.mydomain.com, etc. Once assigned to a server, all subsequent requests would have to also be made to that same server, because of the session data being server -specific Anyone know of how to do this within a managed environment like this?

Next, the database server. I'm thinking replication? Is MySQL replication stable enough for production use? I'm presuming I would have to use the transaction logging version of MySQL, which is slow? Maybe I don't know too much what I'm talking about, but that's why I'm asking.

The point here, is I want to design my application ahead of time to support adding web servers and/or database servers as the need arises.

And, oh,

[Gwaihir]

Apache as the load-balancing redirector doesn't make too much sense. There are much lighter applications for that, as well as affordable hardware solutions.

Which managed environment "like this" are you refering to? If you have a managed dedicated server, you can scale up to multiple managed dedicated servers. If you aren't there yet, you'd first scale up to having a server all to yourself.

The www1-www4 problem with sessions isn't that big anymore these days. There's are easy ways to share such session data between servers, such as storing session data in the database and / or (for PHP) using memcache (a memory shared between servers).

If you have a significant amount of (public) static data to serve (images, downloads), it makes sense to split that off first. Getting that from one (or a few) seperate subdomains will also make your site load faster for your clients.

MySQL replication is production stable (especially the master-slave configuration), but not an easy thing to handle. Is your application such a tight whole? Couldn't your first split your data in several databases, each running on their own server? MySQL 5.0 series has support for accessing the ocassional table from one server to another (FEDERATED engine), as you might need for that.

[Vin DSL]

I'm trying to do a little bit of thinking head towards the future, and I'm interested in what I can do for scalability, for both the web server and the database server...

Which managed environment "like this" are you referring to? ...If you aren't there yet, you'd first scale up to having a server all to yourself.

Personally, I wouldn't be too concerned about planning for the future, e.g. scaling up last century technology.

Considering the exponential growth of JagPC, and it's competitors -- and watching how this is all panning out -- here and elsewhere -- I *think* it's obvious that JaguarPC is going to have to switch to some form of 'Grid Hosting', so called.

If they don't, the sheer size of this place will kill them, and the competition will bury them!

Jag's been 'talking' about going with H-sphere, or something similar.

As a matter of fact, I think these forums are already clustered... so, hang in there...

[Vin DSL]

You might find this thread interesting...

http://jaguarpc.com/forums/showthread.php?t=14964

[Gwaihir]

Considering the exponential growth of JagPC, and it's competitors -- and watching how this is all panning out -- here and elsewhere -- I *think* it's obvious that JaguarPC is going to have to switch to some form of 'Grid Hosting', so called.

Well, at least some philosophy that involves more expendable components, at the machine level. Google did this early on and has been leveraging it ever since. Sites / places (I've been following two over the years) that took the route of making the single machines as robust as possible are doing fairly well too, but keep running into pretty long (planned or unplanned) maintenance windows.

The versions of this technology that are publicly available aren't quite mature yet though and / or they need really good specialists to maintain. For example the "board is too busy" errors are new here since the cluster arrived. So, it much depends on what term you're thinking here. In IT stuffs it's hard to see the long term (> 3 years away) obstacles, no matter what, so I wouldn't usually worry about them. For the shorter terms, I'm not so sure about these clusters. It can easily be a year before JagPC starts offering cluster based hosting. It can easily be another two after that before it is mastered to a point that physical dedicated servers have become the second choice option. I said "JagPC" here, but I don't think others are (far) ahead. So, if the topic starter is indeed afraid he's growing out of one dedicated machine soonish, I wouldn't be thinking clustered hosting yet, if I were him.

As for the size of this place killing JagPC, I don't think that's what I'm witnessing. Rather the opposite: it's not more machines that's giving trouble, it's the size of individual machines that reaps havoc lately. If one of these monsters topples, it's just too big a problem: lots of clients affected at once, lots of work to get it going again or to move over all that data to elsewhere.

[the_ancient]

As for the size of this place killing JagPC, I don't think that's what I'm witnessing. Rather the opposite: it's not more machines that's giving trouble, it's the size of individual machines that reaps havoc lately. If one of these monsters topples, it's just too big a problem: lots of clients affected at once, lots of work to get it going again or to move over all that data to elsewhere.

yep that seems to be the problem......

[Vin DSL]

I wouldn't be thinking clustered hosting yet, if I were him.

YGPM...

[Connie]

Lot of conjecture, based on someone who just wants to balance server load with different sites.

I'm sure Amazon.com does this. I wonder if the person asking the original question really needs this.

I think Jag could do the load balancing if someone wanted to pay the price.

Do most sites hosted at Jag need load balancing? I doubt it.

When considering load balancing you also need to consider the SEs. Load balancing is more than delivering visitor A to site B.

[Vin DSL]

Lot of conjecture, based on someone who just wants to balance server load with different sites.

I'm sure Amazon.com does this. I wonder if the person asking the original question really needs this.

I think Jag could do the load balancing if someone wanted to pay the price.

Do most sites hosted at Jag need load balancing? I doubt it.

You're missing the big picture, Connie...

Most of us are just renting space here, on JaguarPC servers, and it's THEIR servers that need the load balancing, not necessarily our individual web sites.

Secondly, this is the way of the future -- and the future is now! Jag is aware of this, I'm sure, and the train is leaving the station. If they wanna stay on the cutting edge and remain a leader in the industry, they need to hop aboard -- the sooner the better, if they want a good seat!

[Gwaihir]

I'm sure Amazon.com does this.

Auch.. boy, did you pick the wrong example!

I'm sure they used to do this. Amazon however, is one of the pioneers of grid solutions, which they're now offering to others as well (specifically check out the Simple Storage Service and the Elastic Compute Cloud).

[It's pretty different from anything JagPC does, so I hope it's not a problem that I'm this specific out here.]