load balancing servers for gallery submitting...
 

I am thinking about setting up 2 servers each with their own ip. so when visitors a load a gallery they can be taken to either server. is this a good idea for gallery submitting? I am not sure about the features of tgp and mgp scripts so I don't know if they would pick up the 2 different ips as cheating.

I don't think you need to do anything like that just for galleries unless you will be getting more than normal traffic. I can handle over 100k uniques a day of movie gallery traffic with no problem at all on one single server..

When you get up around the 1,000,000 per day mark on movie galleries, you NEED to have two severs. It will pin 100mbits line which could be solved with another NIC .. but the large number of HTML files that galleries create (all the recips) will eventually start to bog down the server.

I'm using two servers, but just have domains seperated. If you're running less than 500,000 hits a day, even the least talented server admin's optimization should eliminate the need for a second server.

I don't know what you consider normal but I get alot more than 100k per day. I need two servers.

Anyone load balance their servers for submitting?

if you want to talk about load balancing i would talk to the people at techiemedia, they host ogrish, ogrish is on loadbalancing servers, so they have theres setup good.

jpv, you can indeed do this. There are several advantages to setting things up this way even if one given server
can handle the load. This is especially true
if you are hosting with places like servermatrix where a
given server with 1200 GB is really cheap, but overage
is expensive.

1) if one of your servers has a hardware failure you still
have the other one in operation and can direct all
the load to it.

2) when your traffic does grow you already have
a plan in place on how to scale.

There are some disadvantages though.

1) you'll need to keep your content mirrored across
the two (or more) servers. rsync over ssh is probably
the most common method of doing this and is
quite fast.

2) if you are using DNS round robin to distribute the load
you can run into some unsual issues if you have approx
23 or more servers in your rotation. The DNS system
normally uses UDP packets to communicate. However,
if the response from the DNS server is over a certain
size, the communication switches to TCP. Now,
this shouldn't pose a problem, but for some reason
it does. I found that when i exceed 23 servers in
my rotation that some people would start complaining
that they couldn't reach my site. I never figured out
the exact cause though and why it only occured on
some computers and not others.

3) 2 or more servers means more server administration.

As for your question regarding TGP/MGP scripts... to the best
of my knowledge, i have never had a gallery rejected
because it is on a load balanced system.


- jpoker

This is what I need to know. I have already decided to go with a round robin between two servers IF the tgp/mgp scripts out there don't see the redirecting between two ips as cheating.

Well these days I don't think there are many "regular" gallery submitters that get even close to 100k uniques per day. It's pretty hard to get that without submitting good galleries every day and using good partner accounts and/or paid spots IMO.

I wouldn't recommend the DNS method of load distribution. DNS entries cache, and tend to produce 'clumps' of hits... not very evenly spread. There's a couple of ways to get load spread out nice and evenly across multiple servers.

The expensive way is to invest in a hardware load balancer. There's a variety of them on the market but they all basically do the same job: Take in hits on a virtual IP address and farm them out to real servers inside your network.

Advantages: They're standalone bits of hardware. All they do is load-balance stuff. You can usually point MRTG at them and get statistics too, if you like pretty graphs. Many come with pretty front-ends so setup is a breeze.

Disadvantages: Can be a substantial cost to small-medium webmasters. Fault-tolerant operation requires multiple units, multiplying the cost.

The cheaper (and in my opinion, more flexible) way is to run a unix box with virtual server support. It's free software, and unless you're pushing a WHOLE bunch of traffic (500mbps+) it'll handle your load easily assuming you put the software on a box with sufficiently advanced hardware. You can easily get 250mbps+ out of a stock P4 Dell 1u box with 2x GigE copper ports.

Advantages: Cheap, and very flexable. If you have a resident geek, he can make this sort of setup do backflips.

Disadvantages: More difficult to set up. Will require someone with reasonably advanced technical know-how to get it running (although it's a fire-and-forget type of tool that requires very little maintenance once set up).

Load balanced servers do require data synchonization. Rsync works but multiplies your data storage requirements (not an issue if your data set is small). Another alternative is to centrally locate your files and mount them as NFS, so changes to the master immediately take effect on the front-end slave boxes.

The only other caveat is data storage. If you have scripts writing data to a file local to the machine when submissions are made, then that submission won't be known to other boxes in your load group until you sync them. If submissions come from 2 or more machines in your group simultaneously, you could lose data in a 'collision'. Scripts which use central database stores (postgres, mysql, oracle etc) don't have these problems. If your script uses MySQL, chances are this won't be an issue for you.

Hope this helps some, and doesn't just confuse you more. :thumbsup

We run a cluster of 4 front end servers that we load balance with LVS and IPVS, if you are going to do it yourself make sure you look into load balancing with ipvs and lvs on linux. The best part of this setup is you can add or remove servers as needed with no impact on the production sites. This setup for us handles massive amounts of traffic and the servers are never bogged down.

We have some load balanced servers at candid that do a great job. Pretty scalable... when needed we just pop a new server up, all the content gets copied over. If one server goes down, the rest take over its load, so little downtime, :-) I think they only charge about $100 each sever.

Just curious what you mean by "Stock P4 dell" I don't know of any "stock" server (Stock meaning no frills) that's going to push 100mbit nevermind 250mbit. (This is purely based on hard drive I/O.)

Also, to answer your question of cheating, with a hardware loadbalancer your domain will point to a vip of the router. So the gallery scanner will see that ip address, the actual ip address of the server will show up in the logs beyond that. (to the sponsors, not that they care)

Come to Jupiter Hosting. We'll show you how load balanced gallery servers are done.

We do them for most of the major adult company on the net.

(quick note... NFS is the way to go vs. rsync). You can scale more effectively. DNS Round Robin = bad)

Contact me with the info below and I'll show you some examples.

Yes, when I say 'stock' I mean basic, no-frills boxes. I could have been more specific there.

When you're dealing with LVS action, the packets never reach the disk subsystem. The packet gets read in, header modified, and right out the other network card. Very efficient.

Now if it had to hit the disk, that WOULD slow things up substantially... but it's not an issue in this case, unless you're talking about disk subsystem performance required to push 250mbit of actual content. That's a case where load balancing pretty much becomes a requirement... 10 boxes with the same data pushing out 250mbit aggregate only requires a max capacity of 30mbit, and any cheapo beige box with IDE drives can handle that.

thanks for the information. I sent some of it on to my host to see what they can do. I thought round robin would be the best because it seemed easy but now I think the best way is through hardware.

Thanks Ray for the info. I will pass on coming to jupiter. I have a solid host now :)

Despite the comments mentioned here, Round Robin is
actually just fine. Its not perfect, but I haven't
experienced any problems with it (other than when i introduced more than 23 servers in the rotation) nor have I have seen
the clumping that was mentioned above.

Load Balancers are indeed the ideal solution but if you
are using only one of them then you are introducing a single
point of failure which defeats one of the purposes of
having a load balanced system. If you can afford it, redundant
load balancers are the nice cadillac solution. But for galleries,
i would tend to stay with simple and cheap.



- jpoker.

What does the end user see when a load balanced server goes down?

What does the end user see what a RR DNS server foes down?

Very different outcomes. With RR, you loose a significant portion of your traffic. Load Balance, the hardware takes the server out of rotation.

Trust me when I say I know what I am talking about. I have been doing this since Local Director was a startup (not owned by Cisco) shipping me demo boxes to test back at my days at Schwab.

I heard good things about F5 Networks. They are not really that expensive on Ebay

BIG-IP® LoadBalancer
http://www.f5.com/f5products/bigip/LB520/

http://www.networktrust.jp/other_ntw...image/6443.jpg

Looks sweet too :Graucho

Big Ray, everything you are saying is accurate and I'm in no
way doubting that you know what you are doing.
You
are correct in saying that when a RR server goes down
that one can lose some traffic until such time as you've
taken the broken server out of the rotation and all
name servers have refreshed with new info. If you
keep your TTL fairly low, 15 minutes or so, and if you
are quick to take broken servers out of rotation (perhaps
automatically by an outage detection script) then this
can be minimized. Anyway, all i'm suggesting is that
its a cost versus benefit situation. For hosting galleries
i personally don't think that having redudant hardware
load balancers is worth the expense (actually, to be honest, i haven''t looked at recent prices - its entirely possible they
are really cheap now, but they certainly were not
a couple years ago) If thats the case, just ignore me
entirely and pretend i didn't open my mouth).

- jpoker

do you really need it?