Slash Boxes

Slash Open Source Project

This discussion has been archived. No new comments can be posted.
Display Options Threshold:
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • LVS is a great system, and works _extremely_ well.
    For best scalability you are best advised to use a VS-DR - direct routing, since this allows the servers to reply directly to the clients. (just make sure you switch off ARPing on the private interfaces or you'll get massive headaches)

    Also I'd also use IP based client persistance since that way you'll make running the service much simpler. Other things - journalling FS? How will you share data between servers? Replication? NFS? Database server? etc...

    I know of a location using LVS to serve ~100 million HTTP requests per day, all load balanced using LVS, so you can be sure LVS will scale :-)
  • We'll be using NFS.

    We're starting to build the machines today (hopefully).

    We're stuck on the network topology. As we have it now, everything is on the same switch. It looks as though the best way to lay the network out is have the web servers isolated behind the VIP (and it's backup). So we would have to add another nic to each webserver so that 2nd nic would be on the normal network backbone.

    I've read about people putting all the machines on the same network on the LVS listserv. But that seems like it's a hack.

    Has anyone done this with the machines on the same network?
    lottadot []
  • The "best" LVS setup - most scalable is VS-DR - which you can use in your situation, and it should work very nicely. If you need a hand with setting up the LVS side of things drop me an email at zathrus AT mad DOT scientist DOT com. Having the web servers isolated behind the VIP is good for security, bad for throughput. Having the setup you're suggesting all on the same switch is good for throughput. Having all the servers at different network locations is the best since you can then take advantage of bandwidth at various locations.
  • i've just finished writing a book for o'reilly on server load balancing. it doesn't cover LVS specifically, but the concepts and topologies are the same. i'd be happy to take a look.