tony bourke wrote:
> Has anybody here had any experience iwth this debate?
> There are three ways to get the packets to get rewritten after leaving the
> Bridging-path: When the return packets traverse the load balancer by
> being in the Layer 2 path. This is how the switches usually work, as well
> as the localdirectors.
> Routing-path: When the return packets traverse the load balancer by being
> in the Layer 3 path, the servers have the load balancer as their default
> route. This is how devices such as F5's BIG-IP work, as well as most of
> the other server-based units. Most of the switches can operate this way,
> as well.
> Direct Server Return: This is where the server re-writes the packet
> itself by implementing some ip/arp trickery.
> I prefer using the layer 3 routing-path method, since layer 3 redundancy
> is better than spanning tree. It also eleminates briding-loops and
> such. Anyone else have a comment on that?
Disclaimer: I've only done this with Alteons. Just starting to work
w/LD and so far I find them annoying and limited.
I prefer your method (layer 3). I don't like having my topology
complicated by having a l-b switch on top of a regular one. Don't like
my redundancy relying on spanning tree rearranging my topology on the
fly. The Alteon limitation of 1 spanning tree domain per box is
annoying. I'd rather have chatty multicast from VRRP than ethernet
Spamming Tree broadcasts, though that's not based on anything
Having said that, my company's networking gods prefer the layer 2
method. I believe this is because they don't like/trust anything that
routes and does stuff like VRRP if it's not a Cisco. Also, I think they
don't want to take a fast box and make it do slower stuff (IP
forwarding). Eh... Every Alteon I've ever seen has never broken a
sweat in it's life...
What's in the third category? Resonate?
This archive was generated by hypermail 2b29 : Fri Oct 13 2000 - 19:37:55 EDT