Re: LISP for web

Frank Goenninger DG1SBG <frank_goenninger@xxxxxxxxxxx> writes:

> "Marco Baringer" <mb@xxxxxxx> writes:
>
> [Snippedy-Snip]
> >>
> >> Need more input? Ask!
> >
> > i'm asking :)
> >
>
> So you dare you ask, hm ? ;-) Here you go:
>
> Any large scale web app that is business critical is based on a few
> fundamental components.
>
> 1. All components are redundant
> -> So you need at least two routers, two switches, two machines ...
> And also two UPS boxes, two power supplies in each devices...
>
> 2. Load balancing is done based on content and on network traffic
> Cisco and Baynetworks are examples for producers of stuff like
> LoadBalancer that direct traffic based on content and network
> bandwidth requirements to parts of the network infrastructure and
> to the application servers
>
> 3. High availability is achieved on application level either by
> having mulptiple instances of a given app on several machines
> (which of course are running in buildings that are at least
> 2 km apart - according to NATO regulations) and/or by
> putting the apps in packages that are switchable between
> machines in case of a hardware failure. Commercial solutions
> are IBM'S HACMP or HP's MC/ServiceGuard. These solutions facilitate
> the switching based on monitoring critical resources and also
> by updating each node in a cluster about health status of each
> node. If a node experiences problems all other nodes are either
> notified actively or passively (when that node is no more
> reachable). Then the packages (a DB, a CL image proces, the Web
> server, whatever) gets shutdown on the problem node and started
> on another, pre-defined failover sequence node. In doing so the
> IP address of that app (NOT of the interface, in a HA cluster
> each application is assigned an IP address) is taken onto the
> new node's network interface. The network infrastructure is
> notified about the new MAC address to IP address mapping using
> the ARP protocol and other mechanisms. Therefore the load
> balancer, the switch and all other devices "know" that the
> app is now running on a new node. Or, more precisely, they
> don't care because the switches next to the nodes take
> care of this using IP protocol facilities.
>
>
> Even Amazon and Ebay are running based on the architecture outlined
> above.
>
> Hey, normally I do charge 1250 EUR for this type of consulting ;-)
>
> Want more? Ask!
>
> Frank
>
And to just add to Frank's outline - Red Hat is now offering a
clustered version of their enterprise solution. They have some good
documentation available and while their cluster version may be a bit
expensive, there is enough information available to "do it yourself"
if your adventurous enough. We did this when we wanted to do a "proof
of concept" at work, but didn't have the funds to invest in full
commercial versions. It worked well. The only warning is that you do
need at least 3 servers participating in the cluster. This is because
with only two nodes, each node cannot easily determine if it is the
node which should be "master" (overly simplified, but should give you
the idea). Note that the commercial clustering solutions are still
IMHO superior to "roll your own" or even Red hats current offering,
but of course everything is relative to your need and your budget.

Tim
..
--
Tim Cross
The e-mail address on this message is FALSE (obviously!). My real e-mail is
to a company in Australia called rapttech and my login is tcross - if you
really need to send mail, you should be able to work it out!
.

Relevant Pages

  • Re: LISP for web
    ... Any large scale web app that is business critical is based on a few ... -> So you need at least two routers, two switches, two machines ... ... Load balancing is done based on content and on network traffic ...
    (comp.lang.lisp)
  • Re: 3 and 4 node heartbeat network
    ... I suspect your team leads were using hubs or low end switches ... isolated vlan for just the nodes used in that cluster. ... and rely on the public network for redundancy. ... The vlans ensure ...
    (microsoft.public.windows.server.clustering)
  • Re: IP Address resource does not come up after failover!
    ... That's not the way cluster works. ... addresses - it brings them online provided the network supporting the IP ... switches, they are on the same VLAN? ... connection between the NICs of both machines and the switches (links are ...
    (microsoft.public.windows.server.clustering)
  • Re: IP Address resource does not come up after failover!
    ... There are no VLANs on the network - just several switches sequentually ... I suspect the cluster service uses IP Helper API to add/remove ... addresses - it brings them online provided the network supporting the IP ... connection between the NICs of both machines and the switches (links are ...
    (microsoft.public.windows.server.clustering)
  • Re: SQL 2008 clustering with Windows 2008
    ... If you are just building a tw0-node cluster, the safest heartbeat network is a crossover cable. ... You can team the public NICs, but you need to put them on different switches to really make that highly redundant. ... So far the failover configuration is working correctly and now I am working ...
    (microsoft.public.sqlserver.clustering)