 rchandraStargate Universe fanPremium
join:2000-11-09
14225-2105 | reply to talz13
Re: I admire the tenacity to even bother. DNS round robin is one way, as another poster has pointed out. There are lots of references on the Web which you can check out. The Ultramonkey project is one such that makes it easier to set up load balancing on RPM based Linux systems. It is basically a "distro" of LVS (Linux Virtual Server). LVS not only checks the health of the servers being load balanced but also (optionally) has a hot standby LVS machine. The key here is there is some way of periodically sending an "are you working?" message to other LVS machine(s) (could be over the network, could be a serial link), and failing over if the other LVS machine seems dead (IP address/MAC address/ARP takeover, that sort of thing). In the router realm, there is HSRP and VRRP, which is similar in concept of having hot standbys. There is also anycasting, which is used, among other things, for the root nameservers. Some root nameserver goes offline or gets DDoSed? No problem, just update the routing tables (BGP I assume) to send the traffic other places, or even to servers which are network-wise closest to the client which is querying.
--
English is a difficult enough language to interpret correctly when its rules are followed, let alone when a writer chooses not to follow those rules.
Jeopardy! replies and randomcaps REALLY suck! |
|
|
|
sk1939Premium
join:2010-10-23
Washington, DC
kudos:9 | The only issue with that though is that there will still be downtime unless you have the backup site designated as a hot-standby or is always online as a failover since BGP is notoriously slow to update routes (important for the internet). |
|
rchandraStargate Universe fanPremium
join:2000-11-09
14225-2105 | Oh sure. There's almost always going to be some downtime. But hopefully it can be measured in seconds or minutes as compared to days or weeks for something like a (utility) power outage. It all depends on the engineering chosen.
BTW, thanks for that. I knew of the technology (BGP), but I have no practical experience in having to reroute traffic.
--
English is a difficult enough language to interpret correctly when its rules are followed, let alone when a writer chooses not to follow those rules.
Jeopardy! replies and randomcaps REALLY suck! |
|
sk1939Premium
join:2010-10-23
Washington, DC
kudos:9 Reviews:
 ·T-Mobile US
1 edit | said by rchandra:Oh sure. There's almost always going to be some downtime. But hopefully it can be measured in seconds or minutes as compared to days or weeks for something like a (utility) power outage. It all depends on the engineering chosen.
BTW, thanks for that. I knew of the technology (BGP), but I have no practical experience in having to reroute traffic.
No problem. For BGP failover it depends on the peers (have to match timing) and varies depending on provider. BGP still fails-over in a matter of minutes though for the most part (Cisco's default timer is 3 minutes I believe), but is still slower than say EIGRP or OSPF which is almost instantaneous. A lot of organizations also run BGP internally if they have a large enough network. Edit: » www.enterprisenetworkingplanet.c···ting.htm» www.petri.co.il/csc_what_is_bgp.htm |
|
