keywords: comptetive uptime, evolution, shared nothing, aws, amazon, ai, skynet, terminator, replication, cloning, survival of the fittest, fantasy, a song of ice and fire, eragon, server, hosting, cloud computing, wintermute, neuromancer
quote: big visions / great dreams / little tiny living space
problem: clients want continuous non-stop uptime
problem: client's don't have an arm and a leg to spend on it
problem: existing technologies (marathon's everun, stratus ftServer, vmware's vspshere) are way beyond the reach of most small companies
goal: get sub 4 hour disaster recovery
goal: get autonomous self healing recovery
goal: use commodity hosting
goal: use off the shelf components
goal: "practice failure" technique compatible (contact josh)
problem domain:
- no single failure point
- must survive network failure
- must survive power failure
- must survive disk failure
- must survive application failure
the above are common - everun, ftServer, vsphere all handle multi failure points
BUT they don't encapusalate network failure - more vendors are needed - network vendors (at least two) - but network hardware vendors etc etc etc
but your too small to have redundant networks and BGP routing so you put your servers in a data centre
BUT everun, ftServer, vpshere don't heal themselves / recover from the failed / degraded state - ie. software can't fix hardware - a process can't replace a failed raid card, hard drive, power supply, nic, or cable.
UNLESS we use libcloud - http://incubator.apache.org/libcloud/ or equivalent (simple cloud php api) with the knights and kings design pattern
quote: big visions / great dreams / little tiny living space
problem: clients want continuous non-stop uptime
problem: client's don't have an arm and a leg to spend on it
problem: existing technologies (marathon's everun, stratus ftServer, vmware's vspshere) are way beyond the reach of most small companies
goal: get sub 4 hour disaster recovery
goal: get autonomous self healing recovery
goal: use commodity hosting
goal: use off the shelf components
goal: "practice failure" technique compatible (contact josh)
problem domain:
- no single failure point
- must survive network failure
- must survive power failure
- must survive disk failure
- must survive application failure
the above are common - everun, ftServer, vsphere all handle multi failure points
BUT they don't encapusalate network failure - more vendors are needed - network vendors (at least two) - but network hardware vendors etc etc etc
but your too small to have redundant networks and BGP routing so you put your servers in a data centre
BUT everun, ftServer, vpshere don't heal themselves / recover from the failed / degraded state - ie. software can't fix hardware - a process can't replace a failed raid card, hard drive, power supply, nic, or cable.
UNLESS we use libcloud - http://incubator.apache.org/libcloud/ or equivalent (simple cloud php api) with the knights and kings design pattern
Comments