My query concerns parallelization processes / applications.
The situation is roughly as follows:
I have x servers in x datacenters, which are highly unreliable
(availability, quality of the store (one hdd etc.), But have large
amounts of power.
Then I have a second "group" of servers, which has less power, but the
quality and fast connectivity and is a reliable store.
The idea is that some of the servers of the second group (a stable, high
quality) were loadbalancers, which would exercise (especially apache)
allocated to individual unreliable store, which would process requesty
(apache php) and the data would take the nfs annexed from reliable servers.
Unreliable services would likely operate on the principle netboot to not
at all rely on one hdd, etc. All the servers could be in one VLAN to a
private IP range. (It should take into account the geographical
diversity of servers, where some are deployed after Europe, some in the
I presume that the rising and outgoing servers would provide
loadbalancers completely transparent.
The question is how to deal with flavor when the user starts to download
a large file 10 gigabytes, for example, will run from one server, but
after the withdrawal of 2 gigabytes fall. It is possible to completely
transparently and without interruption of operations taken over by
It is this consideration is good, or completely zcestná and the other is
a better solution?
Thanks and I look forward to your tips for a solution.