|
Markus Bernhardt wrote:
>
> Hello
>
> Our szenario:
> We have to store 400GB+ in a self written database.
> Main problem is indexing. We need 10 - 15 GB of RAM.
> There is not enough money to by an S/390.
> We want to use Linux.
>
> Our idea:
> We splitt the data in smaller portions and distibute them on several
> machines with 1-2GB RAM.
> Then we are using LVS to present it external as one server.
>
> Our question:
> Is it a good idea, to use the scheduler to analyse a small portion
> of the request
> (it has to analyze, which portion of data is needed)
> before it redirects the packages to the real servers ?
No, it is not a good idea to solve your problem using LVS. In the LVS
cluster, every real server is assumed identical, which mean the same
content (such as web content) and the same services on each server, so
that there is no difference if the request is sent to the real server
A or the real server B.
In your scenario, data is partitioned on the different servers, so
that request task must be partitioned too and those sub-tasks are
shipped to different servers where data is. You might use PVM/MPI to
save your programming effort to build your system, however, it seems
to me that your major work is on how to design a good scalable
architecture of your system, how to partition your data so that the
worse data skew can be avoided.
Wensong
>
> ----------------------------------------------------------------------
> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> To unsubscribe, e-mail: lvs-users-unsubscribe@xxxxxxxxxxxxxxxxxxxxxx
> For additional commands, e-mail: lvs-users-help@xxxxxxxxxxxxxxxxxxxxxx
----------------------------------------------------------------------
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
To unsubscribe, e-mail: lvs-users-unsubscribe@xxxxxxxxxxxxxxxxxxxxxx
For additional commands, e-mail: lvs-users-help@xxxxxxxxxxxxxxxxxxxxxx
|
