On Tue, Dec 28, 2010 at 7:56 AM, Ozz Nixon <ozznixon@gmail.com> wrote:
> Is it possible (and how) to implement a data path on another partition (linux) for an existing system? And then if I
donot gain anything, merging it to the production /data path?
>
> Scenario of what I want to achieve (/mnt/data is already running)
>
> /mnt/data resides on an NFS share
> Contains over 2 Billion web sites crawled (yeah another search-engine site)
>
> want to add:
>
> /opt/data resides on internal drive
> Will contain keyword hash system
>
> Then if I find this does not improve anything - or runs tight (running on IBM Blade center with 76gb internals - so I
maybe limited), that I can simple shutdown postgres, scp /opt/data/folder/ to the NFS - bring up postgres - fix any
conffile, and everything is on the multi-terabyte array?
You can't just merge two pgsql databases into one directory. You have
to pick one to run and dump the other and load it into the other db
with pg_dump / psql / pg_restore. If you completely replace the db
instance in /opt/data/folder with the other one that will work, but
erase the original db that was there.
> Checking the pulse of all you speed freaks that just hit the floor because this is on NFS not iSCSI -- don't worry,
whenthis site goes live it will be iSCSI. We have been in private development for 3 years... and the NFS still out runs
theInternet connections.
It's more an issue of reliability in case of crashes and power losses
than anything else. most iSCSI implementations will honor fsync
properly, but many nfs mounts will lose data / corrupt your data store
if one or the other machine crashes in the wrong way.
