On Tue, May 25, 1999 at 08:11:44PM +0100, Leigh Porter wrote:
> Adam Sherman wrote:
> > On Tue, 25 May 1999, Patrick Greenwell wrote:
> > > > On my system, there is one NetApp, and 1+n database/web servers.
> > > >
> > > > We want scalability and redundancy so each machine should be
> > > > There is a dedicated 100Mbps link between the filer and each server
> > > > bandwith shouldn't immediately be a problem.
> > > >
> > > > The data must all be live because a user may not hit the same machine
> > > > every time he clicks.
> > >
> > > Corrections welcome, but I think you are confusing terms. What you are
> > > describing are replicated front-end boxes with a single back-end box(the
> > > Netapp.) That is not difficult to accomplish, you just have to carry
> > > state around in either the browser and/or the database via transaction ids
> > > or somesuch. What is more of a challenge is to duplicate the actual
> > > database onto multiple back-end boxes(in this case the NetApp.)
> > Lets create an example:
> > 4 Machines, all using the same NetApp for storage,
> > All running the SQL server process.
> > All running Apache.
> > The database data files can be on the NetApp, but what I want to know is
> > what is the performance going to be like if four process are using the
> > same data files.
> Depends upon how fast the NetApp box can deliver the data and they deliver data
i'm not sure what a NetApp thing is, sorry, i live in a
constricted budget world these days.
> fast as anything I have ever seen (Faster than local disk usually!) so there
> will be no
> problem there..
> If they all write to the same datafiles though I duno how it'll handle locking,
> MySQL support multiple processes looking at the same files?
how about seting up someting like 'afs' (andrew file system)
and share it with the required number of 'backup' server
components. thier is another distributed filesystem project
called coda that may offer a way to dummy up a distributed
database technology platform as well.
this sort of technology is rather complicated and mysql
currently lacks some of teh interprocss communications
components inkernel to facilitate this sort of userland tasking.
currently only (hard) realtime operating systems like qnx and os9
(these are the two i am particularly familiar with) have this
sort of functionality at the kernel integration levels required
to make this a safe and reliable exercise. we used to develop
realtime resource (farm and broad acre) mnagement systems as
awell as scada systems.
i have heard that thier are two versions of a what is known as
soft real time operating systems coming to a gpl encumbured or
bsd licenced operating system near you, that is linux (soft)
real time kernel with a freebsd strand as well.
it might be possible to use these developments and someting
like afs to build a more distributed database platfrom with
postgresql as well as having a good facsimilie built on top of
please excuse my hasziness its been some 10 to fifteen years
since i was at the development edge of distributed systems and
information gathering, storage and retrieval .. we used os9
(primarily) and had started teh migration to qnx. we did this
because the motorola subsystems were getting hard to comeby and
very expensive (here in austrlia) and teh intel based ibm pc
mother board was cheap and gave us all teh performance we
needed when under the control of qnx.
sorry i couldn't be much more help, most of my 'distributed'
infotec has fade with many years of not being used.
jonathan, old but not out yet ...
PO Box 144, Rosebery, NSW 1445 Australia