hi ya tom
if the idea is to have a "fail safe" system....
when we were doing doa work for the gov't... we had to have
multiple redundancy....and redundancy for the redundancy...
( gov't had too much $$$ to throw at it )
but basically, how I would like to see a "HA" type linux system
would have:
dual-port scsi controller... read/write by two different servers
-
- this is the key factor...in my mind...
-
dual CPU is nice...but it's worthless(?)...
- many other points of failures...
( mb, cpu, memory, disk, connector, cables, power supp, electricity,
etc )
you really want two motherboards and/or boxes that access
the same disk... if one system fails...the other system
should be able to read/write that data..
- if one "system" fails...
than the other "system" can step in...
( multi-host servers....www1, www2, www3 acting as "www"
( for now sun boxes does this better than linux ? )
when there is a transaction like a mc/visa/amex transaction occur...
the terminal sends that request to two servers... both record the
request... one of the two machines will respond with the authorization...
and they both acknowledge that the task was done...
- if it's not acknowledged... the task is pending still cause
the other system got stuck someplace...
well...guess i just wanted to say that...."dual-port disk controller"
is what does the job best...if not... the disk contoller dies and all
hell breaks loose... or the motherboard dies and ..so on...
( my adaptec controller just died this past week...so it's an issue )
- no monthly tape archive for a while...
have fun
alvin
> Tom Kunz wrote:
>
> SW-RAID List,
> This is slightly off-topic. No, in fact it might be further than just
> "slightly". I have been exploring redundant network filesystems for
> Linux, off and on for the past several months. I need something that
> will replicate a fs across a lan, much the same way RAID-1 duplicates
> fs's. The purpose is for a high-availability system, where several
> nodes participate to keep a set of services active indefinitely. My
> company uses a SCO solution, called "Sentinel", which is just a
> single-master/single-slave arrangement that duplicates disks between two
> machines. When the master node goes down, the other assumes its IP, and
> has an identical filesystem as the dead machine. When the master comes
> back up, it resyncs to the master and assumes the slave position again.
> But we want to dump SCO and go entirely with Linux, and have the same
> functionality.
> I have already explored Coda, InterMezzo, and the Linux-HA website
> (http://www.henge.com/~alanr/ha). So far, the HA site only has the IP
> assumption source, and links to other sites I've already taken a look
> at, none of which contain what I think is necessary for a redundant
> network filesystem. Coda and InterMezzo seem more like "caching"
> solutions, for mobile or remote computing, not as a *redundant*,
> fully-duplicated filesystem. I attended a seminar at LinuxExpo in
> Raleigh, NC on GFS, but that looked like something still in its infancy,
> and relying somewhat heavily upon certain SCSI and Fiber-Channel
> features. I want something that anyone with >1 machine and any
> linux-supported disk media can use as RAID-1 data duplication.
> After considering the RAID-1 code that has come to (relative) maturity
> here, it seems like a good code base to get started on a networked
> RAID-1-type system. Basically, rather than sync-ing between two
> physically local disks, modify the code to sync between a local and a
> remote (or more than one remote) disk. Maybe I've missed something
> about Coda, GFS, or Intermezzo, or maybe there's an entirely different
> system that already does exactly what I want (as open-source, of
> course). But if not, is there anyone else on this list who is
> interested in venturing out into this arena?
> Thanks,
> Tom
>
> --
> Tom Kunz Tool Developer Software Consulting Services
> PGP Key http://www.users.fast.net/~tkunz/pgp.html
> 1452 1F99 E2BB 632E 6EAE 2DF0 EF11 4DFC
> DB62 7EBC 3BA0 6C40 88C0 C509 DA85 91B4 D5E9 EFD3
>
