LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

RE: GFS, intermezzo and ndb [Was Re: Many thanks for help! One last ques

To: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: RE: GFS, intermezzo and ndb [Was Re: Many thanks for help! One last question ;-)]
From: "Clint Byrum" <cbyrum@xxxxxxx>
Date: Wed, 18 Oct 2000 15:08:27 -0700
> -----Original Message-----
> From: Juri Haberland
> [mailto:news-innominate.list.linux.lvs.users@xxxxxxxxxxxxx]
> Sent: Wednesday, October 18, 2000 11:26 AM
>
>
> Joseph Mack wrote:
> >
> > John Cronin wrote:
> >
> > > Also, if only two systems are involved, some kind of network
> block device
> > > could be used:
> >
> > someone came up on the mailing list a while ago saying he had
> it running on an
> > LVS.
> > We descended on him with questions about how it was working and how
> > he set it up. He disappeared off the mailing list without any
> substantial
> > answers
>
> I doubt that drbd, nbd or enbd will help you here because they are just
> distributed block devices on which you have to set a filesystem like
> ext2.
> But ext2 or any other common FS can only be mounted on one node (rw).
> Sure, you can mount it one one node rw and on the other one ro - but
> that other one will never see changes because of caching.
>

I am currently very close to putting a drbd+reiserFS+samba storage server in
production, along with NT web servers. The solution looks something like
this:


                   [LVS 1]---[LVS 2]
                           |
              |------------|------------|
         [NT Web 1]    [NT Web 2]   [NT Web n..]
              |            |            |
               ------------|------------
                           |
                  [drbd 1]---[drbd 2]

Both the LVS and drbd machines are running heartbeat, so only one is active
at a time. When the drbd machines failover, NT happily reestablishes the
connection, and re-opens the file. Because we use ReiserFS, the failover
takes less than a minute.

This works pretty well, and may see production soon. I'm having some issues
with the NT applications failing over(one of them runs a DB, that stores its
data on the drbd server... if that one dies, another must take over for it).
While this would save us a few thousand dollars(we could just buy
Double-Take or NT Enterprise), it may not actually work because its harder
than you think to make NT failover(right now we have to force a reboot to
get the IP alias to work). For Linux web servers, this would be a great
solution however. If only I could use Linux for the web servers..... :P

> What you really need is something like NFS or GFS which can be mounted
> rw on several nodes.
>
> juri
>
<snip>



<Prev in Thread] Current Thread [Next in Thread>