Jeremy Begg
2008-01-17 01:22:00 UTC
Hi,
Process Software MultiNet V5.2 Rev A-X, COMPAQ AlphaServer DS20E 666 MHz,
OpenVMS AXP V8.3
For various reasons I'm about to split a VMScluster consisting of two nodes
which boot from the same system disk and MultiNet root directory. Each node
has its own IP address (a.b.c.1 and a.b.c.2) plus they share an IP Cluster
Alias address (a.b.c.15). All addresses are in the same subnet (a range of
31 addresses).
After the split (which will happen in about 48 hours) each machine will be
operating stand-alone with no shared files. However they will be connected
to a shared SCSI bus so that in the event of one machine failing the other
one can take over. (Sort of a "warm" failover ability.) The startup
procedures on both systems are careful not to both mount the same disks
simultaneously!
The purpose of this email is to ask how best to handle the web services which
are currently associated with the IP cluster alias address. The intention
is that the various web sites hosted on these machines will be offered on an
IP address which will be moved from one machine to the other if the
currently active machine fails. (The failover time will be in the realm of
several minutes rather than seconds or hours.)
Each machine has two Ethernet ports. So I have come up with the following
scenarios and I'd like some guidance as to which would be "best".
1. Use one ethernet port as the machines "primary", unchanging ethernet
address i.e. a.b.c.1 or a.b.c.2, depending on the machine. Use the other
ethernet port as the machines "service" address i.e. a.b.c.15. In normal
operation one machine will set this interface /DOWN and other other one
will set it /UP. The failover process will require the newly active
machine to make its second interface /UP, and the failing machine to make
it's /DOWN (although it will probably be down anyway).
2. Similar to the above, but use a pseudo-interface attached to the primary
ethernet port. The pseudo-interface IP address (a.b.c.15) will be marked
/DOWN on one machine and /UP on the other, as before. (The reason for
adopting this approach is to allow the second ethernet interface on each
machine to be used for a "private" network, but I haven't yet decided if
this is necessary for my purposes or even desirable.)
3. Retain the IP cluster alias feature on each machine but take steps to
ensure it's active only on one machine at a time. I suspect the only way
this can be done is to manually ENABLE or DISABLE the CLUSTERALIAS
service and restart the MULTINET_SERVER process. I suppose it might be
done by having a process running on the "inactive" machine which grabs
the relevant VMS lock before MultiNet gets a look-in, but first I'd have
to find out what the lock is.
4. Any other suggestions for achieving this? I'd rather not rely on dynamic
DNS updates because they take too long to propagate and dynamic zones are
a pain anyway.
I can see that an immediate benefit of option 1 will be to offer a BIND
resolver on the .15 IP address, something which isn't possible using IP
Cluster Alias (because BIND only binds to interfaces) and which I've wanted
to be able to do for a long time.
Also isn't there support in MultiNet for transparently failing over from one
Ethernet port to another in the event of device failure? I can't find
mention of this in the MultiNet V5.2 install & admin guide.
Thanks,
Jeremy Begg
+---------------------------------------------------------+
| VSM Software Services Pty. Ltd. |
| http://www.vsm.com.au/ |
| "OpenVMS Systems Management & Programming" |
|---------------------------------------------------------|
| P.O.Box 402, Walkerville, | E-Mail: ***@vsm.com.au |
| South Australia 5081 | Phone: +61 8 8221 5188 |
|---------------------------| Mobile: 0414 422 947 |
| A.C.N. 068 409 156 | FAX: +61 8 8221 7199 |
+---------------------------------------------------------+
Process Software MultiNet V5.2 Rev A-X, COMPAQ AlphaServer DS20E 666 MHz,
OpenVMS AXP V8.3
For various reasons I'm about to split a VMScluster consisting of two nodes
which boot from the same system disk and MultiNet root directory. Each node
has its own IP address (a.b.c.1 and a.b.c.2) plus they share an IP Cluster
Alias address (a.b.c.15). All addresses are in the same subnet (a range of
31 addresses).
After the split (which will happen in about 48 hours) each machine will be
operating stand-alone with no shared files. However they will be connected
to a shared SCSI bus so that in the event of one machine failing the other
one can take over. (Sort of a "warm" failover ability.) The startup
procedures on both systems are careful not to both mount the same disks
simultaneously!
The purpose of this email is to ask how best to handle the web services which
are currently associated with the IP cluster alias address. The intention
is that the various web sites hosted on these machines will be offered on an
IP address which will be moved from one machine to the other if the
currently active machine fails. (The failover time will be in the realm of
several minutes rather than seconds or hours.)
Each machine has two Ethernet ports. So I have come up with the following
scenarios and I'd like some guidance as to which would be "best".
1. Use one ethernet port as the machines "primary", unchanging ethernet
address i.e. a.b.c.1 or a.b.c.2, depending on the machine. Use the other
ethernet port as the machines "service" address i.e. a.b.c.15. In normal
operation one machine will set this interface /DOWN and other other one
will set it /UP. The failover process will require the newly active
machine to make its second interface /UP, and the failing machine to make
it's /DOWN (although it will probably be down anyway).
2. Similar to the above, but use a pseudo-interface attached to the primary
ethernet port. The pseudo-interface IP address (a.b.c.15) will be marked
/DOWN on one machine and /UP on the other, as before. (The reason for
adopting this approach is to allow the second ethernet interface on each
machine to be used for a "private" network, but I haven't yet decided if
this is necessary for my purposes or even desirable.)
3. Retain the IP cluster alias feature on each machine but take steps to
ensure it's active only on one machine at a time. I suspect the only way
this can be done is to manually ENABLE or DISABLE the CLUSTERALIAS
service and restart the MULTINET_SERVER process. I suppose it might be
done by having a process running on the "inactive" machine which grabs
the relevant VMS lock before MultiNet gets a look-in, but first I'd have
to find out what the lock is.
4. Any other suggestions for achieving this? I'd rather not rely on dynamic
DNS updates because they take too long to propagate and dynamic zones are
a pain anyway.
I can see that an immediate benefit of option 1 will be to offer a BIND
resolver on the .15 IP address, something which isn't possible using IP
Cluster Alias (because BIND only binds to interfaces) and which I've wanted
to be able to do for a long time.
Also isn't there support in MultiNet for transparently failing over from one
Ethernet port to another in the event of device failure? I can't find
mention of this in the MultiNet V5.2 install & admin guide.
Thanks,
Jeremy Begg
+---------------------------------------------------------+
| VSM Software Services Pty. Ltd. |
| http://www.vsm.com.au/ |
| "OpenVMS Systems Management & Programming" |
|---------------------------------------------------------|
| P.O.Box 402, Walkerville, | E-Mail: ***@vsm.com.au |
| South Australia 5081 | Phone: +61 8 8221 5188 |
|---------------------------| Mobile: 0414 422 947 |
| A.C.N. 068 409 156 | FAX: +61 8 8221 7199 |
+---------------------------------------------------------+