Evan Hayes wrote:
>> For the USB, I would *suspect* you would use a data transfer style
>> cable (cross over equiv.?)... But I am not sure, hence my question
>> this morning. :)
> USB is an asymmetric bus with one host and multiple peripheral nodes.
> Plus USB transfer cables probably use proprietary Windows software.
> Firewire is acyclic and more suited to this pseudo-networking

Foo, figured it'd probably have been mentioned more had it been feasible.

>> As for long throw, I haven't even begun thinking of this as I am
>> still working on the basics. For serial, I would probably use those
>> DB9-RJ45 adapters and run a long CAT5(e) cable. Not sure if there is
>> something equivalent for USB.
> They exist, but $50/pair
> 
(http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=2893
288&Sku=C250-3868) 

Ouch, at that point I'd rather look the multiport serial cards. :/ 
Firewire would be no better as I'd need a card anyway as the servers 
don't ship with firewire. I was hoping to save a scarce slot. :)

>> In my case, I would have the nodes in the same rack for now. It's
>> certainly an interesting question, and one I plan to look at down the
>> road as I learn more.
> While it doesn't protect from a TCP/IP stack, it'd probably be the most 
> cost and space efficient to plug a second NIC in each server into a 
> cheap n-port switch on a private IP space.

I've already got a second NIC used for backup heartbeat, but just that, 
it doesn't protect against the TCP/IP or general network stacks going 
down. :/

Madi

--------------------------------------------------------------

I apologize in advance for the broken web mailer that I am using that can't 
seem to quote properly...



I am not sure I am understanding your response, so please forgive me if this 
sounds dumb...

How does this not protect against a failure of the TCP/IP stack on one of the 
machines in the cluster?  If the TCP/IP stack falls over and fails to do what 
it needs to do, the heartbeat will die and trigger the failover condition just 
as it would for any other communications failure using any other means, such as 
serial.  Where this fails to function is if the switch that you are using, 
pre-supposing you're not using direct NIC-to-NIC cabling, fails to communicate 
properly and all of the servers in the cluster make the assumption that the 
other(s) have failed and that they are now the master.

What am I missing?

_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to