Re: [Xen-users] entry-level XCP server using Intel desktop components

["Luke S. Crawford" <lsc@xxxxxxxxx>]

On Fri, Feb 10, 2012 at 08:17:22AM -0600, Scott Damron wrote:
> Why on earth would you need to administer XCP via a serial port?!?

Logging, mostly.   What caused that crash last night?  Was it a hardware 
problem?  a software problem?   With most KVM over IP setups, you have 
no idea.  With some (like the KVM over IP capabilities of some DRAC 
cards)  you can get a little video of what the monitor says right before 
a crash, but aside from the fact you can't cut and paste, a lot of times 
the monitor is on some management GUI and doesn't get the backtrace.
(also, the 'video before a crash' feature is usually only available on
the higher end models, and like I said, I've never seen it work
particularly well.)

At scale, serial ports have the additional advantage of being grepable.   
One place I worked we had thousands of servers logging to a central place.
I could quickly use grep and similar tools to find servers with bad ram,
bad hard drives, and similar problems.  

If you setup serial properly and linux crashes, the backtrace goes
to the serial port.  Setting up text logging of a serial console is
trivial (I use conserver for this, but there are a thousand solutions 
that will work)  

If you only have a few servers, you can setup a 'buddy system' so that
server A watches console for server B and server B watches the console of 
server A for nothing more than the cost of a usb-> serial dongle and a 
null modem adaptor.  (I used the onboard db9 as the serial console for that
server and the usb-> serial to log the console of the other server in the 
pair.)

Once you have more than a few servers, you can get nice, new opengear brand
48 port console servers for under $1500.  You can go on ebay and get used
48 port cyclades ts3000 console servers for around $200 each. 

Personally, I think it's extremely irresponsable to put any server in 
production without a serial console.   Without a serial console, those
irritating 'it reboots once a week'  errors keep coming back until you 
decide to chuck the hardware.  With a serial console, you have a fighting
chance of figuring out the problem after the first crash.

In testing, it's less important, but it's still pretty nice for those times
when you are trying to figure out of the instability is because you
cheaped out on test hardware (and we've all done that)  or because there
is a genuine problem with the product.  

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users