On Mar 9, 2009, at 5:54 PM, Darren Reed wrote:
On 8/03/09 12:27 AM, Dale Ghent wrote:
On Mar 7, 2009, at 4:54 PM, Jason King wrote:
Since the ball has already started to roll on this, I'd like to
'officially' ask for feedback on my current proposed design for link
layer discovery support (specifically LLDP -- possibly in the future
other protocols such as CDP or EDP can be added in). The current
documentation can be seen at
http://www.opensolaris.org/os/project/lld/design
Huzzah! Awesome work, Jason.
I was reading over the Operation document and the paragraph
describing what will populate the System Description TLV and a
thought occurred to me.
One of the components you're using to populate this TLV is
SI_PLATFORM from sysinfo. On SPARC boxen, this will produce a
string akin to "SUNW,Ultra-4" or "SUNW,Sun_Fire_T200" and so on.
However, on x86, this will always be "i386". I think that can be
improved.
That's not his fault.
He's returning the same information that is available from uname.
Getting the specific box platform name, even with Solaris on Sun
x86 hardware, requires some other special command... that escapes
me at present.
I know it's not his fault. He's using whatever sysinfo(2) (and by
extension, the kernel) says SI_PLATFORM is, which on x86 means nothing
of real substantive utility when compared to results one would get/
expect on a SPARC platform.
Example:
SI_PLATFORM on SPARC would return the product name, or a very close
approximation of it such as "SUNW,Netra-T12" and such.
SI_PLATFORM on x86 will, as currently implemented, always return
"i86pc" be it on a Dell R905 or a IBM Xseries... or a Sun X4540.
Granted, the SPARC example I gave with the value of "SUNW,Netra-T12"
does not exactly mirror the actual model of the server it represents
(in this case, this is from a E2900), but it's a bit less vague than
just "sparc" and gives one a general idea of what the actual hardware
model is when viewing LLD telemetry. It's also a given that a
particular platform can be used on differing models... take the Ultra
5 and Ultra 10, for example. Luckily it seems that Sun is giving
different models, even if based on the same or similar motherboards,
their own unique platform strings (eg; the motherboards used in the
T5120 and T5220, while almost identical, have different platform
names). So, it isn't 100% accurate in some older cases but it's
certainly better than being vague... which is what we have on x86
defacto.
Now to your hint on the special commands thing... on both SPARC and
x86 one could get an actual model name. On x86, one would use the
libsmbios facilities to get that, as manufacturers of systems are
pretty good about putting the computer's (or motherboard's) model name/
number in there. On SPARC, one could extract the "banner-name"
property from the eeprom (this is how prtdiag on SPARC gets the model
name into its output)
I'm just suggesting that lldd be detailed in this respect, because
looking at it from the other end, network device vendors that
implement LLD or CDP do put the device's model in their LLD/CDP
packets... You just don't get "switch" or "router".
I like Jason's idea of making these bits configurable, but even for
the default values it might be good for us to try hard and be detailed
about it... to mirror in practice what is found on existing LLD
implementors (ie, the network device vendors). To a network engineer
who is inspecting something through LLD, the machine is no longer
mystery meat, but is something he can put a make, model, and perhaps
even serial number (sneep!) to.
/dale
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss
