Re: Whitepaper on XML-based Network Management

["David T. Perkins" <dperkins@dsperkins.com>]

Phil,

While I'm excited about using some of the technology from XML, I fear
that you are repeating history with a new flag.
The situation is something like:
1) management is hard, so lets make it easier.
2) management app developers don't want to deal with all of
   the inherent problems of writing a distributed application
   where communication is not reliable, and there can be
   variations between different managed devices (and between
   different system software versions on a particular device),
   so lets make it easier.
3) the details of a management protocol don't match the programming
   language that management app developers are using, so lets
   make it easier.
4) etc

Everything that has been said is true, but the managed device
side of the equation has been left out. Also, the evolution of
experience has been left out.

There are high-end devices that have dedicated processors, lots
of memory, and lots of persistent storage (such as flash or even
disk drives) where putting in management mechanisms with high
processing and memory needs does not add a significant burden.
However, there are also devices where this is a burden. A compromise
position is needed.

On the evolution issue, it has been shown time and again that when
a new technology is created because there is no usage base (experience),
it is difficult to guess what aspects of the technology need to be
managed, and to best organize, and identify the management information.
After you gain some experience, a competent engineer can almost
always re-engineer the management to remove unneeded stuff, add
stuff to solve outstanding problems, and to re-organize and possibly
change the identification of management information to make it
more useful and easier to write code in managed devices and management
applications. 

So, does XML really provide an approach that is fundamentally better
than SNMP? I don't believe that your message or the white paper provides
convincing evidence. You have provided some evidence that for one
market segment there is some potential, but you presentation has
been all one sided.

Now, please remember that so far I have found many attributes that
I like about XML, but there are other attributes such as the poor
efficiency of using text based messages that I find horrible. There
is a lot of religious furor wrapped up in the text vs binary encoding,
and it would not be useful for responders to this message to bring
up those old issues to rehash. 

So, I suggest that a balanced presentation of XML vs SNMP be put
together that includes some real numbers. This would help with
an informed decision on the adoption of XML for management.   


At 04:22 PM 7/30/2001 -0400, Phil Shafer wrote:
>A bit of followup.....
>
>>From my discussions with network operators and their internal tools
>groups, the concensus is basically:
>
>- SNMP is great for simple jobs, but gets cumbersome and inefficient
>for more complex tasks. Comments on ops-nm echo this.
>- Expect is the tool of choice for a large number of tasks, but
>the lack of control and variability of the responses mean that scripts
>are too fragile and maintenance costs are too large.
>
>XML-based Network Management (XNM?) gives a better solution than Expect
>for parsing data emitted by a network device, since it provides both an
>exact description of the data being transmitted and a compatibility
>mechanism for handling future (and past) versions of software. It can
>handle complex datasets and hierarchical relationships.
>
>XNM gives tools developer (internal, public, and commercial) the
>ability to handle complex data with standard programming libraries
>and to build on the shoulders of the fresh crop of XML technologies.
>For example, XSLT allows a vendor-neutral configuration file to be
>converted into a vendor-specific one. Once a definition is made
>for vendor-neutral configuration, vendors can provide the XSLT
>transformations to make vendor-specific configuration files.
>
>Most of the customers I've talked with have their own tools group;
>most avoid commercial tools because of version and feature lag. They
>are excited about the possibilities this technology holds. Some folks
>are already using XML internally, and see expanded use in their
>future. 
>
>So the question is: Is my sample of network operators off by more than
>one standard deviation? It's likely skewed toward larger operators,
>but are the above readings representative of the operator community?
>How far off the mark am I?
>
>Thanks,
> Phil

Regards,
/david t. perkins