Re: [MIB-DOCTORS] AD review of draft-ietf-adslmib-vdsl2-06.txt

["David B Harrington" <[e-mail]>]

Hi,

It is almost always important to remember the audience for a MIB. 
This MIB will be used by operators and NMS developers.

Operators and NMS developers are accustomed to Truthvalue TCs.
Many NMS applications already have been coded to present the values as
"true" and "false" for Truthvalue TCs. When those many NMS
applications parse this MIB module and get the special-case TCs, the
NM application writer will need to write new special-case code for
these special-case TCs.

People who helped develop the managed protocol are probably NOT the
same people who are going to work on the helpdesk or in the IT
department monitoring the values and behaviors of the network on a
daily basis, and are not the people who will write NM applications
(unless they will write special-case NM applications for this MIB
module).

So while it might be more "natural" for the protocol designers to
think in terms of 0-1, the end-users of the MIB are accustomed to
TruthValue TCs, and NMS applications already know how to deal with the
standard TCs rather than the special-case TCs being used in this MIB.

Interoperability in the world of network management protocols, like
SNMP and MIBs, is all about making it possible for NM applications to
talk to managed devices in a manner that is consistent and
vendor-neutral, and managed-technology-neutral. That is why SNMP has a
standard set of error codes, rather than different error codes for
each managed technology. That is why SMIv2 organizes data into
two-dimensional tables rather than in deep nested structures, even if
the underlying technology uses deep nested structures. That is why
SNMP has standard datatypes, rather than different datatypes for each
managed protocol. The special-case datatypes being proposed as TCs in
this MIB module seem to be simply designer-preference - we like 0-1
better than 1-2. 

I think there is a strong interoperability justification for using
TruthValue TC rather than MIB-specific TCs with the same purpose.  

Over thwe past ten years or so, we have had it drilled into our heads
by the user community that we need to meet the interoperability needs
of the readers and users of MIB modules, such as operators and NM
application developers, more than the personal preferences of the MIB
module writers.

I think this MIB should use TruthValue, not special-case TCs.

dbh

> -----Original Message-----
> From: Romascanu, Dan (Dan) [mailto:[e-mail]] 
> Sent: Monday, February 16, 2009 7:29 AM
> To: Moti Morgenstern; David B Harrington
> Cc: MIB Doctors (E-mail); [e-mail]; Menachem Dodge; 
> [e-mail]; [e-mail]
> Subject: RE: [MIB-DOCTORS] AD review of 
> draft-ietf-adslmib-vdsl2-06.txt
> 
> See my answers in-line. David, please jump in if you have
objections.
> 
> Thanks and Regards,
> 
> Dan
> 
> 
>  
> 
> > -----Original Message-----
> > From: Moti Morgenstern [mailto:[e-mail]] 
> > Sent: Friday, January 30, 2009 12:40 PM
> > To: Romascanu, Dan (Dan); David B Harrington
> > Cc: MIB Doctors (E-mail); [e-mail]; Menachem Dodge; 
> > [e-mail]; [e-mail]
> > Subject: RE: [MIB-DOCTORS] AD review of 
> > draft-ietf-adslmib-vdsl2-06.txt
> > 
> > Hi Dan and David,
> > 
> > The editors reviewed the additional comments against revision 
> > -06 and our response is the following (numbered according to 
> > the comment numbers):
> > 
> > 1) As explained in our previous reference to this issue, we 
> > prefer the special TCs we use instead of replacing them with 
> > the TruthValue TC, which we do use where appropriate. This is 
> > also because the use of TrutrhValue TC is not mandatory (That 
> > can easily understood from the [RFC2119] interpretation of 
> > "SHOULD NOT"). 
> 
> OK - still not best IMO but I can live with this. 
> 
> > 
> > 2) The comment is not so clear. 
> > Do you mean that, instead of xdsl2Notifications OBJECT 
> > IDENTIFIER ::= { vdsl2 0 } we should use xdsl2Notifications 
> > OBJECT IDENTIFIER ::= { vdsl2MIB 0 } so the notifications 
> > will reside directly under the MODULE-ENTITY?
> > We'll appreciate a quick answer for this, if possible.  
> 
> I clarified this I think in the separate mail sent yesterday. 
> 
> > 
> > 3) We'll fix the abbreviation inconsistency reported.
> 
> Thanks. 
> 
> > 
> > 4-6) xdsl2LineCmndConfBpscReqCount will be renamed as 
> > suggested and we'll also change its syntax. However, we 
> > believe that there is a misunderstanding of the mechanism for 
> > which we use that object. Therefore we intend to change the 
> > DESCRIPTION clauses of this object and xdsl2LineCmndConfBpsc 
> > to clarify the dynamic behavior of the procedure. The main 
> > point to explain is that the SNMP agent does not allow 
> > interrupting a process which is in progress. The counter only 
> > allows the manager to identify the reported measurements are 
> > result of "his" process (i.e., they may belong to a process 
> > that was executed later on, after his process was already 
> > complete).   
> 
> OK - pending on the exact edits you need to come with. 
> 
> > 
> > 7) The results of a line diagnostic tests have their own 
> > table (xdsl2SCStatusSegmentTable) and therefore 
> > xdsl2LineSegmentTable is not overloaded.
> 
> OK
> 
> > 
> > 8) The xdsl2LineCmndConfBpsc and xdsl2LineCmndConfLdsf are 
> > associated with different results tables.
> 
> I am not sure that you understood our concern here. Exactly the fact
> that there are two different tables is the reason of concern 
> because the
> tests can happen on the same line simultaneously, at the 
> request of two
> different managers, one writing  xdsl2LineCmndConfBpsc and another
> xdsl2LineCmndConfLdsf. Can this happen? If not, please explain what
we
> are missing. 
> 
> > 
> > 9)
> > a) The document already describes the coexistence of RFC 
> > 2662, RFC 4706 and the current document. We can summarize 
> > that RFC2662 is relevant only for managing modems that do not 
> > support any DSL technology other than ADSL (e.g., G.992.1/2) 
> > especially if were produced prior to approval of ITU-T 
> > G.997.1 rev 3 . RFC 4706 is more appropriate for managing 
> > modems that support ADSL2 technology variants (with or 
> > without being able to support the legacy ADSL). The current 
> > document supports all ADSL, ASDSL2 and VDSL2 standards but it 
> > assumes a more sophisticated management model, which older 
> > modems (even ADSL2 ones) may not be able to support. 
> > b) It's important to notice that each modem can indicate, by 
> > the ifType, what (specific) MIB it supports. E.g., It's 
> > possible to upgrade from RFC 2662 to 4706 MIB, but it 
> > requires a software change on both sides of the management 
> > interface and also a database conversion.   
> 
> The two issues above seem to be important enough to be explicitly
> mentioned in the document. 
>  
> > c) We do not agree that "rfc2662 does support using the 
> > entity mib". All it says is that implementing the entity MIB 
> > is optional and that, if implemented, it should include a row 
> > for the the ATU-C or ATU-R based on the agent lcation. This 
> > is so obvious that it's not a surprise no other xDSL line MIB 
> > (e.g., RFC 3728 (vdsl), RFC 4319 (shdsl), RFC 4706 (adsl2)) 
> > refers to that. 
> 
> I guess that David's intent was that he expected this reference to
the
> entity MIB to show up here as well. Sometimes it is better to 
> state the
> obvious. 
> 
> > By the way, note that even RFC 2662 understands that there 
> > can be a row in the entity MIB for either the ATU-C or ATU-R, 
> > not both. This strengthen our response to Dan's previous 
> > comment T8 that referred to the entity MIB.  
> > 
> > Thanks for your efforts,
> > Moti Morgenstern
> > 
> > -----Original Message-----
> > From: Romascanu, Dan (Dan) [mailto:[e-mail]]
> > Sent: Thursday, January 22, 2009 10:42 PM
> > To: Romascanu, Dan (Dan); Moti Morgenstern; David B Harrington
> > Cc: MIB Doctors (E-mail); [e-mail]; Menachem Dodge; 
> > [e-mail]; [e-mail]
> > Subject: RE: [MIB-DOCTORS] AD review of 
> > draft-ietf-adslmib-vdsl2-06.txt
> > 
> > Hi Moti,
> > 
> > There are some more comments from David Harrington listed below. 
> > 
> > Please address them in the new version of the MIB module and 
> > of the document. 
> > 
> > 1) as per RFC4181 the TruthValue TC SHOULD be used in the 
> > SYNTAX clause of object definitions
> >     that require a Boolean type.  MIB modules SHOULD NOT use 
> > enumerated
> >     INTEGER types or define TCs that duplicate its semantics.
> > 
> > 2) the MIB organization does not follow the RECOMMENDED OID 
> > assignment scheme for notifications as per section 4.7 and 
> > Appendix D of RFC 4181
> > 
> > 3) Xdsl2LineEntry  ::=
> >    SEQUENCE {
> >       xdsl2LineCnfgTemplate            SnmpAdminString,
> >       xdsl2LineCnfgFallbackTemplate    SnmpAdminString,
> >       xdsl2LineAlarmCnfgTemplate       SnmpAdminString,
> >       xdsl2LineCmndConfPmsf            Xdsl2ConfPmsForce,
> >       xdsl2LineCmndConfLdsf            Xdsl2LineLdsf,
> >       xdsl2LineCmndConfLdsfFailReason  Xdsl2LdsfResult,
> >       xdsl2LineCmndConfBpsc            Xdsl2LineBpsc,
> >       xdsl2LineCmndConfBpscFailReason  Xdsl2BpscResult,
> >       xdsl2LineCmndConfBpscReqCount    Unsigned32,
> >       xdsl2LineCmndAutomodeColdStart   TruthValue,
> >       xdsl2LineCmndConfReset           Xdsl2LineReset,
> >       xdsl2LineStatusActTemplate       SnmpAdminString,
> >       xdsl2LineStatusXtuTransSys       Xdsl2TransmissionModeType,
> >       xdsl2LineStatusPwrMngState       Xdsl2PowerMngState,
> >       xdsl2LineStatusInitResult        Xdsl2InitResult,
> >       xdsl2LineStatusLastStateDs       Xdsl2LastTransmittedState,
> >       xdsl2LineStatusLastStateUs       Xdsl2LastTransmittedState,
> >       xdsl2LineStatusXtur              Xdsl2LineStatus,
> >       xdsl2LineStatusXtuc              Xdsl2LineStatus,
> >       xdsl2LineStatusAttainableRateDs  Unsigned32,
> >       xdsl2LineStatusAttainableRateUs  Unsigned32,
> >       xdsl2LineStatusActPsdDs          Integer32,
> >       xdsl2LineStatusActPsdUs          Integer32,
> >       xdsl2LineStatusActAtpDs          Integer32,
> >       xdsl2LineStatusActAtpUs          Integer32,
> >       xdsl2LineStatusActProfile        Xdsl2LineProfiles,
> >       xdsl2LineStatusActLimitMask      Xdsl2LineLimitMask,
> >       xdsl2LineStatusActUs0Mask        Xdsl2LineUs0Mask,
> >       xdsl2LineStatusActSnrModeDs      Xdsl2LineSnrMode,
> >       xdsl2LineStatusActSnrModeUs      Xdsl2LineSnrMode,
> >       xdsl2LineStatusElectricalLength  Unsigned32,
> >       xdsl2LineStatusTssiDs            Xdsl2Tssi,
> >       xdsl2LineStatusTssiUs            Xdsl2Tssi,
> >       xdsl2LineStatusMrefPsdDs         Xdsl2MrefPsdDs,
> >       xdsl2LineStatusMrefPsdUs         Xdsl2MrefPsdUs,
> >       xdsl2LineStatusTrellisDs         TruthValue,
> >       xdsl2LineStatusTrellisUs         TruthValue,
> >       xdsl2LineStatusActualCe          Unsigned32
> >    }
> > 
> > Check out inconsistent abbreviations, such as 
> > xdsl2LineCnfgFallbackTemplate which points to the 
> > xdsl2LineConfTemplateTable. (cnfg and conf both mean 
> > configuration) I note that RFC2662 uses "conf" consistently.
> > 
> > 4)The counter  xdsl2LineCmndConfBpscReqCount let's management 
> > station M1 know when another management station M2 requested 
> > a measurement, which caused the relevant counter to reset, 
> > such that M1's measurement is no longer valid. 
> >       "SNMP managers can use this attribute to check that the
> >        measurement results retrieved by the manager where not 
> >        interupted by another measurement request."
> > 
> > We discouraged resetting counters in IETF MIB modules. Can 
> > you explain why this is necessary? 
> > 
> > 5) The naming conventions for counters are not being followed 
> > - xdsl2LineCmndConfBpscReqCount should probably be 
> > xdsl2LineCmndConfBpscRequests.
> > 
> > 6) xdsl2LineCmndConfBpscReqCount -  why is this an Unsigned32 
> > and not a Counter32?
> > 
> > 7) I am a bit concerned about the table that actually 
> > contains the measurement - the xdsl2LineSegmentTable  The 
> > actual measurement is in xdsl2LineSegmentBitsAlloc. It 
> > appears the value is overloaded to provide subcarrier 
> > allocation status, as well as the result of a line diagnostic
test. 
> > 
> > 8) from security considerations "Unauthorized changes to 
> > xdsl2LineCmndConfBpsc could cause an unscheduled bits per 
> > subcarrier measurement to be carried out on the line.", and 
> > "Unauthorized changes to xdsl2LineCmndConfLdsf could cause an 
> > unscheduled line test to be carried out on the line." Since 
> > these two measurements report using the same object, and 
> > since one measurement can interfere with the validity of the 
> > measurement requested by a different manager, this seems to 
> > have serious issues of information reliability.
> > 
> > 9) On a larger scale, I am concerned that this appears to be a
> > **competing** standard to RFC2662. This does not complement, 
> > or update, or obsolete RFC2662.
> > 
> > I have not yet gone into the two mib modules to determine how 
> > compatible they are, but section 4 in this document makes me 
> > really uneasy as a
> > (previous) management app designer. Can a device support both 
> > mib modules simultaneously? Can a management app use both MIB 
> > modules simultaneously to manage the same ADSL functions on 
> a device?
> > 
> > I note that rfc2662 does support using the entity mib, while 
> > this mib module does not. Which makes me wonder about the 
> > compatibility between the two mib modules.
> > 
> > Thanks and Regards,
> > 
> > Dan
> > 
> > 
> > 
> > > -----Original Message-----
> > > From: [e-mail]
> > 
>