IBCA mib comments

[email protected] Thu, 15 Sep 2005 20:46:55 -0700 (PDT)
Newsgroups gmane.ietf.ipoib
Message-ID <[email protected]>
Some comments from me personally (WG Chair hat off the head sitting way
across the desk)

inline
Bill
> Hi all,
>
> I am Ali Ayoub, and I work for Mellanox technologies.
>
Welcome to the group.
> I've started implementing the IBCA MIB module, and I would like to share a
> few comments regarding the MIB definition as it appears in
> draft-ietf-ipoib-channel-adapter-mib-07.txt:
>
Great - glad to hear some of the MIBs are being implemented.

> 1. The MIB definition was written according to tables 259 & 260 (Sections
> 17.2.1, 17.2.2) of the IB spec, v1 r1.1. However, it does not include
> important info from section 11.2.1.2 such as: vendor specific information,
> vendor ID, vendor part ID, HW/FW version. This information is important
> (from network administrator's point of view).
>
I would assume such things would belong in the enterprise mib for the
particular vendor.

However if there are specific GENERIC things that belong in this MIB -
write a object up, submit it here and after we look it over Sean can add
it into an updated draft

> 2. The MIB is almost entirely static in the sense that all the info
> supplied
> is determined at driver loading, and does not get changed until the next
> reboot. An exception to this is "ibCaInterpacketDelayValue" which is
> dynamic
> according to the current network topology. I suggest to remove this
> attribute from the mib to keep it all static (this affects the
> implementation data structure too and saves unnecessary queries).
> Dynamic info (such as port state) should be moved to the SMA MIB, and the
> IBCA MIB definition should become completely static.
>
Hate to say it... but this seems like an implementation issue.  But will
take this under advisement... any other implementors out there feel the
same way ?

> 3. What is the ibCaSupportsSubnetManager object? I couldn't find it in
> table
> 260 as mentioned there. It is by any chance the ibCa*Is*SubnetManager?
> Then
> the right place for this object would also be in the SMA MIB.
>
Sean ?

I need to go look at this one closer - but Sean, what was your intention