Re: [Lksctp-developers] Extended SCTP stats, not in SNMP specs

[Jon G. <jg...@us...>]

Hi Nivedita,

Nivedita Singhvi wrote:
> Hi,
> 
> At present there arent any counters for a few things
> I think should be tracked for SCTP in the SNMP MIB specs. 
> There are at least 2 which I think are really necessary, while
> the others would all fall into the nice-to-have category.
> 
> Needed: 
> -------
> SctpInErrs (not counting checksum failures and
> 	    outofblue packets - this covers bad
> 	    formatting, incorrect skb, buffer, option
> 	    handling etc)
> SctpOutErrs (ditto)
> 
> Some nice-to-haves would be a breakout of udp-style
> and tcp-style packets, and a whole host of other
> situations, especially error situations (eg:
> timeouts, retransmissions, etc).
> 
> For now, I'd like to propose adding a linux_sctp_mib{}
> that mirrors currently the TCP extended stats (defined
> as linux_mib). I already need to add an extension for
> raw counters (There is no real need to combine tcp, 
> sctp, raw stats in one global struct unless absolutely
> necessary - this will reduce unnecessary cacheline 
> sharing between them).
> 
> Are there any objections to doing this, and any other
> suggestions? 


Please post this over to linux net (or lkml) too. I want
to do this right first time as its a royal pain to redo
our bk tree just to back out something controversial.

I definately want to avoid an negative affect on
TCP at all.

Your intent and rationale seem justifiable to me.

> 
> This is low priority, but I think it helps to put
> the infrastructure in early, and then add the
> counters as you do the development and find them 
> necessary..
> 

Yep.  I think they are Goodness(tm), too.

BTW, since the SCTP MIB is not really completely
nailed down yet.  You may want to consider if
any of your "extensions" should be in the
base.

just a thought,
jon

> thanks,
> Nivedita
> 
>