[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Priority rules for allow statements. (fwd)



------- Forwarded Message

Date:    Thu, 07 Nov 2002 09:06:21 -0700
From:    Russ Rew <address@hidden>
To:      "Stonie R. Cooper" <address@hidden>
cc:      steve, Steve Chiswell <address@hidden>,
         "Cannon, Declan" <address@hidden>,
         "Wells, Tim" <address@hidden>
Subject: Re: Priority rules for allow statements. 

>To: address@hidden
>From:"Stonie R. Cooper" <address@hidden> 
>Subject: Re: 20021107: Priority rules for allow statements.
>Organization: Planetary Data, Incorporated
>Keywords: 200211070342.gA73gaX09881

Hi Stonie,

> I've got a somewhat unusual question for you.  Is there anyway to
> establish a "priority" of listed "allow" statements in the
> ldmd.conf.  Here's where I'm going, and you let me know if it's
> already in LDM, or would need to be added.

It's not already in the LDM, so would need to be added.

> For argument's sake, let's say NOAA is going to decide to establish
> a internet distribution system to compliment NEX-NOAAPort.  The
> distribution system could be very similar to IDD . . . with NOAA
> establishing main nodes throughout the digital world, that basically
> anyone could do a "request" to.
> 
> The top node, obviously, would be at the Telecommunications Gateway
> and NESDIS; the principle of priorities would be as such that a
> short list of the first tier of nodes would have priority over all
> other requests.  As an example, say the first tier nodes would be
> the regional offices - Southern, Western, Central, and Eastern.  The
> second tier would be Alaska and Hawaii, and all the specialized
> services - SPC, FSL, NCEP, HPC, NCDC.
> 
> The third tier, feeding from either first tier or second tier, would
> be the individual WFOs, military, FAA, etc.
> 
> All would have redundancy built in - like the IDD failover.
> 
> My question revolves around the idea that - say - IBM decides to use
> HPC as their request node.  But HPC also is the primary for the FAA.
> Is there a way that HPC could configure their LDM such to give the
> FAA priority over IBM for servicability?

I'm not sure what you mean by "priority" here.  Possibilities include
priority in use of limited out-bound bandwidth, CPU priority of the
sender processes, or priority in getting access to the shared product
queue, for example.  I'll assume bandwidth priority is what you have
in mind, because that's probably the most limited resource in this
scenario.  

I think it would be difficult to modify the LDM to give some out-bound
connections priority over others in use of out-bound bandwidth, at
least on the commodity Internet, because I don't think the RPC
interfaces used by the LDM support this functionality.  If and when we
convert to IPv6 (which is available for Internet2 connections), there
are mechanisms for specifying priorities (QoS) that might be of use,
but I'm skeptical about this.

Steve Emmerson is the developer for LDM5, so I'm CC:ing him on this
reply, in case he has other comments about this.  It sounds like an
interesting idea, but I think giving priority to certain feeds or
connections might be easier to do with multiple product queues.  A low
tech way of accomplishing this might be running running LDM servers on
multiple hosts at a site, with lower priority servers having a
throttled-down connection to the network ...

--Russ


------- End of Forwarded Message