SV: RE: Hardware shaping

["Ronnie Lindgren" <ronnie@radiolan.se>]

Thanks again,

But, how do I configure my PCA200E to work with the ISP switch?

atmarp -s 192.168.100.1 0.100   qos cbr,aal5:max_pcr=10Mbps (Thanks Yon!)

The problem is that the switch throws away almost all packets from my NIC when they use shaping.  The ISP switch is set to 2.5Mbps shaping. 

If the ISP switch is set to 90 Mbps and my NIC to 166 cps (=2Mbps) shaping, it works ok. But, my ISP is not happy about that (90Mbps), because they have heavy traffic at the moment  :-(

How do I set the QoS parameters to work witch their switch? Are the switch using the same shaping as the NIC?

More info:
> Our config:
> atmarp -s 195.163.xx.yy  0.0.33 qos cbr:pcr=166cps
>
> Problem:
> In this case, our ISP's ATM-switch cuts our traffic when it is heavy loaded. Small packages can get through (5 bytes). Big packages don't.
>
> It seems that the Linux-ATM shaping is different than the switch (unknown brand). When our ISP configures their switch to 80-90 Mbits/s everything works ok. But, when shaping is set to 2.5Mbit/s on the switch (same as our shaping above?) it don't work. Why?


/Ronnie

-----Ursprungligt meddelande-----
Från: Andrew Juniper <Andrew.Juniper@madge.com>
Till: 'Ronnie Lindgren' <ronnie@radiolan.se>; Gianluca Insolvibile <g.insolvibile@cpr.it>
Kopia: linux-atm@lrc.di.epfl.ch <linux-atm@lrc.di.epfl.ch>; Linux Fore 200E list <pca200e-linux@os.inf.tu-dresden.de>
Datum: den 21 maj 1999 12:42
Ämne: RE: RE: Hardware shaping


>Ronnie Lindgren wrote:
>
>> In this case, if I understand it right (i'm a newbe),
>I'm afraid that you do and you don't understand it correctly.
>
>> the PCA200E fills up the stream with idle cells up to 155 Mbps?
>Yes.
>
>> That should explain why it don't work. 
>No. The way that the ATM physical layer works, it requires that cells are
>transmitted all the time.
>
>If the NIC has no user data to send, it sends idle (or unassigned - see
>below) cells instead.
>
>The switch automatically filters these cells and does not process them.
>In fact, the same is also true in reverse. The switch is also constantly
>sending your NIC cells, and your NIC is discarding the idle/unassigned
>cells.
>
>> Is an idle cell equal to an empty cell?
>Technically speaking, the term "empty cell" is not defined.
>An idle cell is a cell which contains no user data.
>An unassigned cell is also a cell which contains no user data.
>
>> What size is an idle cell?
>It is the same size as all ATM cells, 48 payload bytes plus 5 header bytes
>(VPI, VCI etc).
>
>> Are there any other NICs that work without transmitting idle cells when
>> shaping?
>I hope not; they wouldn't work!
>
>> Are there a way to filter these idle cells away? 
>That's what the hardware does. I'd be worried if these are being passed to
>the software.
>
>Gianluca Insolvibile wrote:
>
>>> Is an idle cell equal to an empty cell? What size is an idle cell?
>
>> An idle cell is an 'unassigned' cell, but I have no details about
>> the contents of its header (VCI/VPIs, etc.);
>
>To be 100% precise, that's wrong. Idle cells ARE different to Unassigned
>cells.
>Either can be used to fill in unused space on the physical link when there
>is no user data to send.
>
>Both cells types are detectable by the fact that they are on VPI 0, VCI 0
>with a payload type of 0 and a GFC field of 0. The only difference is that
>the idle cell has the CLP bit set, and the unassigned cell does not.
>
>See the PMC Sierra web site (http://www.PMC-Sierra.com) product
>documentation area (you'll have to register), they've got a very good
>application note called "The ATM Physical Layer", which explains this
>(section 4, page 9).
>
>AJ
>