Re: [16NG] WG last call on draft-ietf-16ng-ipv4-over-802-dot-16-ipcs-02 + RE: regarding my comments on MTUs

[Burcak Beser <[e-mail]>]

As far as I can see we tree options:

1. Get WiMAX Forum to handle 1500 byte packets. I am not sure how this may be achieved other than traveling back in time to impact NWG v1.

2. Change this draft to 'personal'-ipv4-over-wimax-ipcs and use 1400 byte limitation from WiMAX Forum NWG v1.

3. Write the draft for IPCS  assuming L2 tunneling and no MTU limitations; and put the 1400 byte case on the appendix stating that WiMAX is expected to be the widest 802.16 deployment and NWG v1 has the 1400 byte limitation as a special case, and offer the remedies and state the problems.

Personally I like the option 3 which is cleaner and straight forward.

-burcak
________________________________________
From: Samita Chakrabarti [[e-mail]]
Sent: Friday, March 21, 2008 4:07 PM
To: 'Wesley George'; Burcak Beser
Cc: 'gabriel montenegro'; 'Jari Arkko'; 'Samita Chakrabarti'; [e-mail]
Subject: RE: [16NG] WG last call on draft-ietf-16ng-ipv4-over-802-dot-16-ipcs-02 + RE: [16NG] regarding my comments on MTUs

Hi Wes,

Please see in-line.

>
>Samita, your proposed update merely adds more words to say the
>same thing that the current draft says - host MTU should be 1400. It's
>helpful in that it adds more justification to your argument, but I still
>disagree
>with the underlying conclusion.
>
>My recommendation was to treat MTU as something that must be taken into
>consideration at the design phase of a WiMax network and the
>initial configuration of the devices, not a hard and fast limit based on
>the limitations and assumptions of ONE case out of many, especially since
>the
>806.16 over Ethernet draft completely contradicts your recommendation. It
>views MTU as non-negotiable (must be at least 1500), and puts the
>responsibility on the network operator to ensure that they've build the
>network to support the right underlying MTU or deal with the
>fragmentation that will be inevitable.
>I don't see why we wouldn't be doing the same thing here.
[SC>]
       As far as I know for IPCS network WiMAX forum has decided that their
default link MTU would be 1400 bytes. Checkout RFC 5121 (appendix D) which
chose the same value 1400 bytes for that reason. But one difference from
IPv6 case is that IPv6 specification requires to send RA with MTU option if
the MTU supported by the link is not the default value. Thus by following
IETF specification WiMAX AR can send MTU option with 1400 value to the
clients and the clients can adjust the MTU accordingly; while RFC 5121 can
specify 1500 bytes as the default MTU.

Now IPV4CS would be happy to set the same default 1500 bytes following
RFC5121 example - the only major issue is that there is no IPv4 mandatory
requirement from the server to communicate (through RA or DHCP option etc.)
the non-default link MTU unless the client itself asks for it.

The problem that Jari and others are pointing to is that if we specify 1500
bytes and the clients do not know that they are in WiMAX link and start
sending packets over 1400 bytes, the packets will not be delivered at all.
Those clients will not operate in the WiMAX network.

Please note that WiMAX forum can change their default MTU value in the next
revision, but the current specifications (Rls 1.1) cannot change.

This is the data behind the decision on 1400 bytes as initial value; it
matches the criteria that was given by Jari on the list. This IPv4CS
recommends the new clients to request MTU option from DHCP requests for
efficiency.

If you or Burcak can work with WiMAX forum and produce a note from the chair
that 1500byte  IPv4CS payload is OK for WiMAX network for the current NWG
releases in the next one or two weeks, we can really appreciate that and
change to 1500 bytes. Currently, I do not know of any other deployment of
IPv4CS other than WMF.



>
>Also, I do not understand how the statement
>"The WiMAX forum [WMF] has specified the Max SDU size as 1522 octets."
>aligns with the current statement in your draft
>"The Length parameter of IEEE 802.16 MAC frame has a size of 11 bits.
>Hence the total PDU size is 2048 bytes.  The IPv4 payload can be a
>maximum value of 2038 bytes ( Total PDU size (2048) - (MAC Header (6) +
>CRC (4)), which is the maximum possible MTU."
>
>Where are those other ~500 octets going? Forgive my ignorance, but is
>there some layer between SDU and PDU that has that sort of overhead, or is
>this another example of arbitrary limits that conflict with each other?
[SC>]
      I do not know why WMF has specified 1522 bytes as MAX SDU and not
beyond that. You need to ask that question to the WiMAX forum. For
reference, please check RFC 5121 Appendix D.


Thanks,
-Samita





>
>On Thu, 20 Mar 2008, Burcak Beser wrote:
>
>> I have to admit that I fail to understand the reason for the MTU
>limitation in the first place. The draft talks about Layer 2 tunnels
>between BS and AR, and than GRE is being given as an example which causes
>the problems in the first place. Instead of GRE, if 802.1Q VLANs are to be
>used this artificial limitation will not be taken into considereation.
>>
>> Is it possible to clarify the sentence "Hence if a IPv4CS MS is
>configured for 1500 bytes it will have to be communicated by the access
>router(AR) about the default link MTU (1400 bytes) in WiMAX network."? Due
>to fragmentation support in the 802.16 transport aribitrary PDU sizes are
>supported even when the maximum SDU size is limited. In the light of
>fragmentation the "The WiMAX forum [WMF] has specified the Max SDU size as
>1522 octets." only means that a 1600 byte PDU will be transmitted in at
>least two chunks. The 802.16 transport will carry almost anything that the
>underlying native protocol can.
>>
>> The normative 'should' in the sentence "Therefore, IEEE 802.16 frame
>SHOULD support for 1500 byte IP payload..." is unnecessary.
>>
>> The use of normative 'should' in ".. IPv4 and IPV4CS clients SHOULD
>implement DHCP interface MTU option..." sentence looks like conditionally
>mandating the use of DHCP as well. A clear requirement will be better. The
>inclusion of IPv4 clients is first not correct since ARP would work, and
>second is out of scope for this draft by definition.
>>
>> I understand that IPv4CS cannot transmit ARP messages but "An IPv4CS
>client is not capable of doing ARP probing either to find out the link
>MTU." looks like stating that "IPv4CS client MUST NOT use ARP messages". At
>the same time IPv4CS client has to be defined.
>>
>> "Consequently, the clients are encouraged to run PMTU[RFC 1191] algorithm
>or PPMTUD[RFC 4821] for appropriating the bandwidth usage over the route-
>path or end-to-end transport layers. However, PMTU mechanism has inherent
>problems of packet loss due to ICMP messages not reaching the sender and
>IPv4 routers not fragmenting the packets due to DF bit being set in the IP
>packet." Can we state this sentence in normative manner?
>>
>> I have to admit that the draft struggles around the artificial MTU
>limitation which is not justified, and omits other important subjects.
>>
>> -burcak
>>
>> -----Original Message-----
>> From: [e-mail] [mailto:[e-mail]] On Behalf Of
>Samita Chakrabarti
>> Sent: Thursday, March 20, 2008 7:18 PM
>> To: 'Wesley George'; 'gabriel montenegro'; 'Jari Arkko'
>> Cc: 'Samita Chakrabarti'; [e-mail]
>> Subject: Re: [16NG] regarding my comments on MTUs
>>
>> Hi Jari and Gabriel,
>> Thanks for the guidelines on re-wording the text on MTU determination for
>IP4CS link. Thanks to Wes for your proposed text as well; I have taken some
>ideas from your text.
>>
>>
>> Here is the proposed text on MTU based on the wg discussion and the AD
>> guideline:
>> ------------------------------------
>>
>> 5. Maximum Transmission Unit considerations for IPv4CS link
>>
>> [RFC894] specifies 1500 bytes as a maximum length of IPv4 over Ethernet
>and encourages to support full-length packets. Therefore, IEEE 802.16 frame
>SHOULD support for 1500 byte IP payload in the IPv4CS link by supporting
>larger than 1500 bytes link MTU between the MS and the Access router (AR).
>> The extra overhead that IEEE 802.16 frame should support in this case,
>depends on the tunneling protocols or any other virtual or pseudo-wiring
>protocols packet overhead.
>>
>>   The current architecture of IPv4CS in IEEE 802.16 networks is defined
>in
>>   the WiMAX (Worldwide Interoperability for Microwave Access) forum
[WMF].
>>   The addressing and operation of IPv4CS described in this document are
>>   applicable to the WiMAX networks as well. The WiMAX forum [WMF] has
>>   specified the Max SDU size as 1522 octets. However, it is recommended
>>   that IP-payload in WiMAX architecture [WMF] specified network is 1400
>>   bytes due to overhead of GRE tunnel and possible
>>   additional over-head of different types of tunneling (such as IPSec/UDP
>>   tunneling) between the base station (BS) and Access Router.
>>
>>   Hence  if a IPv4CS MS is configured for 1500 bytes it will have to be
>>   communicated by the access router(AR) about the default link MTU (1400
>>   bytes) in WiMAX network.
>>   However, currently in IPv4 client architecture a node is not required
>>   to ask for MTU option in its DHCP messages nor the
>>   IPv4 router-advertisements
>>   are required to advertise link MTU option when the link does not
>support
>>   1500 byte de-facto MTU size. An IPv4CS client is not capable of doing
>ARP
>>
>>   probing either to find out the link MTU.
>>   Thus current specifications of WiMAX network access routers cannot
>>   communicate its link MTU to the IPV4CS MS. On the other hand, it is
>>   imperative for an MS to know the link MTU size if it is not the default
>>   MTU value for de-facto standard in order to
>>   successfully send packets in the network towards the first hop.
>>   This document can not also assume that the legacy IPv4 client
>>   implementation with IEEE 802.16 layer 2 support,
>>   would be able to dynamically sense IPv4CS WiMAX network and adjust
>>   their MTU size  accordingly.
>>
>>   Thus for IPv4CS over IEEE 802.16 the default MTU size is 1400 bytes.
>>
>>   This document recommends that all future implementations of IPv4 and
>>   IPV4CS clients SHOULD implement DHCP interface MTU option [RFC2132] in
>>   order to configure its interface MTU according to the access network in
>>   order to maximize the capacity of the bandwidth of
>>   the network. Consequently, the clients are encouraged to run
>>   PMTU[RFC 1191] algorithm or PPMTUD[RFC 4821] for
>>   appropriating the bandwidth usage over the route-path or end-to-end
>>   transport layers. However, PMTU mechanism has inherent problems of
>packet
>>
>>   loss due to ICMP messages not reaching the
>>   sender and IPv4 routers not fragmenting the packets due to DF bit being
>>   set in the IP packet. The above mentioned path MTU mechanisms will take
>>   care of the MTU size between the MS and its correspondent node across
>>   different flavors of convergence layers in the WiMAX networks,
>>   IEEE 802.16 networks and other types of networks such as Wi-Fi,
>>   Ethernet or 3G networks.
>>
>> --------------------------
>>
>> Review comments are welcome.
>>
>> Regards,
>> -Samita
>>
>>> -----Original Message-----
>>> From: [e-mail] [mailto:[e-mail]] On Behalf Of
>>> Wesley George
>>> Sent: Thursday, March 20, 2008 8:24 AM
>>> To: gabriel montenegro
>>> Cc: Samita Chakrabarti; [e-mail]
>>> Subject: Re: [16NG] regarding my comments on MTUs
>>>
>>> I have been thinking about the discussions regarding MTU, and would
>>> like to propose some alternate language for section 5 of the ipcs
>>> draft. I will caveat this by saying that it is my first attempt at an
>>> ID (or a part of
>>> one) but I thought that it might be helpful to propose an alternative
>>> rather than simply saying that the language that is in the -02 draft
>>> currently in Last Call that IMO is not adequate, due to its conflict
>>> with the .16 over ethernet draft.
>>>
>>> ** proposed draft language to replace the first paragraph of section 5
>>> **
>>>
>>> When considering end host MTU, there are two points at which MTU must
>>> be considered. First, the end to end path MTU between one IP host and
>>> another. In 802.16, this is expected to follow existing convention [rfc
>>> citation needed?] for fragmentation and path MTU discovery between end
>>> hosts, and will not be treated here.
>>>
>>> The point at which MTU must be considered and will be treated by this
>>> document is the path between the Base Station (BS) and the ASN Gateway
>>> or HA, where traffic is usually encapsulated using one of several
>>> methods before being allowed to transit the remaining path as native IP
>>> traffic.
>>> On this portion of the path, there are two cases which must be
>considered:
>>>
>>> 1) Where the underlying MTU of the transport mechanism between the BS
>>> and the ASN GW or HA is configurable to a value > 1600 bytes or is
>>> known to support the same
>>> 2) Where the underlying MTU of the transport mechanism between the BS
>>> and the ASN GW or HA is not a known value, or is not able to be
>>> configured to MTU > 1600 bytes
>>>
>>> In case 1, the end host MTU SHOULD be set to at least 1500 bytes, but
>>> can be set to any higher value supported by the underlying network up
>>> to the maximum supported size on 802.16 of 2038 bytes.
>>>
>>> With an end host MTU of 1500 bytes or more, the underlying network MUST
>>> support an MTU of at least 1500 bytes plus the overhead of the chosen
>>> encapsulation method between the BS and the ASN GW or HA. For ease of
>>> standardization, the underlying network SHOULD support an MTU of at
>>> least 100 bytes larger than the chosen host MTU value, as this is large
>>> enough to ensure that there is no dropping or fragmentation of oversize
>>> packets regardless of the encapsulation method.
>>>
>>> In case 2, the end host MTU SHOULD be set to 1400 bytes to ensure that
>>> there is no dropping or fragmentation of oversize packets regardless of
>>> the encapsulation method. If the end host MTU is set higher, or the MTU
>>> of the underlying transport network is lower than 1500 bytes, the
>>> network MUST be able to accept any fragmentation of packets larger than
>>> the supported transport MTU, or MUST employ a method of MTU discovery
>>> to prevent oversize packets from being generated by the end host.
>>>
>>> In order to preserve the defacto standard of MTU 1500 bytes for end
>>> host use, case 1 SHOULD be followed if at all possible. If adhering to
>>> case 1 is not possible, case 2 is presented as an alternative.
>>>
>>> ** end proposed draft language **
>>>
>>> The second paragraph of section 5 can remain as-is.
>>>
>>> Some additional comments:
>>> It may be a good idea to add in language from the background slides
>>> that were presented in Philadelphia which identify why 100 bytes is the
>>> recommended padding between host MTU andNetwork MTU for further
>>> clarification.
>>>
>>> I concede that it's not really feasible to rely on an MTU discovery
>>> method to compensate for MTU on the transport network that is not
>>> capable of supporting the full 1500 bytes of host MTU. However, I think
>>> that the proposed language provides ways to work this based on whether
>>> that is a known value or not (via the two different cases) and
>>> addresses that concern. I think that this is much closer in idea with
>>> the way that the EthernetCS draft is written, treating 1500 as the
>>> defacto standard that we should prefer, but covering what the reasoning
>>> is for MTU settings and identifying ways to compensate for limitations
>>> in the underlying transport network, rather than simply assuming that
>>> those limitations exist and catering to the lowest common denominator.
>>>
>>> Thanks,
>>> Wes
>>>
>>> _________________________________________
>>>   Wesley George
>>>       Sprint IP Engineering
>>>   703-689-7505 (O)  703-864-4902 (PCS)
>>>         http://www.sprint.net
>>> _________________________________________
>>>
>>>
>>> On Tue, 18 Mar 2008, gabriel montenegro wrote:
>>>
>>>> For the IPv4 case, I have no problem with the MTU staying at 1400
>>>> bytes
>>> per the current draft.
>>>> There is no sure-fire way to indicate to the MS what the MTU is. The
>>> draft currently says:
>>>>
>>>>   The MTU value of the MS can be configured via Path MTU Discovery [6],
>>>>   Packetization layer path MTU discovery [7], DHCP MTU option [8] or
>>>>   static configuration of each MS.As pointed out by Jari, what we're
>>> talking about is the *interface* MTU, not the *path* MTU
>>>> (minimum hop MTU of all the hops in a path). Accordingly, when
>>>> applying
>>> [6] or [7]
>>>> to discover the interface MTU, they must be used between the MS and
>>>> its
>>> immediate IP hop,
>>>> the access router. Of course, in order to use [6] or [7], some
>>>> interface
>>> MTU must be assumed
>>>> to begin the discovery process.
>>>>
>>>> 1400 is a very reasonable value for this initial default MTU. Given
>>>> the
>>> well-known issues with PMTU,
>>>> and the scant support for the DHCP Interface MTU option, 1400 will
>>>> work
>>> well in the absence of
>>>> any further MTU discovery or configuration. It will also avoid
>>> fragmentation in what is expected
>>>> to be the most common deployment for 802.16: WiMAX networks. 1500
>>>> bytes,
>>> the other MTU value
>>>> folks have proposed for IPv4 CS does not have this virtue and will
>>>> incur
>>> fragmentation.
>>>>
>>>> Notice that for DHCP we should really say "Interface MTU option"
>>>> to clarify that we're talking about option 26 (Section 5.1, RFC2132).
>>>>
>>>> Going forward we may have a more reliable and less time consuming
>>> mechanism to
>>>> derive the interface IP MTU by learning the layer 2 maximum SDU via a
>>>> new
>>> TLV in the SBC exchange
>>>> in 802.16. This is being proposed  this week at the Orlando IEEE
>meeting.
>>>>
>>>> As for the ethernet CS MTU, as Max said during his presentation last
>> week,
>>> there is no choice. We must
>>>> preserve the ethernet MTU as expected from "Ethernet" CS, regardless
>>>> of
>>> the fact that 1500 byte IP packets will incur
>>>> fragmentation in common WiMAX networks.
>>>>
>>>>
>>>> -gabriel
>>>>
>>>>
>>>> ----- Original Message ----
>>>> From: Jari Arkko <[e-mail]>
>>>> To: [e-mail]
>>>> Sent: Thursday, March 13, 2008 1:18:55 PM
>>>> Subject: [16NG] regarding my comments on MTUs
>>>>
>>>>
>>>> As a clarification of my comments during the WG meeting I would like
>>>> to state the following:
>>>>
>>>>> From AD perspective the WG is free to decide whatever MTU value that
>>>>> you
>>>> wish to have, as long as you are both aware of and document the
>>>> implications of this choice. Some of the implications include:
>>>>
>>>> - difficulty of manual configuration
>>>> - ability to raise/lower the values automatically
>>>> - capability of 802.16 backhaul networks to handle larger MTUs (ipcs,
>>>> ethcs, wimax and non-wimax)
>>>> - ability or inability to determine what network you are in
>>>> - inability of L2 devices to send ICMP messages
>>>>
>>>> Hope this clarifies,
>>>>
>>>> Jari
>>>>
>>>>
>>>> _______________________________________________
>>>> 16NG mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/16ng
>>>>
>>>>
>>>>
>>>>
>>> _______________________________________________
>>> 16NG mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/16ng
>>
>>
>>
>> _______________________________________________
>> 16NG mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/16ng
>>
>>
>>




_______________________________________________
16NG mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/16ng