most important topics

- chapter in developers guild about signals and statistics, naming convention, what constitues a signal and what does not, etc.

- existing protocols

- 802.11 statistics, logging, display strings, code cleanup, memory management

- complete IPv6

- cross-cutting smaller changes

- review end-to-end delay calculation

- for applications it means the delay of complete application packets

- for UDP it means the delay of datagrams

- for TCP it means the delay of bytes

- review NED documentation, update obsolete parts

- add meaningful display strings to at least the top level submodules of network nodes (with optional formatting parameters)

- finish renaming signals, labels, statistics to have a consistent naming scheme

- review all examples to check if they still work as expected

- new features

- add geographic mobility modules (separate Euclidean and geographic mobilities)

- add new queue module API and use it in all protocols which have queues

- LTE PHY full support and porting to INET 4.0, source/sink separation, signal classes, etc. (e.g. probe/microwave oven)

- port existing 3rd party models

- help ANSA porting to INET 4.0

=====================

- implicit serialization of chunks for conversion must be controllable from the API with flags

- the packet dissector may need to do implicit chunk conversion because it has to choose a representation which may or may not be the same as the one in the packet (e.g. two interoperable frameworks)

- add parameter to mobility base which determines what should happen if the energy storage is depleted. For example, a drone should just stop, but a satellite should continue its motion.

- route source type should be a protocol, CISCO admin dist should be calculated from the source type

- all protocols must set their protocol as source to routes added

- long term: extract OspfRouteData from OspfRoutingTableEntry and add as protocolData into Ipv4Route

- long term: extract OspfInterfaceData from OspfInterface and add to InterfaceEntry as tag?

- OspfConfigReader: search = "strict|relaxed" should be changed :)

- OspfConfigReader: collects all OSPF interfaces based on XML config entries, result may contain the same interface twice, should work the other way around by starting with all the interfaces and finding the first OSPF config entry in the XML for that interface

- OspfArea::originateRouterLSA calculates a different LSA length field than Lsa.cc:B calculateLSASize(const OspfRouterLsa *routerLSA)

- change all protocols (where applicable) to accept any kind of cPacket (not only Packet) for interoperability with 3rd party projects

- add test and check for std::dynamic_pointer_cast<Subclass1>(chunk->peek<Subclass1>()) != nullptr always holds when peek uses a variant type deserializer

- remove obsolete HostAutoConfigurator, Ipv4FlatNetworkConfigurator modules?

- fix mobility non-determinism caused by interpolation from last position during animation

- validate Rotation, Quaternion, EulerAngles against each other and documentation

- validate positioning and orientation IMobility, PhysicalEnvironment, ObstacleLoss

- replace all internal queues with external queue modules and make them configurable

- 802.11 preamble, header, data separation in HT and VHT mode doesn't really make sense the way it's right now implemented

there's no separate header in these modes, the preamble is much more complicated than in the other modes

there's also a bug related to this in the Ieee80211ScalarTransmitter when it computes the data duration

- replace long2string and double2string in Units.msg with something more clever (e.g. 160000000 Hz -> 160 MHz)

- add L2 sockets: EthernetSocket bound to EtherType, LlcSocket bound to SSAP, DSAP, add Ieee80211Socket bound unconditionally

- solve dynamic node creation and address resolution in applications

- don't subclass APSK medium and such with other specific radios, think of APSK radio as yet another specific conceptual radio

+ generalize shutdown, startup, crash operations to applications, protocols, interfaces, etc. (NodeShutdownOperation -> ShutdownOperation, NodeStatus -> OperationalStatus?)

+ add antenna directional selectivity visualizer

- move extra functions that doesn't really belong to L3AddressResolver (e.g. findInterfaceTableOf, findIPv4RoutingTableOf)

- revise protocol base class hierarchy (e.g. LayeredProtocolBase vs IPv4 base class, etc.)

- revise protocol header class hierarchy (e.g. avoid unnecessary interfaces)

- revise NED network node module hierarchy (e.g. MPLS nodes)

- use m instead of double in Coord?

- replace functionality of GenericAppMsg with chunk tags, or better yet get rid of it altogether!

- use shared pointers for transmissions/receptions, etc. because the gui copies packets in the packet log and inspects them later

- factor out physical layer analog domain representation into a separate parameter of transmissions

- ethernet socket vs. ssap/dsap both in StandardHost and EtherHost

- review all examples and check if they still do what is intended

- add ChunkStack (lifo) similarly to how ChunkQueue (fifo) is implemented

- add separate reusable application packet generators

- rewrite serializers that use structure bit fields, this is undefined C++ behavior

- revise all network node module base class hierarchy (e.g. mpls nodes)

- replace cCompoundModule class in hosts/nodes with a specific classes such as StandardHost

providing getInterfaceTable, getMobility, getIPv4NetworkLayer, etc. to avoid random code fragments with getSubmodule(".ipv4.routingTable")???

but we should avoid having these classes for a random collections of utility functions!

- add separate classes/functions for iterating over the network nodes in a network or

finding a network node based on its mac address/ip address or finding the position of a network node, etc.

- add missing node NED files for existing protocols (e.g. sensor nodes, IoT devices)

- update all NED and C++ documentation, remove obsolete documentation

- nicely position submodules in compound modules, prepare for 2 info lines

- extend main protocol modules (and all submodules) with 2 info lines above the submodule icon

- avoid specifying module icon sizes by default, override icon size in submodules

- sort includes alphabetically

- pretty print white spaces

TODO: cleanup MSG files

remove if possible:

- import inet.common.INETDefs-ek

- INETDefs.msg

- cplusplus-ok

- #include-ok

- const-ansok

- typedef-ek

- @opaque

- @primitive-ek -> primitive foo;

@descriptor(false) vs @descriptor(readonly); why do we have descriptor anything at all? why don't we have descriptors for all types all the time?

@editable(false) vs @descriptor(readonly)??? and in general editable is a GUI concept

Chunk.msg:

_getChunk -> getChunk

_packet -> packet

DYMOSequenceNumber: class DYMOSequenceNumber??? why class? primitive class???

PacketDrop: packet_ -> packet

IntServ.msg: *_t is reserved for C++ standard

InterfaceEntry: eliminate @getter if possible

EtherType.msg: replace with Protocol.h?

IPProtocolId.msg: replace with Protocol.h?

TODO: review and refactor all applications which have lifecycle, startTime, endTime parameters, move code from initialize to separate reusable functions (e.g. for lifecycle restart)

TODO: DHCP Client should work multiple interfaces, but works with only one.

TODO: UDP: add a parameter: used protocol number, default value is UDP.

It's useful for UDP-like protocols, e.g. MANET

TODO: revise ethernet/eth-index.ned! e.g it writes about autoconfiguration, and probably half of it is obsolete

TODO: examples/ospfv2/areas does not work: the UDP apps in H3 keep sending, but never receive answer; seems like

sent packets cannot even leave H3 (they are queued up in ARP, due to "address resolution pending")

TODO: TCP,UDP,SCTP: Need revision for PORT_ANY usage:

What will the PORT_ANY value: -1 or 0.

Where do replace from PORT_ANY to a valid port: in TCPSocket or in

TCPConnection::process_OPEN_ACTIVE/process_OPEN_PASSIVE

Check it in other protocols.

TODO: Int128/Uint128 code revision / optimisation

Optimise *= and other operators/functions.

TODO: IPv4 misfeature: in IPv4::handleMessageFromHL(cPacket *msg), if the packet is already

an IP datagram, then it does not encapsulate the packet but just routes it!

Problem: HL cannot control whether it wants IP to route the packet or tunnel (encapsulate) it!!

TODO: fix typos: recived, broadCast

TODO: TCPSegment, UDPPacket: use uint16_t for ports, uint32_t for sequence numbers, etc; also in the code!!!

hint: first replace it somewhere with a type that CANNOT be cast to integer types, then let compile errors propagate the type change everywhere

TODO: StandardHost: do we need tcpType, udpType, sctpType? why not just <default("UDP")>, <default("SCTP")>, etc? they are conditional anyway!

TODO: make copyright headers uniform....

TODO: set up nightly builds:

- should report build errors/warnings;

- should run etc/runallexamples and report the results (runtime errors, fingerprint mismatches, etc)

TODO: execute etc/runallexamples, and fix errors (last run: May 2008)

TODO: implement missing models:

Qeueing (we have: FIFO, RED): CBQ, WFQ, RIO, Round Robin, Fair queueing Round Robin

Suggested by VR:

Wired: Circuit Switch, ATM(AAL2/AAL5,CBR/VBR/ABR), IP over ATM, PNNI, MPLS, DiffServ

Wireless: Adhoc routing:FSR/DSDV/ZRP/TORA/LENMMAR;Adhoc MAC:802.11, WTRP,

TDMA; QoS supports for cross-layer design

and also, Group Mobility, etc.

TODO: derive BGPRouter from OSPFRouter that should be derived from Router

TODO: is the namId needed at StandardHost level? Possibly implement namtrace as a plugin?

TODO: port NAM trace writer to use signals mechanism

TODO: Add ThruputMetering channel (NED definition)

TODO: provide default for RoutingTable.routingTableFile once xmlinline() is implemented in omnet.

TODO: integrate models:

HIPSim++, MobileWiMAX, ...

TODO: PPP, Ethernet: should use signals (ModelChangeNotification) to detect if channel path breaks

TODO: implement node failure/recovery by using signals mechanism

TODO: pcap file recording using signals mechanism

TODO: VideoStrmReq problem on IPv6:

"Error in module (UDP) P62.server.udp (id=21): User error: (UDPPacket)VideoStrmReq arrived from lower layer without control info"

TODO: check if SCTP in StandardHost6 (IPv6) is working properly

BUG: if SCTPServer.echoFactor not 0, then will a runtime error:

"Error in module (SCTP) P5.server.sctp (id=22): User error: stream with id 2 not found."

Network: IPv4; 1 server, 1 router, 10 client

TODO: labelling:

- add labels l2pkt/l3pkt/l4pkt to upper layer input/output of L1/L2/L3 protocols or modules?

i.e. NIC's upperLayerIn could be tagged as l3pkt, and IP's ifOut also as l3pkt

- add @labels(node,configurator,!configurator) to FlatNetworkConfigurator and similar "singleton" modeules,

to make them unique?

- change the @labels(PPPFrame) label on pppg[] gates of hosts/routers to something slightly different

(e.g. @labels(PPPFrame/pk)) so that PPPInterface does not appear in the palette on network level?

TODO: finish making gate names consistent! around the SCTP and RTP modules

TODO: PingApp is strange; it should send payload+controlInfo instead of ICMP packet,

the word ping should not occur in ICMP code, move identifier, sequence number from PingPayload into ICMPControlInfo

TODO: eliminate ARP overhead for PPP interfaces;

TODO: UDPSocket::Callback::socketDatagramArrived() - change arg cMessage* to cPacket*

TODO: examples\wireless\hosttohost: Readme mentions some "attached Excel sheet" which is missing!!!

TODO: toplevel package legyen "org.omnetpp.inet"

TODO: MTU patch from Irene Rungeler

TODO: add SCTP to the whatsnew file

- MAKE IT POSSIBLE TO USE DIFFERENT TYPES FOR EACH ELEMENT OF "LIKE" SUBMODULE VECTORS!

In StandardHost etc, change

udpApp[numUdpApps]: <udpAppType> like IUDPApp;

to

udpApp[numUdpApps]: <> like IUDPApp;

TODO:

- failure/recovery (replacing with failed router is problematic!)

- gate labelling for the NED editor

-----------------------

add support for integrating the TIREM module?

http://www.alionscience.com/index.cfm?fuseaction=products.view&productid=19

"TIREM predicts radio frequency propagation loss over irregular terrain and seawater for

ground-based and air-borne transmitters and receivers. TIREM is commercially

available as a DLL and as a module in leading commercial M&S tools."

-----------------------

mailing list post on 2008-11-17 "INET Framework accuracy":

A really interesting paper is the technical report "Towards Comparable

Network Simulators", from Pengfei Di, Yaser Houri, Kendy Kutzner and Thomas

Fuhrmann [see url below]. They measure 802.11 throughput with ns2 and

OMNeT++ INET Fw, and come to the conclusion that they are hard to compare,

due to differences in the radio model and MAC parameterization. Then they go

on to wrap ns2's 802.11 model into an OMNeT++ simple module, and with that,

the OMNeT++ simulation (not too surprisingly) shows excellent correspondence

with the ns2 simulation :)

http://i30www.ira.uka.de/research/documents/p2p/2008/towards_comparable_netw

ork_simulations.pdf

The paper is excellent work, and literally cries for follow-ups:

1. INET's radio model is pluggable, and as the report authors describe it,

the ns-2 radio model seems to be quite simple (i.e. propdelay is a constant

and does not depend on distance). Thus, it should be possible to reimplement

ns-2's radio model in OMNeT++ with not too much effort (and contribute it to

INET) and re-run the experiments with it.

2. OTOH I'm wondering, why not test the radio model and the 802.11MAC

operation *separately*? I.e. write an ideal radio (ber==0 always) and test

different 802.11 MAC models over that in all scenarios; then have more

realistic radio models and test them with simple (contention-free) 802.11

setups

3. INET 802.11MAC model parameterization could be adjusted so that it's

possible to set it up with *exactly* the same parameters as ns-2.

4. the authors ported ns-2's 802.11 model into OMNeT++; wonder if this code

is publicly available, or is there any plan to release it? there certainly

would be value in doing that.

-----------------------

TODO: eliminate cast after dup(), like (AirFrame*)airframe->dup()

TODO: IRoutingTable6, ITED, etc!

TODO routingTable optimizations from Gamer!

TODO: Document: decide interfaceId vs interfacePointer! issue: IPRoutingDecision contains interfaceID.

if we change it to pointer, messages that are underway in the host during a deleteInterface()

may crash!!! solution: use a vector inside InterfaceTable, and let each deleteInterface()

leave a hole in it?

-----------------------

- move "q=queue" too into the default display string

- add gate @labels

- rename TCPBasicClientApp to TCPRequestReplyClientApp ?

- ethernet: ditch autoconfig and zero-datarate channels, use deliverOnReceptionStart and 10/100/1000Mbps channels!

-----------------------

TODO rename the following TED.h methods to begin with a verb:

unsigned int linkIndex(IPAddress advrouter, IPAddress linkid)

unsigned int linkIndex(IPAddress localInf)

IPAddress peerRemoteInterface(IPAddress peerIP)

IPAddress primaryAddress(IPAddress localInf)

-----------------------

cache gate IDs to speed up sending

-----------------------

put all c++ code into "inet" namespace

- networklayer\extras\FailureManager.cc: contains obsolete, hardcoded module type names!

- factor out common part of hosts/routers etc, and use subclassing

- remove channelInstaller

- examples: add fingerprints!

- create a new executable demo for Windows

--------

- get quagga to work

--------

- add IdealRadio, DistanceBasedRadio, GilbertElliotRadio.

in DistanceBasedRadio: decouple transmission range from interference range

(ie use 2 different module parameters)

--------

- model link failures, via isDown() method of InterfaceEntry. L2 modules

should understand isDown(), and FailureManager should be enhanced with

linkdown/linkup commands. See email on list archive on 9/17/2006 10:34 AM

--------

- create NetworkInterfaces/Base subdirectory: AirFrame, WirelessMacBase,

ChannelAccess, etc.

--------

- Ieee80211Mac to fire TxNotifDetails when Ack arrives for a frame. Mgmt layer

to use this notification to learn when ProbeRequest or AssociationResponse

has been transmitted.

--------

- ChannelControl: grid; instead of having pMax parameter, it should ask all

radios and collect pMax from them! (or, directly the range!)

--------

- radio models: when calculating the probability of bit errors, snirMIN is

assumed for the whole duration of the frame! This means that if snir

changes along the packet duration, we overestimate the probability of

bit error. (there should be proper integration there)

--------

- IReceptionModel:

- improve it to be able to accomodate antenna gain: calculation function

should take node positions, antenna directions (maybe this should be in

some IDirectionalReceptionmodel, plugging into some AbstractDirectionalRadio?)

- allow for implementing "good/bad channel"-type radio models (Gilbert-Elliot)

e.g. containsBadChannelState(starttime, endtime)

- allow the radio model to add extra noise over time, or modify received power

over time, e.g. using functions like

PowerList calculateReceivedPower(...)

PowerList ambientNoise(...)

(maybe this should be some IDetailedReceptionModel, plugging into

a specialized version of AbstractRadio?)

--------

- AbstractRadio: consider: don't sent up the packet if there're bit errors,

just fire some specific radio state change notification! would simplify the

Ieee80211Mac state machine a lot!

--------

- tummiepiggy's TCP bug ("TCP_S_FIN_WAIT_1 timeout" on mailing list)

--------

- quagga/socketmsg: use base/bytearraymessage instead

--------

- implement ICMP rate limiting, see e.g bsd.mod/netinet/ip_icmp.c, badport_bandlim()

- ICMP options: stopOnError (bool param), UDPBadPortSendICMP (bool param)

--------

- problem with NetworkConfigurator + RSVP's FailureManager: after deleting/recreating

LSR, configured IP addresses and host routes to PPP peers get lost.

Solution: implement failure/recovery with NF_FAILURE and NF_RECOVERY notifications!

--------

- check in Quagga documentation

- rename Daemon to QuaggaDaemon

--------

- "ack" in LinkStateMsg redundant? (never read)

Q: what is the TELinkInfo.state flag?

Q: what is UnResvBandwidth[8] indexed with? what is [4] and [7] that gets printed?

--------

- MPLS examples: there's heaps of ICMP errors coming from UDP ("port unreachable")

--------

- BUG: UDP ephemeral port setting: if chosen & stored in UDP, sending further dgrams need to look it up from the SockDesc...

--------

- reading routing files: it doesn't make sense to be able to manually set MULTICAST on an interface

- create example network with both Ethernet and PPP

- IP/IPv6: implement tunnelling

- revise fragmentation in IP

- added userId to TCPCommand -- rewrite TCPSocketMap to make use of userId

- socket must be inserted into map before bind(), so that a userId can be assigned

- what about incoming connections? how to assign userId to them?

IF IT CANNOT BE DONE: remove userId from TCP!!!!

- TCP/UDP: unspec port should be 0 not -1!!!

- NAM trace: doesn't record drops, because DropTailQueue doesn't send notifications

- there's no IPAddress::getPrefix() and getSuffix() (like IPv6Address has)

- apps: add startTime param; IPAddressResolver should only be invoked at startTime not in initialize()!

- TCP tests fail now (TCPDump has changed)

- if a node pings itself, that'll be "destination unreachable" -- fix it

- Eth: restore original conn color when transmission ends

- "headerLength" param in snrEval???

- IPv4 configurator: should fill in next hop addresses too, not only interface (esp with Ethernet)

- IPv4: make it usable with ad-hoc models

- notifboard: switch to the context of the client before calling its receiveNotification()

- todo for omnetpp-3.2:

- add handleParameterChange() to apps (needed for Scenario Manager!)

- use new opp_msgc features? (kind=..., length=...etc)

----------------------

- ExtInterface performance optimization: I am not sure how long the pointer from

pcap_next() is valid [until the next such call perhaps?], but it might be possible

to spare the allocation of the ExtFrame and the cost of copying the bytes into it

inside cSocketRTScheduler::receiveWithTimeout(). I mean, the notification message

could just directly contain the ptr from pcap_next(), plus caplen; so the ExtInterface

module could directly deserialize from the pcap buffer (and also spare copying the

bytes out of ExtFrame). There is no concurrency problem, because events are processed

in strict timestamp order, the notification message has the same timestamp as the

external event, and receiveWithTimeout() is only invoked again when time has passed

that timestamp (``if (timeval_greater(targetTime, curTime))'' line in getNextEvent()).

This could improve the packet rate the simulation can handle in real time.

----------------------

- MPLS models: "gateway" field of routing entries gets lost after TED::rebuildRoutingTable

----------------------

- ICMP: shouldn't we unify ICMP and ICMPv6...? at least types and codes?

ICMPv6 uses different type&code numeric values but this is only of interest

if we want to do emulation

- instead of sending up ICMP packet to UDP & TCP: create an ICMPErrorInfo, and

IP (IPv6) would attach that to the bogus datagram, with message kind IP_I_ICMP_ERROR.

(win: IP/ICMP dependencies can then be removed from TCP and UDP in makemakefiles!!!)

- ErrorHandling is not used anymore! do we need to send a copy of ICMP errors to the

ICMP module itself as well?

- TCP: how to handle ICMP error reports?

-----------------------

- MPLS/LDP/RSVP:

- document! ScenarioManager commands, XML file formats, unimplemented features

- Quagga ospfd: could it serve as OSPF_TE??

- mpls models use interface NAME - why?

-----------------------

- OSPFv2 TODO:

- should use the proper IPAddress class, not its own one...

- OSPFTimer class only contains a timerKind() -- is this class necessary at all...?

- rename ifIndex (SetIfIndex, GetIfIndex, etc) to interfaceId AND change default value to -1!

- OSPFRouting::LoadInterfaceParameters: interfaceType is contained in InterfaceEntry, it's redundant to read it from XML!

- try to reduce size of configuration. If Zebra can do it...

-----------------------

into the ipv6 doc:

"

Currently, IPv6 support consists of several modules. The IPv6 module

implements IPv6 datagram handling (sending, forwarding etc). It relies on

RoutingTable6 to get access to the routes. RoutingTable6 also contains the

neighbour discovery data structures (dest cache, neighbour cache, prefix

list -- the latter effectively merged into the route table). Interface

configuration (address, state, timeouts etc) is held in the InterfaceTable,

in IPv6InterfaceData objects attached to InterfaceEntry as its ipv6()

member.

The module IPv6NeighbourDiscovery implements all tasks associated with

neighbour discovery and stateless address autoconfiguration. Its data

structures are in RoutingTable6 (dest cache, neighbour cache, prefix list).

Neighbour discovery packets are only sent and processed by this module --

when IPv6 receives one, it forwards the packet to IPv6NeighbourDiscovery.

The rest of ICMPv6 (ICMP errors, echo request/reply etc) is implemented in

the module ICMPv6, just like with IPv4. ICMP errors are sent into

IPv6ErrorHandling, which the user can extend or replace to get errors

handled in any way they like.

"

-----------------------

TCP:

Slow Start should be applied every time TCP starts to send "after a

sufficiently long idle period".

"Idle" could be interpreted as when the send queue is empty (there's nothing

to send), and there's no unacknowledged data (i.e. previously sent segments

have all been acknowledged). But what is "sufficiently long"? I guess that

should be measured in RTT rather than absolute time (secs). So maybe we

should say 5*RTT is "sufficiently long"?

-----------------------

802.11 bugs: see mailing list...

-----------------------

FlatNetworkConfigurator: assigns the same address to all interfaces of a router.

This is not the usual way things are done on the internet. But if we assign

different addresses, which addr to use in the routing tables etc?

-----------------------

VOJTA:

-make connect work without bind (autoassign local address based on

destination, currently unspecified address is sent and SYN+ACK will not

be sent back (at least for 127.0.0.1, I didn't check for others))

-----------------------

from http://www.freesoft.org/CIE/Course/Section3/10.htm:

# Per-interface assignment. IP addresses are assigned on a per-interface basis,

so a host might possess several IP addresses if it has several interfaces.

For example, a host with both Ethernet and serial interfaces would have an

IP address for each. This is an important consequence of prefix-based

addressing. An IP address doesn't really refer to a host, it refers to an

interface.

If a host is known by multiple addresses, then every service on this host

can be referred to by multiple names! Addressing this host requires picking

one of these. Since the packet is addressed to the interface and not the host,

path information is introduced into the address. The exact ramifications of

this effect depend heavily on the network design. In particular, careless

design can result in a host becoming reachable by one address but not by

another. The simplest solution to this problem is to select the host's most

reliable interface and advertise its IP address as the host's primary IP

address.

==> current FlatNetworkConfigurator is shit? how to do address assignment?

see also:

"The Network Administrators' Guide" http://www.tldp.org/LDP/nag/node1.html

"IP 101: All About IP Addresses" http://www.networkcomputing.com/netdesign/ip101c.html

- according to this: network numbers and interface addresses don't necessarily

have to do anything with each other! may look COMPLETELY different

from http://www.networkcomputing.com/netdesign/ip101.html:

The important thing to realize is that while a routing table keeps track of

network numbers, no one assigns a network number to any piece of equipment.

Every interface of a router or host connected on the network must have an IP

address and a subnet mask defined (many pieces of equipment will assign a

default subnet mask if none is applied). From this IP address and subnet mask,

the network number is derived by the IP stack and tracked in the routing table.

Q: if routing tables contain network numbers, are IP addresses of router

interfaces also addressable?