Media over IP has been around for almost 20 years now, and during this time a number of habits have settled within the community in setting IP networks, focusing on fast setup and Plug’n’Play instead of taking the opportunity of good network design for better serviceability, stability, and security. This talk goes over the basics of IP and Layer 3 networking such as multicast and then presents the pros and cons of IP addressing techniques such as zeroconf, DHCP, and static addressing. The speaker will share personal experiences of failures that could have been avoided by a careful network design.
File Type:
pdf
Categories:
Network Infrastructure
Presenters :
Nicolas Sturmel - Merging Technologies
Year :
2022
dlp_document_download :
MOIP BasicsIP addressing for AVNicolas Sturmel (merging technologies) Nicolas Sturmel, PhD Senior Software Architect and Media Network Expert @Merging Technologies‘At Merging if a customer or a business case involves third party devices or a complex IT system… it eventually ends up on my desk…’Active participation in AES67 and ST2110 testing and development. Teacher, trainer, analyst, curious…Steve Albiniwanna be in the early 2000s, I guess it didn’t turn out as expected!ENS Cachan–Paris Saclayuniversity –DIGIGRAM By default: Limited connectivityImagine my young self, trying to build a home made routeur to share our brand news 128kb/s cable internet !Ohhh… wonderfull 90’s..Imagine older me discovering, 20 years later, that those addresses a commonly used by professional IP system, processing media for 10’s of 1000’s of people… It’s all about Best practices Source: Wikipedia The components of an IP addressvThe address of the nodevA 32 bit value represented as four dot separated digitsvIt is unique vThe subnet maskvIt allows computation of the network addressvThe CIDR notation uses one number between 0 and 32 appended to the node address with a /vTo contact an address outside of the subnet, the gateway is usedvThe gatewayvThe gateway sends and receives packets from/to other subnetsvIt uses routes {network address, subnet, next hop}10.0.0.5255.255.0.0 (10.0.0.0)/1610.0.0.1 In binary10.0.0.5255.255.0.0 (10.0.0.0)/1610.0.0.100001010.00000000.00000000.0000010111111111.11111111.00000000.0000000000000101.00000000.00000000.00000001 Routing and subnetsØOne node can reach a node on an other subnet if… a route existsØThe internet is built like a gigantic mesh of routes between public subnets and IPs Private addresses –they are not publicØThe IP address are unique on the internetØIPv4 use is close to full, this is partly why IPv6 is used nowØSome addresses are reserved for special use (e.g.: 224.0.0.0/4 for multicast)ØOne of this special use is private network, not directly connected to the internetØ99.99% of MOIP networks use private addressesØ3 ranges of IP addresses exist:Ø10.0.0.0/8Ø172.16.0.0/12Ø192.168.0.0/16 Good practice ØAlways choose an address in a private IP range unless you have a good reason not toØYou probably do not have a good reason to choose something else Giving addresses on the network ØEvery node on the network needs at least one unique IP addressØTo g i v e t h i s y o u c a n u s e :ØZeroConf, automatic modeØManual modeØCentralized mode, known as DHCP APIPA, the p’n’p (Pray and Play)ØAddress are randomly chosen in the 169.254.0.0/16 rangeØThere is a process for collision avoidanceØOn 99,9% of devices, this is a degraded mode ! It is subject to the infamous "DHCP attack"ØAddress collision avoidance does not really work with static addressing part of the systemØWhy would I have static addresses in an APIPA network? Because APIPA is random of course Static addressing, the spreadsheet kingdom ØThe network administrator is responsible of giving a unique address to a node ØOther values, like gateways and dns server addresses also need to be given if usedØOn a big network, documentation is necessary to avoid giving twice the same addressØIllusion of control, micromanagement, it is rarely the best solution always seems convenient. DHCP: it was built for that you know…ØIt is a service on the network to help the network administratorØThe server manages leases of addresses and other parameters given to devices that ask for it.ØYou can easily spot parts of the network that are not connected: devices can not reach the serverØIt is frequently mixed with static addressing and this is not bad (there is always one static address anyway)ØAddresses are leased, can change, but you can tie a device to an address There is "one size fit all" solutions, but we can try to sum that up: An IP address is unique It always comes with a subnet mask or a CIDR suffix If on a private network, use a private addressWe have 2^24+2^20+2^16 addresses to choose fromMore than 17000000 possibilities 169.254.X.X are unstable networks A documentation is never maintained DHCPs can have determinist addressing Invest a proportional time in your network design compared to your project. TimeNumber of devicesAPIPAStatic AddressesDHCPIT management toolsDanger Zone Few devices, no time ØThis is one of the rare case when APIPA might be okØExcept if your computer it connected to more than 1 APIPA networkØExample: one control network and one AoIP…ØThen again you have few devices, manual IP config should not be too hard More device in a very static setupØManual IP configuration is okØBut care has to be taken to check connectivity, homogeneity, and consistanceØThe hardest part: keeping your spreadsheet up to date. ØI am not the laziest, but even I have difficulties… For everything else, consider DHCPØMost administrable switches have oneØTo c h e c k c o n n e c t i v i t y a c r o s s t h e n e t w o r k j u s t u s e a s h o r t lease time (1m, 10m)ØYo u c a n l i n k M A C t o I P i n m o s t o f t h e D H C P s e r v e r sØYo u c a n r e s e r v e a s u b s e t o f I P f o r m a n u a l c o n f i g o f s o m e node Famous last words…I it will never happen !
Downloads:
7