License: Creative Commons Attribution 4.0 International
cs341-notes's Introduction
The internet
Interconnected with communication links and packet switches
Two types of packet switches
Routers: Directs traffic
Link-layer switches: Connects multiple devices to one router
End system/host: Consumer devices connected to the internet
Access network
Connects the end system to the first router
The optical fiber from the central office is split at the splitter to many homes
Packet switching
Store-and-forward transmission
Packets are buffered in a queue
Packet loss occurs if packets arrive when the buffer is full
Circuit switching
Resources are reserved between end systems
Circuit switched links may be multiplexed with frequency-division multiplexing or time-division multiplexing
Packet switching may be more bandwidth efficient than circuit switching
Forwarding table: Maps destination addresses to outbound links
Tiered network
Tier-1 ISP
Has global coverage and connects with many access ISPs
Internet Exchange Point (IXP) meeting point of multiple ISPs
Point of presence: Group of routers in the provider network where customer ISPs can connect to
Regional ISP group together many access ISPs
Content-provider networks
A private connections between data centers of a company
Connects to tier-1 networks
Delay
Nodal processing delay
Time needed to read the packet header
Queuing delay
Packets are queued at a router
Transmission delay
Time needed to transmit the packet
Propagation delay
Time needed for bits to travel through a link
The total delay is the total nodal delay
Traffic intensity
La/R where L=packet bits, a=average packets/sec, R=transmission rate
If traffic intensity is greater than 1, queuing delay increases and packets may be dropped
If traffic intensity is 0, there is 0 queuing delay
Throughput: transfer rate expressed in bits/sec
Internet protocol stack
Application
Network applications and application-layer protocols
HTTP
A packet of information in the application layer is a message
Transport
Controls flow of packets
TCP
A packet of transport information is referred to a segment
Network
Moves data between hosts
A packet is called a datagram
Link
Routes datagrams through routers
A packet is called a frame
Physical
Bits are converted into signals to send through links
Network security
Types of denial-of-service attacks
Vulnerability attack: Exploits vulnerabilities
Bandwidth flooding: Floods the host with packets
Connection flooding: Establishes a large number of open TCP connections
Distributed DoS: Attacker controls compromised hosts to attack a target
Packet sniffer: A receiver that copies packets in transit
IP spoofing: Creating packets that contain a false source address
End-point authentication: A mechanism that authenticates the source address
Application layer
Reliable data transfer
A guarantee of packet validity and delivery
Ensures protection against data loss and data corruption
TCP ensures reliable data transfer with handshakes
Loss tolerant application: Media that is useful even with a small amount of data loss
Bandwidth-sensitive application
Streaming services that rely on a certain bandwidth requirement
Elastic application: Applications that work with any bandwidth, such as e-mail
Persistent connection: TCP requests and responses are sent over the same TCP connection
HTTP/2 reduces latency by multiplexing request and responses over a single TCP connection so that HOL blocking does not occur
SMTP procedure
Alice composes an email
Alice's user agent sends the mail to her mail server, and is placed in the message queue
The SMTP client retrieves the message from the queue and opens a TCP connection to Bob's mail server
SMTP handshake occurs. The SMTP client sends Alice's message to Bob's server.
Bob's mail server places the message in Bob's mailbox
Bob's user agent retrieves the message from the mailbox
Domain Name System (DNS)
DNS is used for: host aliasing, mail server aliasing, load distribution
Types of DNS servers
Root DNS: Provides the IP addresses of the TLD servers corresponding to TLDs (.com, .gov, etc.)
TLD server: Provides DNS servers for authoritative DNS servers
Authoritative DNS server: A name server that maps hostnames belonging to organizations to smaller subdomains
Local DNS server: Acts as a cache
Iterative vs recursive queries: An iterative query means the local DNS server follows mappings to resolve a full query. For a recursive query, a chain of DNS servers gets the full mapping for the local DNS server's behalf
Resource record
(Name, Value, Type, TTL)
If type=A, name is hostname and value is IP address
Maps hostname to IP
If type=NS, name is domain name and value is hostname of DNS server that can obtain IP addresses for hosts in that domain
Routes DNS queries
If type=CNAME, name is the origin name and value is the alias name
Redirect
If type=MX, name is mapped to an alias value
Mail server
Video streaming
Dynamic Adaptive Streaming over HTTP (DASH): Adaptive bitrate streaming protocol
Content Distribution Network (CDN)
Transport layer
Logical communication: The apparent perspective that communication between two hosts are direct. In reality, there are multiple relays along the physical structure.
The transport layer provides logical communication, whereas the network layer deals with physical communication
Socket
Different doors of a network process that data can pass through. AKA port.
Segments are multiplexed by adding the source and destination info to the header, and demultiplexed with the header info
0-1023 are well-known port numbers and are reserved
UDP
UDP characteristics
Finer application level control over data control
No connection establishment and no handshakes
No connection state. Compared to TCP, which does maintain connection state.
Small packet header overhead
UDP segment structure
16-bit source port #
16-bit destination port #
16-bit segment length (Including header)
16-bit checksum for header and data
Payload data
Reliable data transfer protocol
A protocol that has ensures reliable transfer of data across a network that may be subject to data loss and corruption
It doesn't mean that all packets are sent without fail; it means that all such failures are reported as errors
Checksums and ACK acknowledgement enables:
Error detection
Receiver feedback: A success or error response
Retransmission after an error response
A sequence number adds clarification to which packet an ACK is for
A countdown timer ensures that a lost packet can be sent again
Pipelined data transfer
In a stop-and-wait protocol, the utilization of time and bandwidth is low
U_sender = L/R / (RTT + L/R) where RTT = return trip time
By sending multiple packets before ACK, utilization is higher
Go-back-N protocol: If an ACK for sequence N is not returned before timeout, the Nth packet is sent again, regardless of if an ACK for N+M is received
Selective repeat protocol: The sequence number of each ACK is marked. If there is an unmarked sequence number after timeout, that packet is sent again. Marked packets are never sent again.
TCP
Connection-oriented: A handshake must occur before the main message is transmitted
TCP segment
A chunk of data of a data stream
TCP segment header structure
16-bit source port
16-bit destination port
32-bit sequence number field: Byte number of segment
32-bit acknowledgement number field: Byte number of next expected segment
16-bit receive window: Number of windows that the sender can store
4-bit Header length field
Variable-length options field
6-bit flag field: ACK, RST, SYN, FIN, PSH, URG
Receivers send cumulative acknowledgements, or acknowledgements of the last sequence number received
Handling of out-of-order segments is up to the implementation, but implementations typically have a buffer to store out-of-order segments
Piggybacking: Delaying the ACK to send it back in the next segment
Timeout interval is calculated by keeping an estimated RTT
Estimated RTT changes based on the sampled RTT
Exponential weighted moving average: Formula to calculate estimated RTT
Timeout interval is set above the estimated RTT
Timeout interval = Estimated RTT + 4 * DevRTT
Fast retransmit: A receiver may send two duplicate ACKs after an out-of-order segment. The sender sends back the missing segment immediately, rather than waiting for timeout.
Slow-start: Since the receiving side's buffer and bandwidth capacity is unknown, the segment rate is slow at first
Flow-control: Matches the rate of data transfer for the sender and receiver so that the receiver's buffer doesn't overflow, using the receive window field
Congestion control
With a higher rate of segments, bigger buffers, higher bandwidth, and other network resources are necessary
Solution
Additive increase, multiplicative decrease (AIMD)
Increase segment rate until timer rate or triple ACK, which means a segment was dropped
A dropped segment indicates congestion
On segment loss, the segment rate is decreased
Segment rate may be halved (TCP Tahoe) or decreased by one (TCP Reno)
TCP CUBIC: Increases segment rate by logarithmic growth to send more segments until the limit
Explicit congestion notification
Explicit congestion notification: A bit in the segment header that a router can set to indicate that it is experiencing congestion
TCP Vegas: Measurement of congestion based on RTT, not segment loss
Network layer
Although data in the network layer are datagrams, they are also referred to as packets
Forwarding
Router-local action of sending a packet to the appropriate output destination
Managed inside the data plane
Routing
Network-wide process of determining end-to-end paths
Managed by the control plane
Forwarding table: A mapping of IP ranges to output link
Software-defined networking: An approach to routing where software in the control plane sets forwarding tables, instead of through human management
The network layer only guarantees best-effort service
The network layer does not guarantee in-order delivery or reliability
Router
Input port
Physical end termination of incoming link
Decapsulates (decodes) packets
Uses forwarding table to get output link interface
Longest prefix matching rule: Find forwarding table entry with the longest matching prefix
The packet is sent to the switching fabric
Switching fabric
Connects input ports to output ports
Memory switching
Old routing mechanism where a buffer stores and delivers the packet
Limited to memory IO speed
Bus switching
The output is attached to a packet's header and sent along a bus
Only one packet can cross the bus at a time
Interconnect network
A crossbar switch that delivers the packet without blocking
Output port
Queues packets
Size of buffer needed is B = RTT * C / sqrt(N) where C = link bandwidth and N = number of TCP flows
Transmits packet to outgoing link
Output order can be by FIFO or by priority of packet
Weighted fair queuing discipline: A round-robin method of serving one packet per each class of priority
HOL blocking occurs when an output port is busy, and subsequent packets cannot be sent
Routing processor
Performs control-plane functions
Communicates with remote controller to receive forwarding tables
Internet Protocol
IPv4 header
Version (Always 4)
Internet Header Length: Size of header
Differentiated Services Code Point (DSCP): Specifies type of service, such as real-time or non-real-time service
Explicit Congestion Notification (ECN): Congestion control
Total length: Packet size
Identification, flags, fragment offset: Identifies fragmented datagrams, which are datagrams broken into smaller datagrams that are reassembled at the destination
Time to live: A number that limits a datagram's lifetime. Each router decrements this number by 1. When it hits 0, the datagram is discarded.
Protocol: Identifies transport layer protocol, such as 6 for TCP.
Header checksum: For error correction of header. Each router must recompute the checksum since a value may change. UDP and TCP have separate checksums.
32-bit source IP address
32-bit destination IP address
Options: A rarely used field for setting the behavior of the datagram.
Payload
Subnet
A subdivision of an IP network
Subnet mask: Used to specify the size of the subnet
CIDR: Specifies prefix and mask
Classful addressing: 8-, 16-, and 24-bit subnet addresses. These classes were either too small or too large for organizations, and it led to a rapid depletion of unallocated addresses.
The ICANN allocates blocks of addresses to ISPs. The ISPs further divides their block to individual organizations.
IP addresses in an organization can be manually distributed to hosts, or Dynamic Host Configuration Protocol (DHCP) can be used to do this automatically.
DHCP protocol
DHCP discover message is sent via UDP from the connecting client to the broadcast address 255.255.255.255
DHCP offer message is broadcasted to 255.255.255.255, containing the proposed address for the client (YIADDR), network mask, and address lease time
Client broadcasts a DHCP request for the address
DHCP acknowledges the request
Network address translation (NAT)
The port number is used as an extension to an IP address
A client is assigned a private IP address in the address space 10.0.0.0/24
When a client sends a packet, a NAT router replaces the source address with a public address and a unique port number according to a NAT translation table
Incoming packets have the destination address and port rewritten to the private IP address
IPv6 header
Version (Always 6)
Traffic class: Type of service
Flow: Identifies a stream of packets, such as for a media stream
Payload length
Next header: Identifies the transport layer protocol of the payload
Hop limit: Replaces TTL field in IPv6
128-bit source address
128-bit destination address
Payload
IPv6 removes the fragmentation, checksum, and options fields
IPv6 transition
Tunneling: Packaging IPv6 datagrams into IPv6 datagrams so that IPv4 routers can pass IPv6 datagrams
Generalized forwarding
Instead of destination-based forwarding, generalized forwarding is a layer 3 switch that uses header fields of different protocols in different layers
OpenFlow flow table
Implementation of software-defined networking
Match-plus-action forwarding table
Each entry contains:
A set of header field values: Matched to the packet
A set of counters: Can include number of matched packets and time since last update
A set of actions to take: For example, forwarding packet to an output port, drop the packet, modify header fields
Some use cases of SDNs are routers, load balancers, and firewalls
Middlebox
An intermediary box performing functions different from those of an IP router
Some examples: NAT translation, security services, load balancers
Network control plane
Two approaches to router control
Per-router control
A routing algorithm runs in every router individually
Logically centralized control
A centralized controller distributes forwarding tables to all routers
Control agent: An agent in a router that communicates with the central controller
Routing algorithms
Static routing algorithm: Routing tables change very slowly, often because they are manually set
Dynamic routing algorithm: Routing tables are changed often due to automated systems
Load-sensitive algorithm
Link costs vary depending on the level of congestion
Modern routing algorithms are load-insensitive
Link-state routing algorithm
Used for centralized control because all link costs are known by the controller
Dijkstra's algorithm is an implementation of LS routing
Worst-case complexity of O(|E| + |V|log(|V|))
Distance-vector routing
Iterative, asynchronous, and distributed
For per-router controllers
Bellman-Ford equation
Worst-case complexity of O(|V||E|)
The solution to the Bellman-Ford equation provides the forwarding tables
Count-to-infinity problem
A routing loop where a disconnected link causes routers to increment the cost of that link to infinity
Can be avoided with poisoned reverse
Open Shortest Path First (OSPF)
Intra-autonomous system routing protocol
Autonomous system: A group of routers that are under the same administrative control
Routers store path information for their own autonomous system
OSPF
Open standard used for intra-AS routing
Uses Dijkstra's algorithm
Each router is given a graph of the AS, and each router computes the shortest paths to all subnets
Authenticates packets
Border Gateway Protocol (BGP)
Inter-autonomous system routing protocol
Purpose
Obtain prefix reachability information
Determine best routes to prefixes
Types of routers in autonomous systems
Internal router: Connects only to hosts and routers within the AS
Gateway router: Connects to internal routers of its own AS and routers of other ASs
Routing information is exchanged through BPG connections which are TCP-based
External BGP (eBGP): A BGP connection that spans two ASs
Internal BGP (iBGP): A BGP connection within the same AS
Routing information contains BGP attributes
AS-PATH
A list of ASNs that the advertisement has passed through
A router adds their ASN to the front of the list and passes it on
If a router sees its own ASN, the information is discarded as it looped back
NEXT-HOP
IP address of the router interface that begins the AS-PATH
BGP route selection priority
Local preference value
A policy decision by the AS to send packets to a certain router
The route with the shortest AS-PATH is selected
Hot potato routing
Packets are passed to the AS with the lowest immediate cost
IP-anycast
Different hosts have the same IP addresses
If a router sees different routes to the same address, the best route is picked
It is useful for CDNs, as routers will pick the fastest CDN
Also used for DNS servers
Software-defined networking (SDN)
Key characteristics of the SDN architecture
Traditional routers only read the IP address from a packet, but SDN flow tables consider many more header fields of packets
The data plane consists of switches, and the control plane consists of servers and software to manage flow tables
Controllers hold network state information and control the switches
Network-control applications receives state information from the controller and computes routing paths and flow tables
The SDN can be programmed through APIs
The controller API is the northbound interface, and the switch API is the southbound interface
OpenFlow
OpenFlow is a protocol in which the control plane communicates with the data plane
The controller can send the following messages:
Configuration: sets configuration parameters
Modify-state: Adds or deletes flow entries and port properties
Read-state: Collects statistics and counter values
Send-packet: Specifies a message to send
Switches can send the following messages:
Flow-removed: A flow table has been removed
Port-status: A port status was changed
Packet-in: A packet that does not match a table entry is sent for additional processing
Example SDN interaction
The link between switches s1 and s2 fails. Port-status message is sent to inform the controller of the failure.
The SDN controller notifies the link-state manager
The network-control application receives notification of the link-state change
The link-state manager computes new least-cost paths
The flow table manager creates flow tables
The flow table manager sends modify-state messages to update flow tables
Internet Control Message Protocol (ICMP)
ICMP is for error reporting, such as unreachable hosts
ICMP is part of the network layer
Ping and traceroute use ICMP
Network management
Managing server: Operated by network managers and controls the network devices
Managed device: Network equipment such as hosts, routers, and switches
Data
Configuration data: Device information explicitly configured by the network manager
Operational data: Information that a device collects as it operates, such as the list of immediate neighbors
Device statistics: Status indicators and counts
Network management agent: Software running in the managed device that communicates with the managing server
Network management protocol: A protocol that allows communication between the managing server and managed devices
Simple Network Management Protocol (SNMP)
Application layer protocol for conveying network-management control and information messages
Used to set and retrieve MIB object values on a managed device
Also used for sending trap messages (notifications) to a managing server
Some uses for MIB objects are: the number of datagrams discarded, number of carrier sense errors, and the version of a software
Network Configuration Protocol (NETCONF)
Similar to SNMP, NETCONF is used for setting and retrieving operational and configuration data
XML-based RPC
YANG: Data modeling language used for NETCONF
Commands are applied to a set of devices
The difference between SNMP is that SNMP sets values on individual devices, whereas NETCONF sets the configuration of the network as a whole
Link layer
Node: A device that runs a link-layer protocol
Data is transmitted in link-layer frames
Link layer services
Framing: Data is packaged in frames
Link access: A MAC protocol manages frame transmission
Reliable delivery: Even though TCP ensures reliability, nodes can resend frames between links to avoid the need for end-to-end retransmission
Error detection and correction: Error detection bits are included in frames to fix bit errors
Network adapter/network interface controller (NIC): Hardware that handles link layer responsibilities
Error detection and correction
Parity bit: 1 bit included with a string to detect errors. Checks that the total number of 1-bits in a string is even or odd.
Forward error correction: The ability to detect and correct errors without retransmission
Internet checksum
The sum of 16-bit words of the data and the checksum should be zero
Used for IP packets and TCP and UDP messages
Cyclic redundancy check (CRC)
Better algorithm than a checksum for detecting burst errors
Multiple access link
Point-to-point link: A link that consists of a single sender and a single receiver
Broadcast link: A link that has multiple sending and receiving nodes
A multiple access protocol regulates transmission by nodes in a broadcast channel
Channel partitioning protocols
Time-division multiplexing
Frequency-division multiplexing
Code division multiple access (CDMA)
Each node is assigned a code to encode data
Nodes can transmit simultaneously and receivers can still receive data
Used in older mobile phone standards
Random access protocol
Nodes transmit frames at full rate, but when there is a collision, frames are retransmitted after a random delay
Binary exponential backoff: The random delay to retransmit increases by twice the max delay of the previous transmission attempt
Carrier sense multiple access with collision detection (CSMA/CD)
Carrier sensing: A node listens for ongoing transmissions before transmitting to avoid collision
Even if carrier sensing is used, channel collisions occur because of channel propagation delay
Obsolete method for Ethernet
Taking-turns protocol
Polling protocol: A master node polls each transmitter for a certain number of frames
Has a polling delay
Avoids collisions
Token-passing protocol: Nodes exchange tokens to decide who can transmit
Data Over Cable Service Interface Specification (DOCSIS)
For sending data over cable
Uses FDM for data transmission, and TDM to assign slots
New modems can send a request for a slow using random access protocol
Switched local area network
MAC address
A device-specific and (usually) permanent address
6 bytes or 48 bits
Used in link-layer to address frames
Broadcast address is FF-FF-FF-FF-FF-FF
Address Resolution Protocol (ARP)
Translates IP addresses to MAC addresses
Ethernet
Link and physical layer standards
An Ethernet frame contains:
Preamble: A wake-up and sync signal
Destination address: MAC address of destination adapter
Source address: MAC address of sender
Type: Indicates the payload protocol
Data: The payload
CRC: Checksum for frame
Ethernet has various flavors, such as 10, 100, 1000 MB/s, 10, 40 GB/s
Link-layer switch
Filtering: A switch function that determines whether a frame should be forwarded or dropped
Switch table: A mapping of MAC address to switch interface
A switch can take one of three actions for a frame:
If there is no entry for the destination address, that frame is broadcasted
If the entry for the destination address has the same interface as the one the switch received the frame from, the frame is discarded
If the entry for the destination address is not the same interface, the frame is forwarded to the interface
Switches learn which MAC address is connected to which interface
Link-layer switches works automatically, without manual configuration
If no frames are received from an address within a given period of time, the entry is deleted
In contrast to a hub-and-spoke LAN architecture, a mesh of switches reduces the chances of collisions
Switches allow different Ethernet protocols to work together
Routers are network-layer (and above) devices, whereas switches are link-layer
Virtual local area network (VLAN)
A VLAN is a subdivision of a switched LAN that functions like a single isolated network
The ports on a switch are partitioned into VLANs
A VLAN trunk can carry traffic across the VLANs
A frame can contain a VLAN tag which identifies VLAN that the frame belongs to
Multiprotocol label switching (MPLS)
Improves forwarding speed of routers by attaching a fixed-length label to packets
Packets can be forwarded based on the label and not the IP headers
Instead of finding the longest-prefix match, the router only finds the matching label from a lookup table
MPLS enables multiple-path routing
MPLS can be used to implement VPNs
Data center networking
Blade: A modular server computer
Top of rack (TOR) switch: A switch connecting all blades in one rack
Border router: A router that connects the data center network to the public internet
Tier-2 switches connect groups of tier-3 TOR switches and tier-1 switches connect tier-2 switches
Load balancers operate at the tier-1 level
To enable fast communication within a data center network, switches are highly interconnected so that there is no bottleneck link
Wireless links
Signal-to-noise ratio (SNR): The relative strength of the received signal compared to background noise
Bit error rate (BER): Number of bit errors per total bits sent
When SNR is higher, BER is lower
Higher transmission rate = higher BER
CDMA is one of the protocols used for wireless transmission
Data is multiplied by a code
Chipping rate: One bit is subdivided into multiple mini-slots
Each mini-slot is multiplied by the code, and divided by the code at the receiving end to decode
This enables multiple transmissions in the same airwave
WiFi
WiFi is defined in IEEE 802.11
Operates in 2.4 GHz and 5.1 GHz
Basic service set: One or more wireless stations and a central base station or access point
Access point (AP): A device that provides WiFi service, and is connected to a router. Not all wireless stations are APs.
Service Set Identifier: An identifier for a WiFi network that is often in a natural language
A 2.4 GHz service has 11 partially overlapping channels. Channels 1, 6, and 11 are 4 channels apart, and thus do not overlap each other.
Wireless ad hoc network: A peer-to-peer wireless network that does not have a router.
Beacon frame: An message that an AP sends to advertise its service
Passive scanning vs active scanning for beacon frames
A device can passively listen for beacon frames, which are sent periodically by APs
A device can also request a beacon frame from any APs in the area
802.11 uses CSMA with collision avoidance (CSMA/CA)
Devices that use CSMA/CD are listening to collisions as they are transmitting
Devices that use CSMA/CA do not listen for collisions
Instead, the AP sends an acknowledgement message after receiving a message
If the acknowledgement is not received within a timeframe, a device sends the message again after a binary exponential backoff
802.11 has a reservation scheme using request to send (RTS) and clear to send (CTS) control frames
802.11 frame
Has 4 address fields, which contain the MAC addresses of the wireless station, the AP, and the router
Contains the IP datagram in the payload
The AP is responsible for converting Ethernet frames into 802.11 frames and visa versa
Bluetooth can be used to create a wireless network and are called wireless personal area network (WPAN)
Bluetooth uses time-division multiplexing
Unlike WiFi, a Bluetooth network is self-organizing
One centralized controller is designated to manage the network
Mobility management
Home subscriber server (HSS): A database containing information about the cellular subscriber, and is located in the home network.
Base station: The cell tower that connects to the user's device
PDN gateway and serving gateways: Connects the home network to the visiting network. The PDN is on the home side and serving gateway is on the device's side.
Mobility Management Entity (MME): Responsible for connecting to the HSS to retrieve user information and to authenticate the user.
LTE mobility connection steps
Mobile device associates with the nearby base station
The device is identified with an International Mobile Subscriber Identity (IMSI)
Control-plane configuration in the visiting network
The MME uses the IMSI to look up from its database or contact the HSS for authentication and network service information.
The MME informs the HSS that the user's device is in their network.
Data-plane configuration in the visiting network
The MME configures a tunnel from the base station to the serving gateway, and to the PDN gateway.
Handoff steps (Current base station is where the device is connected to and target is where the device is moving to)
The current base station sends a handover request to the target base station
The target station allocates channel resources, and replies to the current base station with a handover request acknowledge message
The current base station informs the device of the target station's identity and channel access information. The device switches to the target base station.
The current base station forwards any packets for the device to the target base station
The target base station informs the MME that it is serving the device. The MME reconfigures the tunnel to go to the target base station.
The target base station informs the current base station that the tunnel was configured, so that the current base station can release resources for that device.
The target base station now starts delivering packets to the device.
Mobile IP: Allows mobile devices to move from one network to another while maintaining a permanent IP address
Wireless in the application layer
TCP mostly works the same whether on a wired or wireless context
Its congestion control response can be improved in a wireless setting with the following approaches:
Local recovery at the network and link-layer levels using checksums and retransmission
TCP awareness of wireless links, so that congestion control is only triggered by packet loss and not packet corruption
Breaking the end-to-end connection into wired and wireless segments so that error recovery works differently depending on the segment that the error occurred
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent