CS 536 ParkKey IssuesFault-toleranceThe larger the network, the more things can gowrong.E.g.: link/node failures, message corruption, soft-ware bugs−→ managing downtime: tier-1 providers−→ 99.999%Two types of failures:• independent• correlatedCS 536 ParkIn a network system with n components, assume a com-ponent fails with independent probability p−→ expected number of failures: n · p−→ probability of no failures: (1 − p)n−→ probability of k simulaneous failures: pkThus correlated failures have miniscule probability.−→ exponentially small in kCS 536 ParkIn reality, failures are not independent.−→ e.g., power outage, natural disastersWe have:−→ Murphy’s Law• issue of reliable communication• reliable network services→ main principle: redundancy• Examples:– routing of messages: alternate/back-up routes– domain name servers: duplication– transmission by space probes: forward error cor-rection (FEC)→ also used for multimedia trafficCS 536 ParkNetwork securityFeatures:• confidentiality: encryption• integrity: message has not been tampered• authentication: sender really is who she claims to be−→ “CIA”−→ foundation: cryptography−→ end-to-end−→ networking problem?CS 536 ParkModern security vulnerabilities:• denial of service (DoS) attack→ e.g., SYN flooding• distributed DoS (DDoS) attack→ e.g., commercial, personal, infrastructure• worm attacks: e.g., CodeRed, Blaster, ...→ buffer overflow: mainly bugs in MS DLLs• spam mail (security issue?)CS 536 Park• with fault-tolerance impacts QoS (quality of service)→ Aug. 04: US broadband deployment exceeds dial-up• security: trade-off with overhead→ what is the desired operating point?→ too much ⇒ too slow→ too little ⇒ too vulnerableFor example: secure routing (S-BGP)−→ “BBN vs. Cisco”CS 536 ParkBig picture:Network Security Fault-Tolerance Quality of Service−→ points in the same spectrum−→ malicious (Byzantine) vs. non-malicious−→ availability−→ service assurancesCS 536 ParkPerformanceIssues:• excessive traffic can cause congestion (analogous tohighways)• traffic volume exhibits large fluctuations→ burstiness• multimedia traffic is voluminous (even for single user)• ubiquitous access→ wired/wireless InternetPotential for bottleneck development−→ spontaneous or persistent−→ similar consequences as failuresCS 536 ParkDifferent applications require different levels of servicequality.Challenges:−→ how to provide customized QoS−→ many users and applications: scalability−→ must interoperate with legacy InternetCurrent state:• overprovisioning→ “throw bandwidth at the problem”→ tier-1 ISPs use sophisticated traffic engineering• still no Internet QoS→ changing with VoIP and content deployment• not economic→ few tier-1 providers make moneyCS 536 ParkData networking, telephony, and content convergence−→ Y2K+ trend• VoIP (Voice-over-IP): wired world→ traditional TDM-based telephony system is entirelyseparate network→ economic factors are dictating convergence→ from KaZaA to Skype• cellular voice networks: 2G, 2.5G, 3G→ what is 4G?→ telcos/cellular providers are concerned→ take-over by WLAN + IP?→ strategy: active participationCS 536 Park• peer-to-peer: rampant content dissemination→ from audio to movies→ content providers need to get into the action→ do not want to get into the action$600 question:−→ what will the wireless/wireline future hold?Mixture of high bandwidth/low bandwidth networks, wire-line/wireless, ...CS 536 ParkNetwork performanceAn overview of key concepts.Three yardsticks or performance measures:• throughput: bps or b/s (bits-per-second)• latency: msec, ms (millisecond)→ signal propagation speed• delay: msec→ includes software processing overhead• jitter: delay variation→ standard deviation etc.CS 536 ParkBandwidth vs. throughput:bandwidth—maximum data transmission rate achiev-able at the hardware level; determined by sig-nalling rate of physical link and NIC.throughput—maximum data transmission rate achiev-able at the software level; overhead of network pro-tocols inside OS is accounted for.reliable throughput—maximum reliable data trans-mission rate achievable at the software level; effectof recovery from transmission errors and packetloss accounted for.−→ true measure of communication speed−→ “goodput” or “effective throughput”−→ vs. “raw throughput”CS 536 ParkTrend on protocol implementation and overhead side:migration of protocol software functionality intoNICs; NIC is becoming a powerful, semi-autonomousdevicenetwork processors: programmable NICs and moresuch as forwarding between NICs, i.e., router−→ as opposed to ASIC based devices−→ trade-off between hardware & software−→ boundary between hardware & software blurredWith proliferation of wireless networks, lower layers havebecome important in network programming and systemdesign−→ possible project topic using iPAQsCS 536 ParkMeaning of “high-speed” networks:• signal propagation speed is bounded by SOL (speed-of-light)→∼300K km/s or ∼186K miles/s→ optical fiber, copper: nearly same→ coast-to-coast latency→ geostationary satellites: ∼22.2K miles/s→ limitation of sending a single bit (e.g., as photon)CS 536 Park• can only increase “bandwidth”→ analogous to widening highway, i.e., more lanes→ simulatenous transmission→ a single bit does not travel faster→ “high-speed” ⇔ “many lanes”→ completion time of large files fasterCS 536 ParkKey issue:−→ fat and long pipes−→ a lot of traffic in transit−→ large delay-bandwidth product (transit traffic)−→ significant damage before recovery−→ reactive cost−→ limitation of feedback systems (e.g., TCP)CS 536 ParkSome units:Gbps (Gb/s), Mbps (Mb/s), kbps (kb/s):109,106,103bits per second; indicates data trans-mission rate; influenced by clock rate (MHz) ofsignalling hardware; soon Tbps.−→ communication rate: factors of 1000Common bit rates:• 10 Mbps (10BaseT), 100 Mbps (100BaseT)• 11 Mbps (and 5, 2, 1 Mbps) for 802.11b WLAN• 100 Mbps (FDDI)• 64kb/s (digitized voice)• 144kb/s (ISDN line 2B + D service)• 1.544 Mbps (T1), 44.736 Mbps (T3)• 155.52 Mbps (OC-3), 622.08 Mbps (OC-12)• OC-24, OC-48CS 536 ParkGB, MB, kB:230,220,210bytes; size of data being shipped; in-fluenced by the memory
View Full Document