Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Main points of this lecture are: Performance Metrics, Network Performance, Formats for Multimedia, Packet Transit Time, Device Transit Delay, Link Transit Delay, Flow Rate, Percent Reliability, Relationships Between Metrics, Kinds of Network Service
Typology: Study notes
1 / 29
This page cannot be seen from the preview
Don't miss anything!
The title of Chapter 7 is "Multimedia Networking" If you read between the lines, the real subject is "network performance". While formats for multimedia are transient, the need for performance is enduring. Performance Performance Monday, March 26, 2012 3:19 PM Docsity.com
Packet transit time Device transit delay: how long to forward from input to output for a device. Link transit delay: how long to transmit over a link (= payload transit time + header transit time + media access delay) = sum(transit times through all links and devices). Flow rate (also known as "throughput", "available bandwidth"): how many packets/how much data can be transferred per second = min(flow rates over all links). packets received/packets transmitted or payload received/payload transmitted. Percent reliability: how many packets actually make it to the destination: Performance metrics Performance metrics Monday, March 26, 2012 3:26 PM Docsity.com
Best effort: "we'll do the best we can, but no guarantees" Guaranteed: "flow rate (or transit time) will be within specified bounds." Kinds of network service The internet is designed as best-effort. What happens if we need guarantees? A basic quandary: Kinds of network service Monday, March 26, 2012 3:42 PM Docsity.com
Streaming audio (or video). Want to avoid "dropouts" (loss of stream) and "jitter" (momentary dropouts). This is mainly a flow rate issue (transit time is not as relevant). Typical guarantee use-case Choose resolution of stream based upon typical flow rate.
buffer stream before playback so that the buffer is "large enough" that best-effort transmission of the remaining stream beats playback to the end of the stream.
Typical solution: A typical guarantee use-case Monday, March 26, 2012 4:43 PM Docsity.com
It used to be that phone signals were sampled at very low sample rates -- as low as could retain intelligibility. Then AT&T advertised higher sample rates: "you can hear a pin drop" (that everyone knew wasn't possible with existing technology). Result: everyone conformed. An aside: "can you hear a pin drop?" An aside: "can you hear a pin drop?" Monday, March 26, 2012 5:58 PM Docsity.com
Producer: the stream source. Consumer: the media player. Concept of time: wallclock. P bits/second produced = bits transmitted from source. C bits/second consumed = bits played for user. Pre-buffered bits = B. Total bits = T. Rates: Producer/consumer timing race (T-B)/P Time at which producer ends is T/C Time at which consumer ends is (T-B)/P < T/C Producer should "win the race" and deliver data first: B > T(1-P/C) Or (where B, T, P, C are averages (Little's Laws)) What is B? Producer/consumer logic Monday, March 26, 2012 4:49 PM Docsity.com
Unreliability can be modeled as a lower effective flow rate. the raw flow rate is P packets/second and that q is the probability of a successful transmission. Suppose Then on average, every second -- assuming independent trials -- the "estimated value" of Pq packets get through successfully, so this is "a first approximation" of the effective flow rate. Modeling unreliability Independence of packet transmissions is a terribly naïve assumption. In general, two subsequent transmission attempts often face the same extenuating circumstances (e.g., congestion). But in practice, the "independence" assumption provides "usable" estimates. Aside: independence Modeling unreliability Monday, March 26, 2012 5:18 PM Docsity.com
In our basic model, there is an absolute limit on how small P can be made and retain utility. This is a limit of human perception and tolerance. (fortunately, cellphone screens are smaller than laptop screens!) But we're not done yet Meanwhile, internet paths are limited in capacity, and there is contention between paths for use of each link. So, how do we "guarantee" media delivery in the context of everything else going on on the internet? But we're not done yet Monday, March 26, 2012 5:42 PM Docsity.com
We've studied how internet routing exploits "the best link" to a destination. The "max capacity" on a path is the minimum flow capacity among the components on the path. Paths that overlap create "contending flows" on the same link. It is common for a link on the backbone to handle many contending flows. Thus, when a link oversaturates (beyond maximum flow rate/throughput), all contending flows are degraded. Flow physics What can we do about this? Flow physics Monday, March 26, 2012 6:31 PM Docsity.com
" traffic shaping ": setting priorities so that higher-priority traffic gets through. " traffic policing ": stopping traffic that does not comply with policies and agreements. Two strategies for traffic management Policy: no more than 5 gb of downloads per month at high speed, after which speed is lower. Policing : setting my traffic at a lower priority after I've exceeded the limit. Shaping : ensuring that other traffic gets priority over my traffic when it is tagged that way. Example: my cellphone: Shaping and policing Tuesday, March 27, 2012 9:04 AM Docsity.com
Diffserv = Differentiated Service Protocol "Tag" each packet with a "flow type" as it enters the internet (at an edge router ). Diffserv When is a packet dropped? What packets are given priority? Use tags to determine what happens when contention (over-subscription of a link) occurs. Diffserv Monday, March 26, 2012 6:53 PM Docsity.com
"Think globally, act locally" Bandwidth reservation requires that every link on a path reserve (at least) a given portion for serving a class. On each link, each class of packets is assigned a portion of total link bandwidth. Traffic shaping is done on each link, but the effect is upon the whole path. The throughput for the path is the minimum of the throughputs for the links Implementing Diffserv Implementing Diffserv Tuesday, March 27, 2012 9:12 AM Docsity.com
Determine the total available flow bandwidth over a link. Decide upon a percentage of flow to assign to each differentiated service class. Percentages must not exceed a total of 100%. The leaky bucket algorithm Represent the traffic for a class as a "bucket of water". Bucket "fills" from transmitting packets in the service class. Bucket "leaks" as time passes (like a water clock). To shape traffic, bucket is prevented from filling. Then The leaky bucket algorithm Monday, March 26, 2012 7:03 PM Docsity.com
Tuesday, March 27, 2012 6:45 PM Docsity.com
