Quantitative Evaluation of Embedded Systems 1. 2. 3. 4. Periodic schedules are linear programs Latency analysis of a periodic source Latency analysis of a sporadic source Latency analysis of a bursty source • Determine the MCM and choose a period μ ≥ MCM • For each actor a initialize a start-time Ta := 0 • Repeat for each arc a—i—b : Tb := Tb max (Ta + Ea – i μ) until there are no more changes Here, i denotes the number of initial tokens on an arc, and Ea is the execution time of an actor a x1 S µ A x3 B 2ms 1ms x2 C y • Choose a period μ ≥ MCM • Initialize a start-time Ta := 0 • Repeat for each arc a—i—b : Tb := Tb max (Ta + Ea – i μ) until there are no more changes 3ms 𝟎 ≤ 𝒙𝝁 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 ≤ 𝝁 + 𝒙𝝁 Quantitative Evaluation of Embedded Systems 1. 2. 3. 4. Periodic schedules and linear programs Latency analysis of a periodic source Latency analysis of a sporadic source Latency analysis of a bursty source Time (s) Throughput y(n) u(n) Latency supy(n) u(n) n 1 Tokens Time (s) Throughput y(n) u(n) Latency supy(n δ) u(n) n 1 Tokens Given a source: u(n) (n 1)μ And a periodic schedule: 𝟎 ≤ 𝒙𝝁 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 ≤ 𝝁 + 𝒙𝝁 We inductively derive the following latency bound: 𝒙 𝟏 − 𝒖 𝟏 = 𝟎 − 𝟎 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 𝒙 𝒏+𝟏 −𝒖 𝒏+𝟏 = 𝑨𝒙 𝒏 𝒎𝒂𝒙 𝑩𝒖(𝒏) − 𝒖 𝒏 + 𝟏 = 𝑨𝒙 𝒏 𝒎𝒂𝒙 𝑩𝒖(𝒏) − 𝒖 𝒏 − 𝝁 = 𝑨(𝒙 𝒏 − 𝒖 𝒏 ) 𝒎𝒂𝒙 𝑩𝟎 − 𝝁 ≤ 𝑨 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 𝒎𝒂𝒙 𝑩𝟎 − 𝝁 ≤ 𝑨 𝒙𝝁 + 𝝁 𝒎𝒂𝒙 𝑩𝟎 − 𝝁 ≤ 𝑨 𝒙𝝁 + 𝝁 𝒎𝒂𝒙 𝑩𝝁 − 𝝁 = 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 Given a source: u(n) (n 1)μ And a periodic schedule: 𝟎 ≤ 𝒙𝝁 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 ≤ 𝝁 + 𝒙𝝁 We derive the following latency bound: 𝒙 𝒏+𝜹 −𝒖 𝒏 = 𝑨𝜹 𝒙 𝒏 𝒎𝒂𝒙𝟎<𝒊≤𝜹 𝑨𝜹−𝒊 𝑩𝒖(𝒏) − 𝒖 𝒏 = 𝑨𝜹 𝒙 𝒏 − 𝒖 𝒏 𝒎𝒂𝒙𝟎<𝒊≤𝜹 𝑨𝜹−𝒊 𝑩𝟎 ≤ 𝑨𝜹 (𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎) 𝒎𝒂𝒙𝟎<𝒊≤𝜹 𝑨𝜹−𝒊 𝑩𝟎 = 𝑨𝜹+𝟏 𝒙𝝁 𝒎𝒂𝒙𝟎≤𝒊≤𝜹 𝑨𝜹−𝒊 𝑩𝟎 ≤ 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 + 𝜹𝝁 Given a source: u(n) (n 1)μ We derive the following latency bound: 𝒚 𝒏+𝜹 −𝒖 𝒏 Theorem: = 𝑪𝒙 𝒏 + 𝜹 𝒎𝒂𝒙 𝑫𝒖(𝒏) − 𝒖(𝒏) 𝟎 ≤ 𝒙𝝁the ≤ 𝐀𝒙latency 𝝁+a 𝒙𝝁 dataflow graph for a source with 𝝁 𝐦𝐚𝐱 𝑩𝟎 ≤ of ≤ 𝑪 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 + 𝜹𝝁 𝒎𝒂𝒙 𝑫𝟎 period 𝝁 and periodic schedule 𝒙𝝁 is smaller than: 𝑪 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 + 𝜹𝝁 𝒎𝒂𝒙 𝑫𝟎 And a periodic schedule: Given a source: 𝒖 𝟏 ≥𝟎 𝒖 𝒏+𝟏 ≥𝒖 𝒏 +𝝁 We inductively derive the following latency bound: 𝒙 𝟏 − 𝒖 𝟏 ≤ 𝟎 − 𝟎 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 𝒙 𝒏 + 𝟏 − 𝒖 𝒏 + 𝟏2): (monotonicity = 𝑨𝒙 𝒏 𝒎𝒂𝒙 𝑩𝒖(𝒏) − 𝒖 𝒏 + 𝟏 𝟎 ≤ 𝒙𝝁 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 ≤ 𝝁inter-arrival + 𝒙𝝁 Larger in𝑩𝒖(𝒏) the source ≤ 𝑨𝒙times 𝒏 𝒎𝒂𝒙 −𝒖 𝒏 −𝝁 = 𝑨(𝒙 the 𝒏 − latency. 𝒖 𝒏 ) 𝒎𝒂𝒙 𝑩𝟎 − 𝝁 will not worsen ≤ 𝑨 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 𝒎𝒂𝒙 𝑩𝟎 − 𝝁 ≤ 𝑨 𝒙𝝁 + 𝝁 𝒎𝒂𝒙 𝑩𝝁 − 𝝁 = 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 Theorem And a periodic schedule: Given a source: 𝒖 𝟏 ≥𝟎 𝒖 𝒏 + 𝜷 ≥ 𝒖 𝒏 + 𝜷𝝁 As an exercise, derive that: 𝒙Theorem: 𝒏 − 𝒖 𝒏 ≤ 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 + (𝜷 − 𝟏)𝝁 the latency of a dataflow graph for a sporadic bursty And a periodic schedule: source with period 𝝁 burst size 𝜷, 𝟎 ≤ 𝒙𝝁 ≤ 𝐀𝒙𝝁 𝐦𝐚𝐱 𝑩𝟎 ≤ 𝝁 + 𝒙𝝁 and periodic schedule 𝒙𝝁 is smaller than: 𝑪 𝑨𝒙𝝁 𝒎𝒂𝒙 𝑩𝟎 + 𝜹𝝁 𝒎𝒂𝒙 𝑫𝟎 + 𝜷𝝁
© Copyright 2026 Paperzz