M/M/1/K与M/D/1/K队列拒绝概率差异及服务时间影响咨询

Great question—this is a classic queueing theory result that clicks once you unpack the dynamics of random vs. deterministic processes. Let’s break this down step by step.

Why M/D/1/K Has a Lower Rejection Probability #

The core difference comes down to service time variability and how it drives queue overflow:

• In an M/M/1/K system, service times follow an exponential distribution. This means there’s a real chance of extremely long service times (the "long tail" of the curve). When one of these slow services hits, new arrivals keep coming at the average rate, quickly filling the limited queue capacity K. Any arrivals after that get rejected.
• In an M/D/1/K system, service times are fixed—no surprises. Every customer takes exactly the same amount of time to process. This eliminates the risk of a sudden, prolonged service that causes the queue to blow up. The queue grows and shrinks in a steady, predictable rhythm, making it far less likely to hit the capacity limit K.

To make it concrete: Suppose average service time is 1 minute. In M/M/1/K, you might have a customer that takes 5 minutes to serve—while that’s happening, 5 new customers could arrive (if arrival rate matches service rate), filling up a small K. In M/D/1/K, each customer takes exactly 1 minute, so the queue never gets a chance to spike unexpectedly.

Another way to frame this is queue length variance: M/M/1 has a much higher variance in queue length than M/D/1. Higher variance means more frequent extreme values (like hitting K), which directly translates to a higher rejection probability.

General Effects of Deterministic Times on Queues #

Whether you’re talking about deterministic service times (D) or deterministic arrival times (replacing the M arrival process), reducing variability has consistent, positive impacts on queue performance:

Deterministic Service Times (vs. Random Service Times) #

• Lower rejection probability (finite capacity systems): No sudden long services mean fewer queue spikes that hit capacity limits.
• More predictable waiting times: Customers experience less variation in wait times—no one gets stuck waiting hours because of a single slow service.
• Slower waiting time growth at high utilization: For infinite capacity systems, M/D/1’s average waiting time grows as ρ²/(2(1-ρ)) (where ρ is utilization), while M/M/1’s grows as ρ/(1-ρ). At high utilization (ρ close to 1), M/D/1’s waiting time increases far more slowly than M/M/1’s.

Deterministic Arrival Times (vs. Random Arrival Times) #

• Reduced queue buildup: No sudden bursts of arrivals (like 3 customers showing up at once when you expect 1 per minute) mean the system never gets overwhelmed unexpectedly.
• Lower average waiting time: Steady arrivals pair with either fixed or random service times to keep queues shorter on average.
• Lower rejection probability (finite capacity): Just like deterministic service times, predictable arrivals avoid filling the queue to capacity in one go.

Key Takeaway #

At the heart of all these effects is a simple principle: less variability = fewer extreme system states. Deterministic service or arrival times eliminate the random "spikes" that cause queues to overflow or wait times to skyrocket. This is why M/D/1/K outperforms M/M/1/K in rejection probability—you’re removing one major source of uncertainty that can push the system beyond its capacity limits.

内容的提问来源于stack exchange，提问作者PassingBy