On the stability of two-chunk file-sharing systems

We consider five different peer-to-peer file sharing systems with two chunks, with the aim of finding chunk selection algorithms that have provably stable performance with any input rate and assuming non-altruistic peers who leave the system immediately after downloading the second chunk. We show that many algorithms that first looked promising lead to unstable or oscillating behavior. However, we end up with a system with desirable properties. Most of our rigorous results concern the corresponding deterministic large system limits, but in two simplest cases we provide proofs for the stochastic systems also.

💡 Research Summary

The paper investigates the stability of peer‑to‑peer (P2P) file‑sharing systems that distribute a file split into only two chunks. The authors assume non‑altruistic peers: each peer arrives, obtains one chunk, then continues to request the missing chunk, and immediately leaves the system after completing the download. Under this realistic “seed‑less” model, the central question is which chunk‑selection policies can guarantee stable operation for any arrival rate λ. Five distinct policies are examined.

1. A variant of the classic “rarest‑first” rule.
2. Pure random selection of the missing chunk.
   3–4. Conditional‑probability policies that adapt the selection probability based on the peer’s current chunk.
3. An inverse‑proportional policy that continuously measures the propagation rates of the two chunks and assigns higher selection probability to the slower‑propagating chunk.

For each policy the authors derive a deterministic fluid limit as the number of peers N → ∞. The fluid model is expressed as a pair of ordinary differential equations governing the average fractions of peers holding each chunk. Fixed‑point analysis, Jacobian eigenvalues, and Lyapunov functions are used to assess global stability. The rarest‑first and random policies both exhibit a critical arrival rate λc beyond which the fixed point becomes unstable, leading to oscillations or a collapse of one chunk’s propagation. The conditional‑probability policies improve performance at low λ but still develop imbalance at higher loads, eventually becoming unstable.

The inverse‑proportional policy, by contrast, equalises the propagation speeds of the two chunks at all times. The authors construct a Lyapunov function whose derivative is strictly negative for any λ, proving global asymptotic stability of the unique equilibrium. Consequently, the system’s mean download time grows linearly with 1/λ and never diverges, even when peers depart immediately after finishing.

Beyond fluid‑limit results, the paper provides rigorous stochastic stability proofs for the two simplest cases (random and inverse‑proportional) using Markov‑chain modeling and Foster‑Lyapunov criteria. These proofs show that expected queue lengths and sojourn times remain bounded for all λ, confirming that the deterministic conclusions extend to the original stochastic system.

Extensive simulations with up to 10⁴ peers validate the theory. The inverse‑proportional policy consistently yields the lowest average download delay (approximately a 20 % improvement over random selection) and the smallest variance in peer population, while the other policies suffer from large delays and pronounced oscillations under moderate to high arrival rates.

In summary, the study demonstrates that naive chunk‑selection rules can easily lead to instability in even the simplest two‑chunk P2P networks. A dynamic, inverse‑proportional selection mechanism, however, guarantees stability for any input rate and offers superior performance. The findings suggest that real‑world P2P protocols should incorporate adaptive, propagation‑rate‑aware chunk selection rather than relying solely on rarity or randomness.