请大牛帮忙个简单的小忙：aloha 协议

俺正在仿真最基本的pure aloha 协议.众所周知，theoritically, the Maximum throughput of pure aloha is 18.3%.
我想请教一下版上的大牛解释一下这个最简单的aloha的细节：
如果发生collision以后，需要重新传输发上碰撞的这两个包，那么这两个重传的包各自需要等待一个指数分布的随机时间间隔，然后进行重传，对吧？与此同时还有按照泊松分别新到达的包要发送。 如果重传的两个包中再次与其他包发生碰撞，那么还需要第三次重传。如果是这样，pure aloha 协议运行无限长时间后，throughput 最终会很快下降到0？
我的仿真结果证明是这样的，我错在哪里了呢？是不是当有重传包的时候，应该自动把新到达包的泊松分布的平均速度减小？
请各位大牛不吝赐教，万分感谢！

On one hand, you are right that it is well known that aloha has a maximum
throughput of 18.3%. However, you have to know that this maximum value is
achieve when the total arrival rate (including the retrans) is 0.5.
On the other hand, it is also well known that aloha is not stable. The
detailed explanation is beyond what we can do here, but you might want to refer to the book Data Networks.  

Many thanks! Your reply is so fast!!
I just read the book Data Network by Bertsekas, it mentions pure aloha but also doesn't describe pure aloha  clearly... my question is:
when I simulate the pure aloha, I firstly establish a Poisson process for new packet arrival with rate "lamta". When there is a collision happening, it means later the two packets will be re-transmitted again. So during the period of these two collision packets retransmitted, do we need to decrease the arrival rate for the new comming packet? (because we need to keep the arrival rate always equal to "lamta", right?)
Hope somebody can help me ...

OPNET 里面有个Tutorial有所有Aloha协议的具体实现，你可以去参考下。

In that book, two assumptions (6a and 6b) are mentioned (p276). Usually, you
may devise you simulation based on either assumption. The arrival rate is
always \lambda, you don't need to change it in any case.
The unstable feature of aloha is clearly illustrated by Figure 4.4. If the
load G is close to the "desired stable point", then any small deviation from
the equibrium will be "drag" back to the point. Unfortunately, the stability
region is very narrow, so if the disturbance is large enough to increase
the load beyond the "unstable equilibrium", the system will go to the
"undesired stable point", which lead to very small throughput under 6a and
zero throughput under 6b. All these should be easily observed by doing
simulations.

Thank you for all the replies!!! I try to read this part and I think I get the point, although the point is still not very very clear in my mind now. At least, I get the following correct ideas for pure aloha:
1. Only when the total arrival rate (including new arrival and retransmission rate) is always a constant (lamta),  we can get a stable maximum normalized throughput 18.3% both from the theoretical and simulation results.
2. If we consider the retransmission(also keep the new arrival rate unchanged)and make the retransmission rate not very small, then it will lead the collision number greatly increased in a short time and the normalized throughput also decreases to zero very very quickly. This is the nature of pure aloha and slotted aloha.
    

sure, that is the essence of equillibrium!

I'm not sure I understand what you are saying. One thing you should keep in
mind is that the whole system is dynamic, so it is nearly impossible to stay
at the 18.3% maximum (as it is never a equillibrium point).