3+2+3+3 = 11

http://news.google.cn/news?complete=1&hl=zh-CN&q=%E5%9B%9B%E5%B7%9D+%E5%9C%B0%E9%9C%87&um=1&ie=UTF-8&sa=X&oi=news_result&resnum=1&ct=title

设1号囚犯摸到的绿豆数为N。
则2号囚犯摸到的绿豆数为N+1或N-1。因为2号囚犯可以通过摸剩余绿豆的方法得知1号
囚犯摸到的绿豆数,2号囚犯摸到的绿豆数为N的话就会重复是找死，如果摸到的绿豆数与N
相差大于1的话，又会使得3号囚犯有机会使摸到的绿豆数居中。
3号囚犯也会使自己摸到的绿豆数与1、2号的紧密相邻，即使自己摸到的绿豆数比1、
2号的之中最大的大1，最小的小1。因为3号囚犯可以通过摸剩余绿豆的方法得知1、2号囚
犯摸到的绿豆总数，又知1、2号囚犯摸到的绿豆数相差为1，从而判断出1、2号囚犯各自摸
到的绿豆数。
4、5号囚犯与3号囚犯想法基本相同。即使自己摸到的绿豆数比自己前面所有的之中最
大的大1，最小的小1。
综上所述，5个囚犯摸到的绿豆数为5个连续整数。
1号囚犯存活机率。1号囚犯有两种情况必死：摸到的绿豆数最大或最小。摸到的绿豆
数最大或最小，只能由后4位囚犯决定，由分析可知后4位囚犯的摸到绿豆数的位置都只有
两个，即一组连续整数的两边。因此1号囚犯摸到的绿豆数为最大时的机率为（1/2）*（1
/2）*（1/2）*（1/2）=1/16，最小时的机率也为1/16，1号囚犯存活机率为1-（1/16）*2
=7/8
2号囚犯存活机率。由对称性可知2号囚犯存活机率与1号相同，也为7/8。
3号囚犯存活机率。3号囚犯摸到的绿豆数为最大时的机率为（1/2）*（1/2）*（1/2）
=1/8，最小时的机率也为1/8，1号囚犯存活机率为1-（1/8）*2=3/4。
4号囚犯存活机率。4号囚犯摸到的绿豆数为最大时的机率为（1/2）*（1/2）=1/4，最
小时的机率也为1/4，4号囚犯存活机率为1-（1/4）*2=1/2。
5号囚犯存活机率。5号囚犯摸到的绿豆数不是最大就是最小，必死无疑。5号囚犯存活
机率为0。

[本题到此告一段落。但是5个囚犯的策略似乎有点问题：5号囚犯在必死无疑的情况下
，还会为前4人保驾护航吗？他会不会临死拉个垫背的？于是有了以下分析。]

5号囚犯的“觉醒”（临死拉个垫背的，在必死无疑的情况下多杀人）

1-4号囚犯策略如前，则4个囚犯摸到的绿豆数为4个连续整数，而5号囚犯的“觉醒”
促使他多杀人。要多杀人，他摸到的绿豆数必须为4个连续整数的中间两个，这样有4人必
死，只有1人存活。5号囚犯必死，4号囚犯摸到的绿豆数为4个连续整数的最大或最小值，
也必死，1-3号囚犯有可能存活。

先不考虑5号囚犯。

1号囚犯存活机率。1号囚犯摸到的绿豆数为4个连续整数的最大或最小值，则必死。1
号囚犯摸到的绿豆数为最大时的机率为（1/2）*（1/2）*（1/2）=1/8，最小时的机率也为
1/8，1号囚犯存活机率为1-（1/8）*2=3/4
2号囚犯存活机率。由对称性可知2号囚犯存活机率与1号相同，也为3/4。
3号囚犯存活机率。3号囚犯摸到的绿豆数为最大时的机率为（1/2）*（1/2）=1/4，最
小时的机率也为1/4，3号囚犯存活机率为1-（1/4）*2=1/2。

考虑5号囚犯。

由于5号囚犯摸到的绿豆数必为4个连续整数的中间两个，故1-3号囚犯存活机率都将减
半。即1、2号囚犯存活机率为（3/4）*（1/2）=3/8，3号囚犯存活机率（1/2）*（1/2）=
1/4。

[5号囚犯的“觉醒”等于宣判了4号囚犯的死刑，4号囚犯考虑到这一点后，随之“觉
醒”。]

4、5号囚犯共同“觉醒”

此情况很简单，大家同赴九泉。

综合考虑后，1、2号囚犯存活机率最大。

Please see our other posting for details on the work we did on the effected file server:

http://www.dreamhoststatus.com/2008/04/06/30-min-blingy-downtime-tonight/

We are also continuing to offload data and are making good progress (it’s never as fast as we would like it to be of course). There’s excellent detail here in case you missed it:

http://blog.dreamhost.com/2008/04/07/another-anatomy/

which chronicles the situation and fills you in pretty much up to today. We’re seeing the data dip to 90% so we’re hoping to have it down in the 80’s by the end of the week (every percent we gain helps and as performance improves we can speed up the rate of moving but we’re still looking to hit critical mass where you get the proper level of performance).

Update, April 9, 2008

As we had hoped progress is speeding up as we free up more space - while the file server is showing 95% usage, around 11% of that is data that has already been moved and is no longer in use. Due to a software issue we haven’t been able to remove it yet (the admin team is working on the best way to execute that), but once that is gone we should be around 85% usage which is another large step forward.

In terms of effect, I have already seen improvement in site function for many customers as well as greatly increased speed in moving chunks of data off as well as receiving reports that mail is functioning quite a bit better. That said this issue remains at the High severity rating and in unresolved status as we have not reached a normal level of service. I can’t stress enough how sorry I am that our customers have had to put up with this but I thank those of you who have stuck with us (check the newsletter for details on what we’re doing for Blingy customers) and look forward to providing you with the level of service we strive for at DreamHost.

I apologize for the late update but we’ve been going over our options (while moving data of course). While we’re not seeing any real relief in terms of data uploads we do have some very large moves that are almost complete. Once those finish we can start deleting the data (for example one is around a half TB or around 500 GB which will be 4-5% of the total but it’s going to take until around Friday to delete it all, so we’re dealing with a ton of data). Tomorrow’s update should be earlier in the day and hopefully we’ll have some progress to reports from the large moves being complete.