创建LPAR时参数desired_procs的内涵

这个我完全当作备忘录记录一下，因为今天看处理器性能时，发现的这个问题，对于我来说是第一次看到，当然PowerVM的老手可能早就习以为常了。

创建LPAR时（其面我有博文专门讲命令行在vios[ivm]下面创建一个lpar)，指定的desired_procs实际上是什么呢？如果这个LPAR上装了OS，我们跑到OS里面看，那么就是逻辑处理器，不过这个逻辑处理器的数量挺有意思，如果SMT被enable的情况下，在OS里看到的logical processor的数量是desired_procs * SMT,所以我更倾向于认为创建lpar时指定的desired_procs是物理cpu的数量，反映到操作系统里就是逻辑cpu，这二者的对应关系前面刚列出来。如果SMT没有打开，那么逻辑处理器的数量就是物理处理器的数量，如果SMT打开，并且比如SMT=2，那么在desired_procs=1的情况下，逻辑处理器的数量就是desired_procs * SMT = 1 * 2 =2;
(这个我刚才又看了一下IBM的红包书，其官方说法是：The simultaneous multithreading feature in the POWER technology allows the core to execute instructions from two ro four independent software threads simultaneous.To support this feature, we use the concept of logical processors.The operating system(AIX, IBM i, or Linux)sees one physical processor as two or four logical processors if the simultaneous multithreading feature is on. If simultaneous multithreading is off, then each physical processor is presented as one logical processor and thus only one thread).

至于如何看物理处理器，逻辑处理器以及SMT的信息，推荐个链接：

---------------------------------丑陋的分割线--------------------------------

Update (Nov.2) -- About the CPU and Memory utilization

(http://www.ibm.com/developerworks/forums/thread.jspa?messageID=14205339)

There is no way via the HMC to determine the actual utilization of memory by a partition. All you can tell is how much memory has been allocated to the lpar - whether the partition is utilizing that memory or not is unknown without running a tool such as topas on the partition itself.

Now for a discussion on Current, Pending and Runtime:

Current: How much of the resource the hypervisor currently has allocated to the lpar. 

Runtime: How much of the resource the operating system currently has 'active'. This almost always matches the 'Current' value unless you are in the middle of a DLPAR operation, or are using a workload management application that tweaks the runtime value. This value is only useful when the lpar is powered on.

Pending: How much of the resource the user has requested be allocated. This usually matches 'current' except when performing a DLPAR operation.

So, to calculate the actual memory assigned to an lpar, you just need to look at the runtime memory value (when the lpar is on), or the current memory value (when the lpar is off. There is no need to divide them, or multiply them by 100.

Now for CPU utilization. In the case of dedicated processors, just look at the current (or runtime) values as appropriate. For shared processors, the Hypervisor does in fact keep track of how many CPU cycles have been allocated to an lpar. This may be less than the runtime value, or may be more than the runtime value (in the case of an uncapped lpar). The HMC can keep track of this utilization via the lslparutil command. There are some good examples in the man page.