I have been doing some virtualization and am now wondering if it is a good idea to use a journalled filesystem inside a virtual machine or not.
Of course a journal makes things a bit slower, but would the journal of the hosting filesystem be able to keep consistence of the image disks stored on it, in case of a power failure, or not?
This is quite a reasonable question, but, alas, it does not have a simple answer. If I understand you right, what you are saying is: by using a VM with its own journaling file system on top of a host with its own journaling file system, aren't we actually doing our job twice? Couldn't we simply use only one journaling system (most reasonably, the hosts's) and thus gaining in speed without compromising safety?
The answer to that is not completely well-known. Let me refer you to this academic paper where the two authors study exactly this effect. However, just to jump to their conclusions, at bottom of page 2, first column:
From our experiments, we have made the following interesting observations: (1) for write-dominated workloads, journaling in the host file system could cause significant performance degradations, (2) for read-dominated workloads, nested file systems could even improve performance, and (3) nested file systems are not suitable for workloads that are sensitive to I/O latency. We believe that more work is needed to study performance implications of file systems in virtualized environments. Our work takes a first step in this direction, and we hope that these findings can help file system designers to build more adaptive file systems for virtualized environments.
What I get out of this is that the situation is much more complicated than suggested by your argument, and even much less investigated than desirable. On the plus side, we may hope for a definite performance improvement as the nesting of different file-systems is further studied and understood.
