VE lo cal time [message #4876] |
Tue, 01 August 2006 09:56 |
devotional
Messages: 2 Registered: July 2006 Location: Stockholm, Sweden
|
Junior Member |
|
|
Must be getting old, cause I was sure I posted a message on this last night, but can't seem to find it. Sorry if it turns out to be a double post after all.
I am evaluating OpenVZ as a base for running multiple test environments in one server. However, the tests involves sequences where passed time has to be simulated by changing the system time. I was hoping to find a solution where time could be changed per VE, but I'm afraid I'm out of luck...? It seems system time cannot be changed at all within a VE and changing it in VE0 will naturally affect all VE's running on that HN.
I had kind of imagined a solution where each VE could keep its own offset to the system time...wouldn't that be kind of a useful feature?
|
|
|
|
Re: VE lo cal time [message #5128 is a reply to message #4878] |
Mon, 14 August 2006 12:29 |
devotional
Messages: 2 Registered: July 2006 Location: Stockholm, Sweden
|
Junior Member |
|
|
Thanks for the pointer...I managed to locate a patch made by Jeff Dike at http://lwn.net/Articles/179825/
The patch was not completely compatible with 2.6.16-026test015.2 but no big issues. However, the solution was not complete as it only worked for the gettimeofday syscall and not for time or clock_gettime. I have added code to handle those cases and are in the middle of testing this solution.
If anyone is interested in these patches, drop me a PM and I'll make it available. If our own tests works out, we will probably keep the patches up-to-date for our own system.
I have to argue that this should be a very useful feature in a virtualization environment that aims at providing multiple isolated test environments. The solution doesn't seem to be that complicated compared to the benefits - although it could be a hairy thing to try and solve things like "what should happen when the base system clock is changed?" and how to handle timers etc. But having a solution where a VE could be given an offet at start time and then not modifiable after that would be useful enough for most test scenarios.
|
|
|