RE: [Xen-users] [XCP] Strange bug this avaible memory for guests

[George Shuklin <nge@xxxxxxxx>]

Thank you.

I think, I found an source of problem:


name-label ( RW)            : one
           power-state ( RO): running
         memory-actual ( RO): 8589934592
    memory-dynamic-max ( RW): 11739856896
    memory-dynamic-min ( RW): 11739856896
     memory-static-min ( RW): 8589934592

(11739856896 - 8589934592 is my lost 3Gb of memory)

I do not understand one thing: how memory-actual can be actualy lower,
than memory-dynamic-min?

How I do this:

1) I boot machine with memory-dynamic-max/min about 11 Gb. 
2) I found I have no memory.
3) I decide to reduce memory to lower values:

xe vm-memory-dynamic-range-set vm=one max=8GiB min=8GiB
(no errors).

After few retries I got reverse situation:

name-label ( RW)            : one
           power-state ( RO): running
         memory-actual ( RO): 11739856896
    memory-dynamic-max ( RW): 9663676416
    memory-dynamic-min ( RW): 9663676416
     memory-static-min ( RW): 8589934592

And I think this is REALLY bug, because memory-actual shall not be
higher, than memory-dynamic-max.


... and I think (I'm not sure) this related to second bug I wrote few
days ago about impossibility to raize dynamic-max over value, set up
prior to VM statup. I reconfirm it at XCP 0.11, XCP 0.5RC3 and XCP0.5,
so I'll repost it in separate letter.

Ð ÐÑÐ, 07/07/2010 Ð 20:40 +0100, Dave Scott ÐÐÑÐÑ:
> Hi George,
> 
> I think I can explain this one, comments inline:
> > I got really strange situation with memory on host:
> > 
> > 
> > xe host-list params=memory-free
> > memory-free ( RO)    : 10305536
> > (10Mb)
> 
> I bet this is the same value as "free_pages" (suitably converted) as shown by
>   /opt/xensource/debug/xenops physinfo
> 
> This represents the actual free memory on the host at this point in time. I 
> bet you've configured VM.dynamic_min and VM.dynamic_max so that most memory 
> is being shared between the VMs. The ballooning daemon (squeezed) always 
> keeps 9-10MiB free on the host because (IIRC)
> * only free memory with physical addresses < 4 GiB (IIRC) can be used for 
> certain xen structures
> * xen allocates memory from the highest physical address downwards
> => therefore the 10MiB we keep free will be 10MiB with low physical addresses 
> which can be used for certain xen structures
> 
> I think before we kept this memory free we got into bad states where there 
> was lots of memory free but you couldn't create a domain. If we had something 
> like memory defrag then we could let this number drop to 0, which would be 
> ideal because it would represent maximum memory utilization.
> 
> > 
> > xe host-compute-free-memory
> > 3147059200
> > (3Gb)
> 
> I had to look into the code to understand this one :) This function is 
> calculating, "the maximum amount of memory we *could* free if we ballooned 
> every domain down to its VM.dynamic_min". This is the amount of memory which 
> is considered when deciding whether we could in theory start another VM.
> 
> Does that help explain what's going on or are there still discrepancies?
> 
> Cheers,
> Dave
> > 
> > A huge difference! And xentop show different numbers too!
> > 
> > I will keep this host online "as is" and I am ready to provide
> > information for debbugging or perform needed actions.
> > 
> > --
> > wBR,George.
> > 
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-users


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users