Add design for mond's CPU load collector

[ganeti-local] / doc / design-partitioned.rst

diff --git a/doc/design-partitioned.rst b/doc/design-partitioned.rst
index c059ccd..2af77a2 100644 (file)
--- a/doc/design-partitioned.rst
+++ b/doc/design-partitioned.rst
@@ -86,12 +86,22 @@ accounted separately in allocation and balancing.
 There is already a concept of spindles in Ganeti. It's not related to
 any actual spindle or volume count, but it's used in ``spindle_use`` to
 measure the pressure of an instance on the storage system and in
-``spindle_ratio`` to balance the I/O load on the nodes. These two
-parameters will be renamed to ``storage_io_use`` and
-``storage_io_ratio`` to reflect better their meaning. When
-``exclusive_storage`` is enabled, such parameters are ignored, as
-balancing the use of storage I/O is already addressed by the exclusive
-assignment of PVs.
+``spindle_ratio`` to balance the I/O load on the nodes. When
+``exclusive_storage`` is enabled, these parameters as currently defined
+won't make any sense, so their meaning will be changed in this way:
+
+- ``spindle_use`` refers to the resource, hence to the actual spindles
+  (PVs in LVM), used by an instance. The values specified in the instance
+  policy specifications are compared to the run-time numbers of spindle
+  used by an instance. The ``spindle_use`` back-end parameter will be
+  ignored.
+- ``spindle_ratio`` in instance policies and ``spindle_count`` in node
+  parameters are ignored, as the exclusive assignment of PVs already
+  implies a value of 1.0 for the first, and the second is replaced by
+  the actual number of spindles.
+
+When ``exclusive_storage`` is disabled, the existing spindle parameters
+behave as before.
 
 Dedicated CPUs
 --------------
@@ -125,7 +135,7 @@ Under KVM or LXC memory is fully shared between the host system and all
 the guests, and instances can even be swapped out by the host OS.
 
 It's not clear if the problem can be solved by limiting the size of the
-instances, so that there is plenty of room for the host OS. 
+instances, so that there is plenty of room for the host OS.
 
 We could implement segregation using cgroups to limit the memory used by
 the host OS. This requires finishing the implementation of the memory
@@ -172,8 +182,9 @@ change: there is one for the whole cluster.
 
 For example, a policy could be set up to allow instances with this
 constraints:
+
 - between 1 and 2 CPUs, 2 GB of RAM, and between 10 GB and 400 GB of
-disk space;
+  disk space;
 - 4 CPUs, 4 GB of RAM, and between 10 GB and 800 GB of disk space.
 
 Then, an instance using 1 CPU, 2 GB of RAM and 50 GB of disk would be