Creating a Shared Cache Object

If you need to create several instant space-optimized snapshots for the volumes in a disk group, you may find it more convenient to create a single shared cache object in the disk group rather than a separate cache object for each snapshot.

To create a shared cache object:

1. Decide on the following characteristics that you want to allocate to the cache volume that underlies the cache object:
   • The size of the cache volume should be sufficient to record changes to the parent volumes during the interval between snapshot refreshes. A suggested value is 10% of the total size of the parent volumes for a refresh interval of 24 hours.
   • If redundancy is a desired characteristic of the cache volume, it should be mirrored. This increases the space that is required for the cache volume in proportion to the number of mirrors that it has.
   • If the cache volume is mirrored, space is required on at least as many disks as it has mirrors. These disks should not be shared with the disks used for the parent volumes. The disks should also be chosen to avoid impacting I/O performance for critical volumes, or hindering disk group split and join operations.

2. Having decided on its characteristics, use the vxassist command to create the volume that is to be used for the cache volume. The following example creates a mirrored cache volume, cachevol, with size 1GB in the disk group, mydg, on the disks mydg16 and mydg17:
   # vxassist -g mydg make cachevol 1g layout=mirror init=active \ 
     mydg16 mydg17

   The attribute init=active is specified to make the cache volume immediately available for use.

3. Use the vxmake cache command to create a cache object on top of the cache volume that you created in the previous step:
   # vxmake [-g diskgroup] cache cache_object cachevolname=volume \
   [regionsize=size] [autogrow=on] [highwatermark=hwmk] \
   [autogrowby=agbvalue] [maxautogrow=maxagbvalue]]

   If the region size, regionsize, is specified, it must be a power of 2, and be greater than or equal to 16KB (16k). If not specified, the region size of the cache is set to 64KB.

   ---

     Note    All space-optimized snapshots that share the cache must have a region size that is equal to or an integer multiple of the region size set on the cache. Snapshot creation also fails if the original volume's region size is smaller than the cache's region size.

   ---

   If the cache is to be allowed to grow in size as required, specify autogrow=on. By default, the ability to automatically grow the cache is turned off.

   In the following example, the cache object, cobjmydg, is created over the cache volume, cachevol, the region size of the cache is set to 32KB, and the autogrow feature is enabled:

   
   # vxmake -g mydg cache cobjmydg cachevolname=cachevol \
     regionsize=32k autogrow=on
4. Having created the cache object, use the following command to enable it:
   # vxcache [-g diskgroup] start cache_object

   For example to start the cache object, cobjmydg:

   
   # vxcache -g mydg start cobjmydg

Tuning the autogrow Attributes

The highwatermark, autogrowby and maxautogrow attributes determine how the VxVM cache daemon (vxcached) maintains the cache if the autogrow feature has been enabled:

• When cache usage reaches the high watermark value, highwatermark (default value is 90 percent), and the new required cache size would not exceed the value of maxautogrow (default value is twice the size of the cache volume in blocks), vxcached grows the size of the cache volume by the value of autogrowby (default value is 20% of the size of the cache volume in blocks).
• When cache usage reaches the high watermark value, and the new required cache size would exceed the value of maxautogrow, vxcached deletes the oldest snapshot in the cache. If there are several snapshots with the same age, the largest of these is deleted.

If the autogrow feature has been disabled:

• When cache usage reaches the high watermark value, vxcached deletes the oldest snapshot in the cache. If there are several snapshots with the same age, the largest of these is deleted. If there is only a single snapshot, this snapshot is detached and marked as invalid.

---

  Note    The vxcached daemon does not remove snapshots that are currently open, and it does not remove the last or only snapshot in the cache.

---

If the cache space becomes exhausted, the snapshot is detached and marked as invalid. If this happens, the snapshot is unrecoverable and must be removed. Enabling the autogrow feature on the cache helps to avoid this situation occurring. However, for very small caches (of the order of a few megabytes), it is possible for the cache to become exhausted before the system has time to respond and grow the cache. In such cases, use the vxcache command to increase the size of the cache, or to reduce the value of highwatermark.

If necessary, you can use the vxcache set command to change other autogrow attribute values for a cache. For example, you can use the maxautogrow attribute to limit the maximum size to which a cache can grow. To estimate this size, consider how much the contents of each source volume are likely to change between snapshot refreshes, and allow some additional space for contingency.

See the vxcache(1M) manual page and the "Administering Volume Snapshots" chapter in the VERITAS Volume Manager Administrator's Guide for more information including how to grow, shrink and remove a storage cache.