American Megatrends ManageTrends 2.7, User Manual

Page: 335 / 367

Glossary 317
COW
Copy-On Write. In a Copy-On-Write (COW) implementation, snapshots protect the old
data by copying it to a separate snapshot space when there is a new write that needs to
overwrite it.
After the old data has been copied to the new location, the data in the old location is
overwritten with the new data. Until the next snapshot is taken, all subsequent writes that
arrive for the same location overwrite the new data, without touching the old data in the
snapshot space. In copy-on-write snapshots, therefore, every first write that tries to
overwrite an old one also spawns an additional read (of the old data) and a write (of the
old data to the snapshot space).
The snapshot space may be organized in different ways depending on the implementation.
Regardless of the implementation, however, it is necessary to maintain a table that keeps
track of the logical addresses of data that have been copied to the snapshot space. Often,
this table is just a hash (by logical address) of the location of old data versus that of new
data. This table needs to be consulted when a snapshot is mounted and read, and
preferably should be stored in the system’s main memory to allow for rapid reads from
the snapshot.
Since the quantity of old data that needs to be protected may be arbitrarily large, these
tables can grow large and unwieldy if they are maintained on a per-sector level. For this
reason, most implementations choose not to track data on a per-sector level; instead, a
larger granularity is chosen, such as 64kB. In this way, data is always copied out in 64kB
portions, regardless of the size of the new write request. Although vendors may refer to
this granularity variously as the chunk size, physical extent, snap granularity, copy size,
and so on, the implementation is the same.
DAS
Direct Attached Storage. As the name implies, this is basically any storage device that is
directly attached to the server. DAS can be a JBOD (Just-a-Bunch Of Disks) unit, hard
disk drives internal to the server, external USB hard disk drives, and so on. Suitable for
small installations. Limitations includes, inefficient provisioning, storage manage through
the server, adding capacity usually means adding another server (resulting in duplicated
overhead information), and scaling requires downtime.
Data
Distinct pieces of information usually formatted in a special way.