Beruflich Dokumente
Kultur Dokumente
Abstract
This white paper describes the SnapView product and how it can be used to accelerate business processes.
Basic concepts for SnapView are discussed, as well as various potential implementations for customers.
Published 12/3/2004
12/3/2004
Table of Contents
Executive Summary............................................................................................ 4
Intended Audience.............................................................................................. 4
Introduction ......................................................................................................... 4
SnapView Terminology ................................................................................................................ 4
Business Operations Supported by SnapView ................................................ 5
Backup and Recovery Operations ............................................................................................... 5
Report Generation ....................................................................................................................... 5
Data Warehouse/Decision Support System Population .............................................................. 6
Application Testing and Upgrades ............................................................................................... 6
User Training................................................................................................................................ 6
BCVs Compared with Snapshots ...................................................................... 7
Data Access ................................................................................................................................. 7
Resource Requirements .............................................................................................................. 8
Performance Considerations ....................................................................................................... 8
BCV Synchronization Effect on Source I/O.............................................................................. 8
Snap Session Effect on Source I/O.......................................................................................... 9
Use of ATA Drives.................................................................................................................... 9
SnapView Implementation ................................................................................. 9
Managing Replicas ...................................................................................................................... 9
Number of Replicas ................................................................................................................... 10
Data Recovery ........................................................................................................................... 10
Conclusion ........................................................................................................ 11
Appendix ........................................................................................................... 12
BCV Performance ...................................................................................................................... 12
Snapshot Performance .............................................................................................................. 14
Performance Conclusions.......................................................................................................... 14
Executive Summary
This paper describes how EMC SnapView replicas can be used to help accelerate business processes.
Specific business application uses are discussed, illustrating the benefit of being able to perform various
business tasks concurrently. Differences between business continuance volumes (BCVs) and snapshots are
discussed so that users can understand the tradeoffs of each feature, and can thus determine which feature
may best suit their business need.
Intended Audience
The intended audience for this paper is members of the EMC field community who interact with customers
and need to understand how SnapView can assist with business processes. Customers are also invited to
read this paper, to get a first-hand understanding of the benefits of using SnapView in their environments.
Introduction
SnapView is an optional software package for the EMC CLARiiON storage system1. Using SnapView,
users can create secondary replicas of data to support secondary tasks in their environment. For instance,
users can create a replica of their data to be used in backup operations, while their production data remains
online. In this way, SnapView replicas provide concurrent data access, providing for greater flexibility and
productivity in customer environments.
SnapView provides two types of replicas: business continuance volumes (BCVs) and snapshots.
SnapView BCVs are full copies of the original production data. That is, if the original production data is
250 GB, the BCV must also be 250 GB. SnapView snapshots implement a pointer-and-copy-based design,
where pointers are directed to the original production data until a write request comes in to change that
data. At that point, the original data is copied in a reserved area, and accordingly, the pointers to that data
are redirected to the new location to preserve the point-in-time view of the data. In this way, only a
fraction of the space of the original source data is required to enable snapshots; that is, snapshot capacity is
proportional to write I/O volume. BCVs and snapshots can be presented to servers as separately
addressable volumes for both reads and writes and thus provide concurrent data access. For more
information on SnapView functionality and operations, refer to the white paper SnapView Snapshots and
Sessions.
SnapView Terminology
The following SnapView terms are provided to help clarify the components and features of SnapView.
BCV The LUN that is a separate disk copy of the source LUN. BCVs must be the same size as the
source LUN, but can be a different RAID and disk type. Fracturing a BCV is the process of breaking off
the BCV from the source so that it can be presented to a secondary server. To update the contents of the
BCV, the user must take it offline from the secondary server and then reconnect it to the source (by
choosing to synchronize it).
SnapView Session The point-in-time reference of a source LUN. Note that the session is not visible to
a server until a snapshot is activated to it.
SnapView Snapshot The virtual device that is activated to a session and presented to a server to enable
visibility into that session. Active snapshots are fully read- and write-capable. Once the snapshot is
deactivated, however, all writes via the snapshot will be deleted.
Source LUN The LUN containing production data of which you want to make a replica.
1
SnapView is supported on all CLARiiON models, CX300 and higher. Also, SnapView is supported on the
AX100. (The only CLARiiON platform on which SnapView is not supported is the CX200.)
SnapView allows users to create up to eight BCVs of each source logical unit (LUN), and likewise, up to
eight snapshots of each source. Furthermore, users can create snapshots of BCVs; since the BCV is serving
as the snapshot source in this case, users can create up to eight snapshots of each BCV. Taken to its
maximum, users can create up to 8 BCVs per source and can create up to 8 snapshots of each BCV, which
calculates to 8 BCVs and 64 snapshots. In addition, users could have an additional 8 snapshots directly off
the source for a total of 72 snapshots of the source (plus the 8 BCVs).
Report Generation
As with running backup operations on SnapView replicas, users can also run reports on SnapView replicas.
Just as with backup operations, reports are typically generated when organizations can afford to take down
the production application. With SnapView replicas, reports can be run whenever needed, on data points
that are relevant, facilitating report generation based on relevance, rather than schedule availability.
User Training
In many environments, employees need to use a particular data set for training or practice, where they can
make modifications and changes as theyre using the data set; yet at the end of the training period, it would
be beneficial to reset the data to its original state. SnapView replicas allow the creation of these
training data sets. In some instances, users have created SnapView sessions at the start of the training
period and activated snapshots to those sessions so that the employees can use these virtual LUNs for
their training. At the end of the training period, the session can simply be stopped and then re-created at
the beginning of the next training period, thus allowing users to reset their training environment in a
matter of moments.
Where the data set is production data, and avoiding performance impact on that production data is critical,
users often opt to first create a BCV of the production data, and then run the Snap sessions against the
BCV. This way, the BCV absorbs the performance impact of the Snap sessions. Furthermore, the BCV
preserves the consistent gold copy of the data since its contents will remain unchanged until the user elects
to update its contents via a resynchronization.
Production Host
Training
Training 1
Replica 2
Replica 3
Replica 4
Production Replica 5
Information Replica 6
Replica 7
Tape Replica 8
Backup Replica 1
Replica 2
Production
Report Information
Generation
Replica 3
Decision
Support Tools
Data Access
Users can access BCVs or snapshots from a secondary server. BCVs can be accessed once they are
fractured from their source, and snapshots can be accessed once theyve been associated with (activated to)
a Snap session. A distinction between BCVs and snapshots, however, is how soon after creation the replica
can be accessed. Because snapshots function on a pointer-and-copy mechanism, the processing to start a
Snap session is minimal and the Snap session data is available virtually instantaneously. BCVs on the
other hand, being full-image copies, must have sufficient time for the initial copy (referred to as
synchronization by the software) to occur before they can be used for the first time. This paper does not
provide in-depth performance information, but for the purpose of comparison with starting Snap sessions as
described earlier, LUN synchronizations on a CX700 can run as quickly as 75 MB/s for a single BCV.
Once the initial synchronization has been performed, subsequent updates (resynchronizations) will be
incremental. The rate of data transfer will be comparable, but the amount needing to be transferred will
only be that data which changed on the source while the BCV was fractured.
In consideration of data availability, it should be noted that creating the snapshot (or, more specifically,
when starting the session to which the snapshot will be activated) or fracturing the BCV from the source, is
what defines the point-in-time characteristic of the replica. As such, the application writing to the source
volume must have logical consistency. Typically this is done by using the online quiesce capability if the
application has that capability. If not, the application must be stopped (and host buffers flushed to disk)
while starting the session or fracturing the BCV, and then restarted once the session has been started or the
BCV fractured.
Finally, on the note of access to replicas, it should be noted that general EMC recommendation is that users
avoid accessing replicas from the production server (or, likewise, that users not access replicas of the same
source from the same server). This recommendation stems from the fact that unless drive signatures are
modified correctly, data corruption may occur. The only exception to this restriction is if users are using
EMC PowerVolume (available with PowerPath 4.x), and/or a member of the EMC Replication
Manager family.
Resource Requirements
Since BCVs are full image copies, they require the same amount of usable disk space as the source volume.
BCVs do not have to be the same RAID type or drive type as their source, however. Consequently, users
may opt to put production data on RAID 1/0 drives for maximum protection, whereas they may opt to put
BCVs on RAID 5 drives to decrease the relative drive cost for BCVs. Furthermore, users may opt to put
BCVs on ATA drives to for additional cost savings. In this way, users can leverage the flexibility of
CLARiiON to provide tiered storage and protection in their environment.
The pointer-and-copy-based design for snapshots minimizes the amount of space needed to support these
replicas. Since only the data that is changed during the snapshot session must be accommodatedthat is,
copied to the reserved LUN2 the reserved LUN needs only to be large enough to hold that changed data.
In other words, instead of requiring another LUN that is the same size as the source LUN (as BCVs do),
snapshots require only a fraction of additional space. How much space is required will vary between
implementations, but depends on variables such as the rate of change of the source LUN data, how many
session(s) will run, and for how long. A conservative size would be 20 to 30 percent of the source LUN for
a single session, and then increasing that figure by 10 to 20 percent for each additional session. Although
these estimates are only general guidelines, they should nevertheless demonstrate how snapshots will
typically require significantly less disk capacity than BCVs. Also, as with BCVs, users may opt to put the
reserved LUNs on ATA drives for greater cost savings.
Performance Considerations
Users may find that the difference between BCVs and snapshots in terms of performance may help clarify
which replica is right for their environment.
2
The private LUNs contain original data to maintain the point-in-time view of the session. Reserved LUNs
are assigned on a per-source-LUN basis, such that source LUNs have a one-to-many relationship to their
reserved LUNs. The design of reserved LUNs allows for additional reserved LUNs to be assigned to
sources when needed (provided that extra reserved LUNs are available in the reserved LUN pool).
SnapView Implementation
Managing Replicas
SnapView tasks can be managed via Navisphere Manager graphical user interface (GUI) or Navisphere
command line interface (CLI)3. In some cases, users may find the Navisphere Manager interface easier for
initial configuration of SnapView objects and resources (such as first creating the source-to-BCV
relationship). Subsequently, users will likely find the CLI more appropriate for ongoing management of
replicas (such as managing replicas with backup jobs).
Additionally, the Replication Manager family of products helps users integrate their use of SnapView
replicas with server applications such as Microsoft Exchange, Microsoft SQL, and Oracle. Replication
Manager also allows users to coordinate replication operations with backup and recovery operations,
allowing for greater flexibility and, at the same time, greater simplicity. For more information on
Replication Manager products, refer to the white paper Introduction to Replication Manager Technology.
White papers discussing application-specific considerations with Replication Manager are also available.
3
admsnap is a server-based utility that facilitates management of SnapView objects at the server level, such
as bringing snapshots online.
Number of Replicas
As noted earlier, SnapView allows users to have up to eight snapshots and/or eight BCVs per source.
Table 1 provides a more complete listing of the allowed replicas per source and per storage system. For the
most up-to-date SnapView limits information, refer to the most recent version of the CLARiiON Open
Systems Configuration Guide.
Data Recovery
In addition to providing point-in-time views of data for concurrent access, SnapView point-in-time replicas
provide users a way to recover data in the event of a data corruption on the source LUN. The data from
BCVs or Snap sessions can be used to restore the contents of a source LUN virtually instantaneously to the
point in time associated with that replica.
These restore operations function independently of any of the other SnapView replicas that may be defined
on a source LUN. Therefore, it is completely nondestructive to the source LUNs other replicas. In other
words, if a user has any given mix of BCVs and SnapView sessions, the data relative to each replica is not
affected by the restore operation.
As shown in Figure 2, a restore operation is started using the Tuesday replica, while the other sessions
continue to operate without interruption.
Tuesday Friday
X
view ... view
reference that is likely different from the source; so any source data that has not yet been updated to the
BCV would be lost.
Conclusion
SnapView BCVs provide parallel data access facility with low impact to production data access. This
capability significantly reduces the time production environments would normally have to be offline for
backups or decision-support-type operations. This accelerates business processes and increases
productivity.
With the option of snapshots, users can have immediate access to replicas (which require a fraction of the
source LUN capacity),such as for training purposes or, in some cases, backup operations. On the other
hand, users may opt to use independent full copies of data that provide an additional level of protection
with better performance, such as for backup operations, or running reports, or data warehousing.
In addition to providing for concurrent access to data, SnapView provides the ability to instantly restore
source data in the event of a data corruption. Any SnapView replica (snapshot or BCV) can be used to
perform the restore operation.
SnapView is part of EMCs industry-leading replication family.. EMCs replication products are designed
to provide users the highest levels of flexibility combined with functionality. SnapViews dual-feature
capabilities have benefited customers in all industries by providing users the tools to fit their specific
business application needs.
Appendix
The charts in this appendix provide performance information for BCVs and snapshots. The information is
intended to provide a conceptual understanding of the differences among various performance parameters,
rather than any specific sizing or performance estimates. For specific sizing and performance estimates,
users should consult their EMC account team.
BCV Performance
As noted earlier, users can select among high, medium, and low sync rates. In some situations, users might
select the high synchronization rate so that they can synchronize their BCV as quickly as possible; whereas
in other cases, users might select the low or medium synchronization rate to free up SP resources to
perform other tasks while the synchronization takes place. The tradeoff is between how quickly the BCV
needs to be synchronized, and what other tasks are taking place during the synchronization.
Figure 3 shows the bandwidth at each synchronization rate. ATA performance is shown as well.
FC
High Sync Rate
MB/Sec
ATA
High Sync Rate
FC& ATA
Medium Sync Rate
FC & ATA
Low Sync Rate
1 2 3 4
Number of Target LUNs
ATA
ATA
BCV Operations
Snapshot Performance
As with BCVs, users must consider the impact of snapshots and Snap sessions on the source LUN. Figure
5 compares the effect on OLTP response time with multiple Snap sessions running on a source.
Additionally, the results compare the performance of reserved LUNs bound on Fibre Channel drives with
those bound on ATA drives.
ATA ATA
ATA
Average Response Time (ms)
Fibre
ATA Fibre
ATA
Fibre ATA
Fibre
ATA ATA Fibre 719
Fibre Fibre
Fibre
1 2 3 4 5 6 7 8
Number of Sessions
Performance Conclusions
To conclude, SnapView BCVs and snapshots essentially offer users a tradeoff between space capacity and
performance. Likewise, users can select between Fibre Channel and ATA drives, again depending on the
price-to-performance balance that is appropriate for the data that is under consideration. This performance
data is provided for the purpose of helping users make educated decisions about the replication options that
best suit their environment.