1. technology and computing

White Paper
Cisco UCS Invicta OS:
Data Reduction for
Scalable Flash Memory
Arrays
What You Will Learn
Flexible and effective data reduction helps you get the best performance, capacity, and value from scalable flash memory arrays.
Deduplication and thin provisioning both offer ways to limit space consumption and extend longevity for flash memory devices.
Cisco UCS Invicta™ OS offers innovative, optional inline deduplication for Cisco UCS Invicta Series Solid State Systems. With
deduplication rates of up to 10:1, this technology can greatly reduce the cost of flash memory arrays, while extending their
useful lifetime. And Cisco® data reduction technology doesn’t negatively affect performance for those applications that don’t
benefit from deduplication.
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
Contents
Introduction......................................................................................................3
Data Reduction and Flash Technology..............................................................4
Unique Considerations for Flash Technology.........................................................4
Traditional Deduplication........................................................................................4
Cisco’s Unique Approach to Deduplication............................................................5
Cisco UCS Invicta Data Reduction Overview.....................................................6
Deduplication in Cisco UCS Invicta OS..................................................................6
Error Detection and Correction..............................................................................8
Thin Provisioning and Space Management............................................................8
Conclusion.......................................................................................................9
For More Information........................................................................................9
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Page 2
Cisco UCS Invicta OS:
Data Reduction for Scalable Flash
Memory Arrays
White Paper
February 2015
Introduction
Powered by Cisco UCS Invicta OS, Cisco UCS Invicta Series Solid State Systems
deliver powerful and easy-to-manage application acceleration for I/O- and
bandwidth-intensive workloads. Designed to work with the industry-leading Cisco
Unified Computing System™ (Cisco UCS®), the solution represents a unique and
scalable application of flash memory technology. Applications ranging from email
and databases to virtual desktop infrastructure (VDI) can all benefit from Cisco UCS
Invicta Series Solid State Systems.
Although flash memory technology can accelerate workloads, you need to use care
to apply effective data reduction techniques to make the most of flash capabilities:
• Capacity must be used effectively without needless waste to help ensure value.
• The longevity of flash devices must be enhanced with data reduction techniques.
• Data integrity must be preserved through effective error correction.
• Architecture must capitalize on the speed of flash memory without needlessly
encumbering it.
Cisco UCS Invicta OS was designed from the outset to implement effective data
reduction strategies in concert with the unique properties of flash memory. Unlike
with other solutions that mandate deduplication, however, organizations deploying
Cisco UCS Invicta Solid State Systems have a choice of assigning workloads to
either performance or deduplication nodes within a Cisco UCS Invicta Scaling
System. Alternatively, organizations can choose whether to focus an individual Cisco
UCS Invicta C3124SA Appliance on performance or deduplication according to
their application needs. Cisco’s deduplication technology also enables data integrity
through automatic error correction, and Cisco’s thin provisioning lets organizations
get the most from their flash memory investment.
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information. Page 3
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
Data Reduction and Flash Technology
Cisco UCS Invicta OS provides data reduction technologies that extend beyond
those typically provided for flash storage technology.
Unique Considerations for Flash Technology
Data reduction technologies are particularly important for flash-based arrays.
Beyond reducing the amount of storage required, data reduction ultimately results
in a lower number of write operations for the media, and correspondingly greater
longevity for flash devices.
• Deduplication: Through deduplication, only unique segments are written to
the media, and write requests for duplicate data result only in the storage of a
pointer. This approach eliminates data redundancy at the block level and reduces
the amount of data written to the disk. Physical storage capacity is effectively
increased, allowing commitment of more logically mapped data than is physically
available on the device. In addition, the lower number of write operations
increases the longevity of the flash media.
• Thin provisioning: Thin provisioning is a data reduction approach that provides
the logical space requested on the device using only the physical space that
is actually required. By not consuming requested unused space in a volume,
thin provisioning allows better utilization of flash media. Storage can be
overcommitted so that more logically mapped data can be committed than is
physically available on the media.
Both of these data reduction approaches require special care when applying them
to flash technology.
Traditional Deduplication
Traditional deduplication can take a variety of forms, and it is implemented differently
by different vendors. Data can be deduplicated at the source (where the data is
created) or at the target (where the data is stored). It can be deduplicated as it
enters the device (inline) or after it is stored (postprocessing). Data can also be
deduplicated in fixed data segments (fixed block) or variable-length segments
determined by an algorithm (variable block).
Mainly associated with backup storage, traditional deduplication is often seen as a
significant performance detractor. The hashing algorithms used by deduplication
can be very CPU intensive, and many systems cannot hold in memory the metadata
used by the data reduction operations used to manage the underlying media. In
addition, locality issues with hard-disk-drive (HDD) media and deduplication can
result in significant I/O overhead for the system.
Moreover, deduplication is not a benefit for every workload. It is most effective for
cases in which redundant data is likely, such as persistent virtual desktops, system
volumes for virtual servers, and unstructured data stores. Deduplication is least
effective for cases in which relatively unique data is stored, such as databases,
video, encrypted volumes, and swap spaces. Poor deduplication rates needlessly
encumber application performance and should be avoided (Table 1).
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Page 4
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
Table 1. Good and Poor Candidates for Deduplication
Good Deduplication Candidates
Poor Deduplication Candidates
• VDI (persistent virtual desktops)
• Server virtualization (system volumes)
• Unstructured data stores
•
•
•
•
Databases
Video
Encrypted volumes
Swap space
Cisco’s Unique Approach to Deduplication
Unfortunately, many vendors support an all-or-nothing approach to data
deduplication for flash memory, requiring that it be enabled for an entire system if it
is enabled at all. This limitation makes it difficult for organizations to support multiple
applications within the same storage system in a flexible way without reducing
performance for applications for which deduplication is not as effective.
To address this challenge, Cisco UCS Invicta OS supports a flexible combination
of deduplication and performance nodes within the Cisco UCS Invicta Scaling
System. In addition, an individual Cisco UCS Invicta appliance can be configured
as either a deduplication or a performance appliance, according to the capability of
deduplication to benefit the application.
With Cisco UCS Invicta OS, deduplication is licensed on a per-node or perappliance basis. Organizations can choose to deploy a specific mix of performance
and deduplication nodes within the Cisco UCS Invicta Scaling System cluster to
meet their actual application requirements. In this manner, Cisco UCS Invicta OS
doesn’t force a performance penalty on applications with data that doesn’t benefit
from deduplication. Organizations also don’t incur unnecessary licensing costs for
that data.
Cisco UCS Invicta OS deduplication is performed inline and doesn’t require
computing or memory resources on the flash drives. Not only does this approach
preserve the performance characteristics of flash memory, but the hash table is
not constrained by relatively meager flash-drive memory resources. The resulting
deduplication rate provided by Cisco UCS Invicta OS can be up to 10:1, resulting
in up to a tenfold cost-per-bit reduction (assuming that the data can be reduced at
that rate).
The deduplication ratio is the ratio of logical write operations to physical write
operations, and it is visible in the GUI (Release 5.0.1.2; Figure 1). Through the GUI,
storage administrators can determine the effectiveness of deduplication for their
data and make appropriate decisions about storage strategies. The GUI also shows
physical space consumption, which is critical in managing flash-based storage.
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Page 5
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
Figure 1. Cisco UCS Invicta OS GUI Shows Administrators How Well Data Is Being
Deduplicated
Cisco UCS Invicta Data Reduction Overview
Cisco UCS Invicta OS data reduction is designed for performance, efficiency, and
data integrity in the context of scalable flash memory array technology.
Deduplication in Cisco UCS Invicta OS
Cisco UCS Invicta OS deduplication is performed inline, before data is even placed
into the write block, employing a very fast CPU-efficient hash algorithm. A fixedblock hash is run using a 4-KB block size for a balance between the number of I/O
operations per second (IOPS) and latency. The 4-KB size means that deduplication
is applied to even small changes to large files. Larger block sizes would increase the
minimum latency per I/O operation. The hash physical block address (PBA) is stored
in system memory, eliminating locality issues.
Collision handling is an important innovation in Cisco UCS Invicta OS, implemented
to achieve data integrity. Collisions occur when two distinct pieces of data generate
the same hash value. To cope with this issue, the system doesn’t trust a mere hash
match as absolute. Instead, it completely reads the on-disk block to confirm every
hash match and verify that duplicate data actually exists. Read operations from
the media have little impact on overall performance due to the system’s high read
speeds, and this approach achieves nearly 100 percent data integrity.
Cisco UCS Invicta OS employs a very high-performance Murmur3 hash function.
The block translation layer (BTL) of Cisco UCS Invicta OS calculates a 128-bit hash
for data. Thirty-one bits of that hash are stored in memory tables, with the complete
128-bit hash stored on disk. Collision lists are maintained for 31 bits of the 128-bit
hashes. Data is always checked byte by byte before deduplication is declared.
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Page 6
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
When deduplication is enabled, the BTL experiences a performance impact; only
write operations are affected, with read operations unaffected. The smaller hash
boundary presents an effective trade-off, because more collisions will occur
and more time will be required to identify duplicate data. However, the overall
performance of the deduplication engine is improved with smaller hash scale.
Figure 2 illustrates the Cisco UCS Invicta OS deduplication process, which is
designed to efficiently handle original, duplicate, and updated data. Every inbound
write is hashed to a 4-KB boundary. The hash is confirmed as either new or
duplicate or as a collision using the hash array index, resulting in the following
assessments of the data itself:
• Original data: In the case of new or original data, the hash position is null, and the
hash engine determines that there is no match in the hash table. All the data is
then written to the media.
Original File
Duplicate
Solution Brief
Updated
Solution Brief
January 2015
March
2014
Solution Brief
January 2015
March
2014
Highlights
January 2015
March
2014
Highlights
Change the Way You Do Business
• We change the way you do business.
Integrating Cisco UCS Invicta™
Series Solid-State Systems with the
Cisco Unified Computing System™
(Cisco UCS®) makes virtual desktops
perform better and delivers a better
user experience.
Simplified and Scalable Solutions
• Cisco’s simplified and scalable
solutions provide broad architectural
support for virtual desktops.
High Performance
• Achieve fast performance for virtual
desktops and applications with lower
costs.
Outstanding Capacity
• Get outstanding capacity and
economy with up to 10x deduplication
efficiency.
Integrated into Cisco UCS
• Gain all the value of Cisco UCS with
accelerated virtual desktop access.
Comprehensive Infrastructure
Management
• Simplify and manage the entire
solution with Cisco UCS Director.
Users are more and more demanding. They expect technology to respond
immediately and flawlessly to their every action and encounter. Using traditional
architecture to provide a high level of user access and efficiency, virtual desktop
infrastructure (VDI) deployments have been slowed by several major obstacles. High
acquisition costs, management complexity, improper planning and sizing: all of these
challenges get in the way.
We deliver better performance at affordable costs. We integrate Cisco UCS Invicta™
Series Solid-State Systems with Cisco Unified Computing System™ (Cisco UCS®)
for greater user satisfaction and business efficiency. Your users are no longer
frustrated—and you exceed business expectations while operating within budget.
In virtual desktop deployments, storage is the most challenging element. Small and
random write requests push traditional hard-disk-based storage to the limit—even
during normal operations. This phenomenon is exacerbated during startup or login
storms, which occur at the beginning of the workday. With traditional storage, virtual
desktop users often start their day frustrated with the lack of responsiveness.
Inbound
Data
While solid-state memory performance is commonly acknowledged to be
measurably superior to that of spinning disks, the importance of properly integrating
solid-state memory into the overall virtual desktop solution is less well known.
Cisco UCS with the Cisco UCS Invicta Series accelerates VDI to efficiently handle
high-demand virtual desktops and applications with reduced capital and operating
expenses.
© 2014-2015 Cisco Systems, Inc. All rights reserved. This document is Cisco Public Information.
Highlights
Change the Way You Do Business
• We change the way you do business.
Integrating Cisco UCS Invicta™
Series Solid-State Systems with the
Cisco Unified Computing System™
(Cisco UCS®) makes virtual desktops
perform better and delivers a better
user experience.
Simplified and Scalable Solutions
• Cisco’s simplified and scalable
solutions provide broad architectural
support for virtual desktops.
High Performance
• Achieve fast performance for virtual
desktops and applications with lower
costs.
Outstanding Capacity
• Get outstanding capacity and
economy with up to 10x deduplication
efficiency.
Integrated into Cisco UCS
• Gain all the value of Cisco UCS with
accelerated virtual desktop access.
Comprehensive Infrastructure
Management
• Simplify and manage the entire
solution with Cisco UCS Director.
Change the Way You Do Business
• We change the way you do business.
Integrating Cisco UCS Invicta™
Series Solid-State Systems with the
Cisco Unified Computing System™
(Cisco UCS®) makes virtual desktops
perform better and delivers a better
user experience.
Users are more and more demanding. They expect technology to respond
immediately and flawlessly to their every action and encounter. Using traditional
architecture to provide a high level of user access and efficiency, virtual desktop
infrastructure (VDI) deployments have been slowed by several major obstacles. High
acquisition costs, management complexity, improper planning and sizing: all of these
challenges get in the way.
Simplified and Scalable Solutions
• Cisco’s simplified and scalable
solutions provide broad architectural
support for virtual desktops.
We deliver better performance at affordable costs. We integrate Cisco UCS Invicta™
Series Solid-State Systems with Cisco Unified Computing System™ (Cisco UCS®)
for greater user satisfaction and business efficiency. Your users are no longer
frustrated—and you exceed business expectations while operating within budget.
High Performance
• Achieve fast performance for virtual
desktops and applications with lower
costs.
In virtual desktop deployments, storage is the most challenging element. Small and
random write requests push traditional hard-disk-based storage to the limit—even
during normal operations. This phenomenon is exacerbated during startup or login
storms, which occur at the beginning of the workday. With traditional storage, virtual
desktop users often start their day frustrated with the lack of responsiveness.
4-KB
Write Operations
While solid-state memory performance is commonly acknowledged to be
measurably superior to that of spinning disks, the importance of properly integrating
solid-state memory into the overall virtual desktop solution is less well known.
Cisco UCS with the Cisco UCS Invicta Series accelerates VDI to efficiently handle
high-demand virtual desktops and applications with reduced capital and operating
expenses.
© 2014-2015 Cisco Systems, Inc. All rights reserved. This document is Cisco Public Information.
Outstanding Capacity
• Get outstanding capacity and
economy with up to 10x deduplication
efficiency.
Integrated into Cisco UCS
• Gain all the value of Cisco UCS with
accelerated virtual desktop access.
Comprehensive Infrastructure
Management
• Simplify and manage the entire
solution with Cisco UCS Director.
Users are more and more demanding. They expect technology to respond
immediately and flawlessly to their every action and encounter. Using traditional
architecture to provide a high level of user access and efficiency, virtual desktop
infrastructure (VDI) deployments have been slowed by several major obstacles. High
acquisition costs, management complexity, improper planning and sizing: all of these
challenges get in the way.
We deliver better performance at affordable costs. We integrate Cisco UCS Invicta™
Series Solid-State Systems with Cisco Unified Computing System™ (Cisco UCS®)
for greater user satisfaction and business efficiency. Your users are no longer
frustrated—and you exceed business expectations while operating within budget.
In virtual desktop deployments, storage is the most challenging element. Small and
random write requests push traditional hard-disk-based storage to the limit—even
during normal operations. This phenomenon is exacerbated during startup or login
storms, which occur at the beginning of the workday. With traditional storage, virtual
desktop users often start their day frustrated with the lack of responsiveness.
While solid-state memory performance is commonly acknowledged to be
measurably superior to that of spinning disks, the importance of properly integrating
solid-state memory into the overall virtual desktop solution is less well known.
Cisco UCS with the Cisco UCS Invicta Series accelerates VDI to efficiently handle
high-demand virtual desktops and applications with reduced capital and operating
expenses.
© 2014-2015 Cisco Systems, Inc. All rights reserved. This document is Cisco Public Information.
HASH ALGORITHM
1 2
3 4
Hash Data
1 2
1 5
3 4
3 4
DEDUP
ENGINE
Determines whether the hash already exists in the in-memory hash table
NO MATCH
MATCH
Existing data
is read to
confirm
duplication
MATCH
NO
MATCH
Pointers to the existing data
are stored
All Data
Written
Pointers
Written
Update
Written
Data is sorted into write blocks and sent to media
Figure 2. The Cisco UCS Invicta OS Data Deduplication Process Provides Fine-Grained
Deduplication with Effective Collision Detection and Data Integrity
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Page 7
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
• Duplicate data: If the hash array positions show a possible duplicate, a further
evaluation is made to determine whether the physical location shows the same
data (duplicate) or different data (collision). Upon confirmation of duplicate data,
pointers to the existing data are stored. Upon a collision, a mismatch forces a
unique store of incoming data.
• Updated data: The 4-KB hash boundary allows highly specific updates to existing
data. As shown in the example in Figure 2, a portion of a webpage can be
updated while retaining deduplication on the remainder of the page.
Error Detection and Correction
The deduplication algorithm in Cisco UCS Invicta OS is also used for enhanced error
detection and correction. If an optional read hash check is turned on, data is read
and hashed. The hash is then compared to the expected hash from the logical block
address (LBA) table. If the hashes don’t match, then an error-correction flow is
performed to correct the data. The error correction flow repairs the RAID 6 algorithm
that manages the flash devices, as follows:
• The process begins with an N-1 read of the RAID set while calculating parity.
• The resulting read data is hashed and compared to the expected hash.
• The system continues excluding a different RAID element until a hash match is
found.
• The excluded element contains the error.
Thin Provisioning and Space Management
Traditional provisioning (also known as thick provisioning) is inefficient and generally
not well suited for flash storage. In traditional provisioning, storage is preallocated,
and fixed physical space is allocated whether it is used or not. This approach
often traps significant unused space, resulting in poor utilization of the storage
media. Making matters worse, when a logical unit number (LUN) runs out of space,
applications can encounter performance problems and need to be brought down. It
is often difficult to add storage to the original LUN. Typically, you must create a new
LUN with more space and then restore the original LUN contents to it.
Thin provisioning generally addresses these concerns by not preallocating storage,
instead waiting until the storage is actually needed. By allocating physical space only
as it is needed, this approach allows overprovisioning, and more logical space can
be committed than the actual physical space available. For example, you can create
a 4-terabyte (TB) LUN even though only 500 GB of physical space is set aside. In
this manner, thin provisioning does not consume unused space in the volume.
Cisco UCS Invicta OS provides thin provisioning at the volume-group level. This
approach differs from traditional thin provisioning in that it does not allow the
addition of more storage capacity to the volume group if it runs out of physical
space. Running out of space can have dire effects on applications. To address this
concern, Cisco UCS Invicta OS holds space in reserve to prevent the file system
from being completely filled. If the device becomes full, the reserved storage is
unmapped and used as a workspace to move data from other LUNs. The physical
space consumption should still be monitored closely through the GUI (see Figure 1)
to help ensure that applications do not run out of physical space. To ameliorate
growing storage needs, additional volume groups can proactively be created in the
same or additional storage media and made available to the applications.
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information.
Page 8
Cisco UCS Invicta OS: Data Reduction for Scalable Flash Memory Arrays
February 2015
Conclusion
Effective data reduction is critical for flash storage such as that embodied in the
Cisco UCS Invicta Scaling System and Cisco UCS C3124SA Appliances. Cisco
understands that deduplication isn’t right for all data, so Cisco UCS Invicta OS
provides deduplication as a per-node option within a Cisco UCS Invicta Scaling
System. Individual Cisco UCS Invicta appliances can also be configured for
performance or deduplication. The resulting modularity enables the Cisco UCS
Invicta Scaling System to be built from a combination of performance- and
deduplication-enabled nodes, letting organizations tune their deployment to the
needs of their specific applications.
Cisco’s efficient and high-performance inline approach to deduplication provides
up to a 10:1 deduplication ratio with very low overhead, offering space savings,
error correction, and media longevity. The GUI provided by Cisco UCS Invicta OS
lets storage administrators see to what degree their data is being deduplicated—
feedback that can help them decide whether deduplication is appropriate and
cost effective for specific applications and data. Further enhancing flash memory
management, Cisco UCS Invicta OS thin provisioning offers good utilization of
storage media, and system checks set aside storage to avoid costly and disruptive
storage overruns. Together these data reduction technologies result in a system
that can deliver significant performance, efficiency, and data integrity for demanding
applications that need fast data access.
For More Information
For more information on Cisco UCS Invicta Solid State Systems, visit: http://www.
cisco.com/go/ucsinvicta.
Americas Headquarters
Cisco Systems, Inc.
San Jose, CA
Asia Pacific Headquarters
Cisco Systems (USA) Pte. Ltd.
Singapore
Europe Headquarters
Cisco Systems International BV Amsterdam,
The Netherlands
Cisco has more than 200 offices worldwide. Addresses, phone numbers, and fax numbers are listed on the Cisco Website at www.cisco.com/go/offices.
Cisco and the Cisco logo are trademarks or registered trademarks of Cisco and/or its affiliates in the U.S. and other countries. To view a list of Cisco trademarks, go to this
URL: www.cisco.com/go/trademarks. Third party trademarks mentioned are the property of their respective owners. The use of the word partner does not imply a partnership
relationship between Cisco and any other company. (1110R)
LE-44402-00 02/15