are you ready for the ‘ava’ generation?...monitor and manage a distributed avamar deployment....
TRANSCRIPT
EMC Proven Professional Knowledge Sharing 2010
Are You Ready for the ‘AVA’ Generation?
Randeep Singh
Randeep SinghSolution ArchitectHCL Comnet
2010 EMC Proven Professional Knowledge Sharing 2
Table of Contents
Figures………………………………………………………………………………………………….3
Summary……..…………………………………………………………………………………………4
Abstract……….…………………………………………………………………………………………4
Rise of De-Dupe Concept……………………………………………………………………………..5
Terms and Concept in Avamar………………………………………………………………………..7
Introduction of RAIN Technology……………………………………………………………………..9
Active/Passive Replication…………………………………………………………………………….10
Protection of Data in VMware Environment………………………………………………………….13
Tuning Client Cache to optimize Backup Performance…………………………………………….17
Integration with Traditional Backup Civilization………………………………………………………21
Protection of Remote Offices…………………………………………………………………………..23
Architecting Avamar to Ensure Maximum System Availability……………………………………..26
Pro-active Steps to Manage Health and Capacity…………………………………………………..27
Daily Server Maintenance……………………………………………………………………………...29
Tape Integration…………………………………………………………………………………………30
What’s New………………………………………………………………………………………………32
Why EMC Avamar………………………………………………………………………………………33
Disclaimer: The views, processes, or methodologies published in this article are those of the
author. They do not necessarily reflect EMC Corporation’s views, processes, or methodologies.
2010 EMC Proven Professional Knowledge Sharing 3
LIST OF FIGURES
Source-based Data De-Duplication ………………………………………………………………...5
Global De-Dupe Concept …………………………………………………………………………….6
Traditional versus Daily Full Backups……………………………………………………………….6
Avamar Client Plug-ins………………………………………………………………………………..8
RAIN Architecture……………………………………………………………………………………...9
Active/Passive Replication……………………………………………………………………………11
Avamar Server Deployed as a Virtual Appliance…………………………………………………..14
Backup Solutions for VMware………………………………………………………………………..15
Processes Used to Filter Out Previously Backed Up Files………………………………………..18
Effectiveness of File Cache for Backing up File Servers versus Databases…………………....18
Avamar and Networker Integration…………………………………………………………………...22
Avamar and Networker Integration (Networker Console)……………………………………….....22
Protection of Remote Offices……………………………………………………………………........24
Avamar Data Transport Components………………………………………………………………...31
2010 EMC Proven Professional Knowledge Sharing 4
SUMMARY Once when I was having a cup of coffee with my students I realized every creature on this earth
whether it’s a man, women, or child produces data. In fact, by 2010 most database sizes
crossed into double digits in terms of petabyte. But data growth is not our most significant
challenge; the main challenge arises when this data needs to be backed up with “on-demand”
24X7 accessibility. And, of course, how can we forget about the IT investment.
My students ask “Is there any way that one can minimize the backup window?” To this, I reply
“well, now it’s time to step into a new AVA generation.” Traditional backup solutions store data
repeatedly and total storage under management gets expanded by 5-10 times. Consider the
case of government regulations; the risk of shipping tapes is one of the greatest security
concerns in today’s IT infrastructure. Also these traditional backup solutions require a rotational
schedule of full and incremental backups. Due to unnecessary data movement enterprises are
often faced with backup windows that roll into production hours, network constraints, and too
much storage to manage. Tape drives, media, and libraries are prone to mechanical failure and
transport of tapes can be unreliable. While disk storage has been used to meet these
challenges, it provides a fraction of the data protection challenges faced by the organizations.
Consider the case of remote offices where the risk of data loss or exposure is extremely high.
The answer to the above challenges is Avamar®.
ABSTRACT EMC Avamar enables fast and reliable backup and recovery across the enterprise which can
include LAN/SAN servers in the data center, VMware environments, remote offices, etc. In order
to achieve this, Avamar utilizes patented global data deduplication technology to identify
redundant sub-file data segments at the source, before it is transferred across the network and
stored to disk. Only new, unique sub-file variable length data segments are moved, thereby
reducing daily network bandwidth and storage by up to 500x. Moreover for security, we can
encrypt data both at rest as well as in flight. For remote offices, Avamar uses centralized
management which protects hundreds of remote offices quite easily and efficiently. Total back-
end storage is also reduced by nearly 50x as Avamar stores only a single instance of each sub-
file data segment. It also utilizes patented RAIN technology to provide fault tolerance across
nodes and eliminate single points of failure. To achieve scalability, Avamar has scalable grid
architecture. It can also work with existing traditional tape solutions such as NetWorker®.
2010 EMC Proven Professional Knowledge Sharing 5
RISE OF DE-DUPE CONCEPT
Looking at the data of enterprises today, we find that data is highly redundant with lots of
identical files and data stored within and across systems. Performing backup using traditional
backup software will store this redundant data again and again. This is where the concept of
deduplication that Avamar uses comes into play, i.e. global, source data deduplication that
eliminates redundancy at both the file and sub-file data segment level. The challenge of
redundancy is solved in backup at the source before it is transferred over the network, whether
LAN or WAN. To ensure that each unique data segment is backed up only once across the
enterprise, Avamar agents are deployed on the systems to be protected that identify and filter
repeated data segments stored across the enterprise over time. As a result, only a small
amount of incremental backup data is generated, whether it is file system or database data.
EMC Avamar uses an intelligent variable length method for determining segment size that looks
at the data itself to determine logical boundary points. This is opposite to the fixed block or fixed
length segment method that traditional backups use, where even a small change in a data set
can change all fixed length segments in a data set. The algorithm used in Avamar analyzes the
binary structure of a data set, i.e. 0s and 1s, which make up a data set to determine segment
boundaries, using which Avamar client agents will be able to identify the exact same segments
for any data set. A variable length segment size averages about 24 KB and then compressed to
an average of 12 KB. For this 24 KB segment, a unique 20-byte ID using SHA-1 encryption
algorithm is generated by Avamar. This unique ID determines whether or not a data segment
has been previously stored.
FIGURE 1 SOURCE-BASED DATA DEDUPLICATION
2010 EMC Proven Professional Knowledge Sharing 6
Figure 1 shows the deduplication concept. Figure 2 shows how it is actually occurring, i.e.
deduplication is occurring before data is transferred across the network. This happens by
identifying the sub-file variable length data segments at the source, breaking data into atoms
and then, using deduplication, sending each atom only once, thereby reducing the data to be
backed up by up to 500 times.
FIGURE 2 GLOBAL DE-DUPE CONCEPT Assuming there is 5 TB of data to be backed up, the data reduction factor will be 100-150x if
there will be a mixture of file system and database data at source and data to be backed up at
source will remain only 35-50 GB. For Windows and UNIX, file data is reduced by nearly 300x
and data to be backed up will remain approximately 20 GB at source. Not only this, it also
results in 85 % reduction in total client CPU utilization.
Figure 3 shows the comparison of traditional backup versus Avamar daily full backups.
FIGURE 3 TRADITIONAL VS AVAMAR DAILY FULL BACKUPS
2010 EMC Proven Professional Knowledge Sharing 7
Below are some of the factors that impact data deduplication ratios:
Type of data
• Duplication in user-generated data is greater than from natural sources
• Encrypted and compressed data are not ideal candidates for deduplication
• More user-created content = higher deduplication ratio
Data change rate
• Small data change rates = more duplicate data in subsequent backups
• Less change = higher deduplication ratio
Retention policy
• Longer retention increases chances data will be repeatedly backed up
• Longer retention policy = higher deduplication ratio
Ratio of full backups to incremental backups
• More full backups increase the amount of data being repeatedly backed up
• More full backups = higher deduplication ratio
TERMS AND CONCEPTS OF AVAMAR Let us begin by learning some terms: Avamar Servers – Logical grouping of one or more nodes used to store and manage client
backups.
Node – Self-contained, rack-mountable network-addressable computer that runs Avamar server
software on the Linux operating system. It is the primary building block in any Avamar server.
Nodes can be any of the following types:
Utility Node – Dedicated to scheduling and managing background Avamar server jobs. In
scalable multi-node Avamar servers, a single utility node provides essential internal services for
the server which can be Management Console Server (MCS), cronjob, External authentication,
Network Time Protocol (NTP), and Web access.
Storage Nodes – store the actual backup data. Multiple storage nodes are configured with
multi-node Avamar servers based upon performance and capacity requirements. Storage nodes
can be added to an Avamar server over time to expand performance with no downtime required.
Avamar clients connect directly with Avamar storage nodes; client connections and data are
load balanced across storage nodes.
2010 EMC Proven Professional Knowledge Sharing 8
NDMP Accelerator node – specialized node that utilizes NDMP in order to provide data
protection for NAS filers (for example, EMC Celerra®, Network Appliance) and Novell NetWare
servers.
Access Nodes – used to run the Avamar File system
Single Node Servers – combines all the features and functions of utility and storage nodes in a
single node.
Hard Disk Storage – Disks used for storage.
Stripes – Unit of disk drive space managed by Avamar to ensure fault tolerance.
Object – Single instance of deduplicated data. Each Avamar object inherently has a unique ID.
Objects are stored and managed within stripes on the Avamar server.
Avamar Administrator – graphical management console software application that is used to
remotely administer an Avamar system from a supported Windows client computer.
Avamar Enterprise Manager – Web-based management interface that provides the ability to
monitor and manage a distributed Avamar deployment.
Avamar Clients – filter out redundant data before sending data over networks, making it
possible to protect systems even over congested LANs or WANs.
FIGURE 4 AVAMAR CLIENT PLUG-INS Agents – Platform-specific software processes running on the client that communicates with the
management console server (MCS) and with any plugins installed on that client. There are two
types of plugins supported.
File system Plugins – used to browse, back up, and restore files or directories on a specific
client file system which can be HP-UX, IBM AIX, Linux, MAC OS X, Microsoft Windows (2000,
2003, 2008, XP, and Vista), Sun Solaris, Novell Netware, and VMware.
Application Plugins – support backup and restore of databases or other special application.
Supported application plugins include; DB2, NDMP for NAS filers, Microsoft Exchange,
Microsoft SQL Server, and Oracle databases.
2010 EMC Proven Professional Knowledge Sharing 9
Avamar Replicator – enables efficient, encrypted, and asynchronous replication of data stored
in an Avamar server to another Avamar server deployed in remote locations without the need to
ship tapes.
INTRODUCTION OF RAIN TECHNOLOGY Assume that your traditional backup solution fails. What will you do? Avamar uses a new
concept of RAIN (Redundant Array of Independent Nodes) technology that provides failover and
fault tolerance across the nodes in an Avamar server grid. If a server node fails or becomes
unavailable, then data stored on any node can be reconstructed from any other node, thus
providing reliable data protection and access. In addition to providing failsafe redundancy, RAIN
can be used when rebalancing the capacity across the nodes once we have expanded the
Avamar server by adding nodes. Through this, we will be able to manage the capacity of the
system as the amount of data added to the system continues to grow. Figure 5 shows the RAIN
architecture.
FIGURE 5 RAIN ARCHITECTURE Operational best practices for using RAIN technology in Avamar
• Always enable RAIN for any configuration other than single-node servers and 1x2’s (two
active data nodes). Minimum RAIN configuration is a 1x4 (four active data nodes plus a
utility node and spare node). Double-disk failures on a node or a complete RAID
controller failure can occur, which can corrupt the data on a node. Without RAIN, the
2010 EMC Proven Professional Knowledge Sharing 10
only recourse is to reinitialize the entire system and replicate the data from the
replication target.
• When deploying non-RAIN servers, we must replicate the data on these servers to
ensure the data is protected. Non-RAIN servers have no data redundancy and so, any
loss of data can have significant and widespread impacts to the backups stored on that
Avamar server.
• When deploying a RAIN configuration, always include an active spare node in the
module. The best way to minimize system downtime is to have a spare node readily
available.
• For performance reasons, Avamar recommends limiting the number of nodes to 16
active data nodes (1x16 configuration). Limit starting configurations to 12 to 14 active
data nodes so that nodes can be added later if needed to recover from high capacity
utilization situations.
ACTIVE/PASSIVE REPLICATION PHASE Replication transfers data from a source Avamar server to a destination Avamar server. The
data replicated to the destination server can be directly restored back to primary storage without
having to be staged through the source Avamar server. Replication is accomplished by way of
highly efficient, asynchronous Internet Protocol (IP) data transfers, which can be scheduled
during off-peak hours to optimize use of network bandwidth. Replication also uses data
deduplication technology that finds and eliminates redundant sequences of data before it is sent
to the destination server, thereby reducing network traffic and promoting efficient use of hard
disk storage.
Using replication, an enterprise can centrally protect and manage multiple remote branch offices
that are using individual single-node servers for local backup and restore. The centralized multi-
node server can then be used for disaster recovery in the event of catastrophic data loss at any
remote branch office. Replication can also be used to replicate data stored in a multi-node
server to any other multi-node server in an enterprise, enabling multi-node servers to provide
peer-to-peer disaster recovery for each other.
If the primary server and the replication target are configured in an active-passive configuration,
then with one change to the DNS entry and, with a few commands, all clients can continue daily
backups to the replication target as shown in Figure 6. In this example of failover, clients still
back up to host name BACKUP_SRVR, but in the DNS server, BACKUP_SRVR is associated
with IP address 192.168.200.1, the utility node of REPL_DST.
2010 EMC Proven Professional Knowledge Sharing 11
Replication is useful for more than recovering from the traditional disaster associated with a site
failure. Replication is the most reliable form of redundancy that the system can offer because it
creates a logical copy of the data from the replication source to the destination. It does not
create a physical copy of the blocks of data, and therefore, any corruptions, whether due to
hardware or software, are far less likely to be propagated from one Avamar server to another.
Also during replication, multiple checks of the data occur to ensure that only uncorrupted data is
replicated to the replication target.
Figure 6 ACTIVE/PASSIVE REPLICATION
There are two kinds of Replication:
• Normal Replication
• Full root-to-root replication.
In Normal replication, user data on the source Avamar server is replicated to a destination
Avamar server. A special “REPLICATE” domain on the destination server is created during the
first replication operation. This domain contains a mirrored representation of the entire source
2010 EMC Proven Professional Knowledge Sharing 12
server client tree on the destination server. All data within the REPLICATE domain is read-only.
The only operations allowed on these backups are:
• Redirected restores to other clients not within the REPLICATE domain.
• Changing a backup expiration date.
• Validating backups on other clients not within the REPLICATE domain.
• Viewing backup statistics.
• Deleting a backup.
Full “root-to-root” replication creates a complete logical copy of an entire source server on the
destination server. Replicated data is not copied to the REPLICATE domain; it is added directly
to the root domain just as if source clients had registered with the destination server and source
server data replicated is fully modifiable on the destination server.
To fully replicate an Avamar server, all of the following data must be copied from the source
server to the destination server during each replication operation:
• Client backups
• Domains, clients, and users
• Groups, datasets, schedules, and retention policies
• State of the server-like contents of the activity monitor and server monitor databases at
the time of the last MCS backup or flush.
LIMITATIONS of REPLICATION:
• Only static data is replicated, i.e. data that is quiescent on source server gets replicated.
Thus, any operation that writes data to the source server and is not completed yet or
running will not be replicated to the destination server during current replication
operation. No doubt, it will be replicated during the next replication operation.
• Avamar Administrator can only manage one server at a time. In an environment with
more than one server, Avamar Enterprise Manager multi-system management console
is used.
BEST PRACTICES FOR REPLICATION:
• Protect the data on the Avamar server by replicating the data to another Avamar server.
• Set up active-passive replication pairs to take advantage of failover capability, i.e. one
Avamar server (active) is dedicated for taking backup of clients and the other Avamar
server (passive) is the dedicated replication target for backup source server. If the active
2010 EMC Proven Professional Knowledge Sharing 13
server fails, then by changing just one DNS entry, all clients will begin backing up to the
passive server as described in the previous image.
• Ensure that available bandwidth is sufficient to replicate all of the daily changed data
within a 4-hour window so that the system can accommodate peaks of up to 8 hours per
day. If using daily replication, avoid using --include flag option as every time a new client
is added to the Avamar system, client data is not replicated until editing repl_cron file to
add --include option for new client. In other words, every time we add a new client to the
active Avamar server, the client data is not replicated unless we edit the repl_cron.cfg
file to add a new --include option for that client. Also, a system that specifies the --
include option might not include the MC_BACKUPS data. That means none of the
policies are then replicated.
• For best results, ensure that the target server is running either the same or later version
of Avamar software as the source Avamar server.
• Use a large time-out setting for initial replication. Recent backups might not get
replicated as replication processes can time out before all backups are successfully
replicated. This is because replication always replicates data backups alphabetically by
client name and earliest backups before later backups.
• Normal source server maintenance tasks such as hfs check and garbage collection
should not be running during the replication session.
• Schedule back ups during period of low backup activity as this will ensure that the
greatest number of client backups will be replicated during each replication session.
PROTECTION OF DATA IN VMWARE ENVIRONMENT There are many benefits of VMware view to IT organizations, but as the amount of data stored
on virtual machines increases, traditional backup finds it difficult to back up such environments.
EMC Avamar delivers fast and efficient protection for such environments by deduplicating data
at the source, thereby sending only unique, sub-file variable length data segments during daily
full backups. Traditional backup solutions will move approximately 200% of primary data on a
weekly basis. In comparison, Avamar, utilizing deduplication, moves approximately 2%. Results
are:
• Up to 95% reduction in data moved
• Up to 90% reduction in backup times
• Up to 50% reduction in disk impact
• Up to 95% reduction in NIC usage
• Up to 80% reduction in CPU usage
• Up to 50% reduction in memory usage
2010 EMC Proven Professional Knowledge Sharing 14
EMC Avamar Virtual Edition for VMware is the industry’s first deduplication virtual appliance for
backup, recovery, and disaster recovery. Avamar Virtual Edition enables users to deploy
Avamar’s deduplication technology easily, effectively, and in a repeatable fashion on VMware
ESX Server hosts. Each virtual appliance supports up to 2 TB of deduplicated backup capacity
compared to a traditional backup schedule that would require approximately 70 TB of tape or
disk storage and can leverage the existing VMware shared server and storage infrastructure to
lower costs and simplify IT management. Avamar Virtual Edition supports VMotion for
deployment flexibility, and up to two Avamar Virtual Edition virtual appliances per ESX server to
provide scalability. Replication between Avamar virtual appliances or from Avamar virtual
appliances to physical Avamar servers eliminates reliance on offsite tape shipments and the risk
of losing unencrypted data. Figure 7 shows the Avamar Server software deployed as a virtual
appliance.
FIGURE 7 Avamar Server Deployed As a Virtual Appliance Key components that need to be protected are:
• Microsoft Active Directory that provides authentication and user configuration to the
VMware view solution.
• View Manager that interacts with vCenter to manage virtual desktop instances.
• Virtual desktop templates.
• vCenter, the centralized management application for the virtual infrastructure.
• ESX Server that provides the virtualization of the desktop instances.
• User data disk is the location of unique user data that is mapped to the virtual desktop
instance upon user login and where users store their individual data.
2010 EMC Proven Professional Knowledge Sharing 15
There can be two data protection strategies for a VMware environment. The first strategy uses a combination of hardware and software to make a simultaneous copy of
the LUNs where the virtual machine desktops are stored, as well as the view manager
application and configuration information. It will allow a total recovery of the entire environment.
A requirement for using this strategy involves making a duplicate copy of the VMware view
environment for presentation to the backup solution. The second strategy involves protecting the key components of VMware view infrastructure
independently using Avamar client software agents. This will avoid the infrastructure cost we
would have incurred if using the first strategy. Figure 8 shows the Avamar client backup solutions for VMware.
Figure 8 Backup Solutions for VMware In the first solution, i.e. at Guest level, the Avamar agent resides inside each virtual machine,
deduplicates data within the virtual machines as if they are physical servers, moving minimum
data to thus reduce resource contention, accelerate backup, and provide file level restore for
Windows, Linux, and Solaris. Advantages and considerations for this method:
Advantages
• Highest level of data deduplication, resulting in maximum storage efficiency
• Guest-level backup easily fits into most existing backup schemes; day-to-day
backup procedures do not change
• Support for fast partial restores of individual directory (folder) or files
• Optional support for application-level support for DB2, Exchange, Oracle,
and SQL Server databases
• No advanced scripting or VMware knowledge required
2010 EMC Proven Professional Knowledge Sharing 16
Considerations The only significant consideration to this approach is that although file and directory restores are
a simple one-step process, full system recovery is a two-step procedure in which you first load a
known-good operating system image inside the virtual machine, then restore the unique data
from the guest-level backups stored on the Avamar server.
In the second solution, i.e. VCB, the Avamar agent resides on the proxy (ESX) server,
deduplicates within and across the VMDK files, and supports both VCB file (Windows only) and
image level backup. Avamar replication provides disaster recovery for backed up VMDKs files.
Advantages and considerations for this method:
Advantages
• Backup resources are offloaded to the VCB proxy server
• Full-image backups of running virtual machines
• File-level backups for Windows virtual machines
• Image virtual disks compressed
• Consolidation ratios optimized for VMware
Considerations
• Restores are a multi-step process
• VCB does not support iSCSI or NAS/NFS
• VCB can only back up a virtual machine with a disk image stored on a device that the
proxy can access
• VCB cannot back up virtual disks that are RDM in physical compatibility mode
• VCB can only back up a virtual machine that has an IP address or domain name server
(DNS) name associated with it
• VCB can only perform a file-level backup of a virtual machine running Microsoft
Windows NT 4.0, Windows 2000, Windows XP, Windows XP Professional, or
Windows 2003
• All backups and restores must be initiated using the Avamar Administrator
Graphical management console; it is not possible to initiate backups and restores
directly from the virtual machine interface In the third solution, i.e. at the Console level, the Avamar agent resides on the ESX console
operating system, deduplicates within and across the VMDK files, and provides VMDK-image
level (no file level) backup without VCB. No proxy server or SAN server is required for this
configuration.
2010 EMC Proven Professional Knowledge Sharing 17
Advantages and considerations for this method:
Advantages
• Virtual machine image-level backups without VCB
• No proxy or SAN required
Considerations
• No file-level backup
• Two-step restore
• Advanced configuration (without proper scheduling and setup, backup jobs can
impact system resources and running virtual machines)
• If we perform a backup by suspending or powering down a virtual machine, and then
back it up directly, the virtual machine must be powered down or suspended before
backup begins. Virtual disk images are not compressed (empty space is processed).
TUNING CLIENT CACHE TO OPTIMIZE BACKUP PERFORMANCE Whenever backup begins, the avtar process loads two cache files into memory from the var
directory that is in the Avamar installation path. The first of the cache files is the file cache
(f_cache.dat). It stores a 20-byte SHA-1 hash of the file attributes, and is used to quickly identify
which files have previously been backed up to the Avamar server. When backing up file servers,
the file cache screens out approximately 98% of the files. When backing up databases,
however, the file cache is not effective since all the files in a database appear to be modified
every day.
The second cache is the hash cache (p_cache.dat). It stores the hashes of the chunks and
composites that have been sent to the Avamar server. The hash cache is used to quickly
identify which chunks or composites have previously been backed up to the Avamar server. The
hash cache is very important when backing up databases.
These client caches are used to:
• Reduce the amount of time required to perform a backup
• Reduce the load on the Avamar client
• Reduce the load on the Avamar server
Figure 9 shows the process that avtar uses to filter out previously backed up files and chunks. It
shows the values that are incremented and reported in the avtar advanced statistics option. First
of all, it starts the file and computes SHA-1 hash of file attributes and checks whether file hash
exists in the cache. If chunk hash exists in hash cache, then it processes next chunk. If it does
not, then a request is sent to the server which checks whether the chunk hash exists on the
2010 EMC Proven Professional Knowledge Sharing 18
Avamar server. If it does, it then processes the next chunk; otherwise, it will send chunk and
chunk hash to the Avamar server.
FIGURE 9 Processes Used to Filter Out Previously Backed Up Files Also consider Figure 10 that shows the effectiveness of file cache when backing up file servers
in contrast to databases. In this example, we have considered 1 TB of file system and 1 TB of
database.
2010 EMC Proven Professional Knowledge Sharing 19
Figure 10 Effectiveness of file cache, backing up file servers vs. databases It is clear from Figure 10 that file cache provides 98% reduction in client load, approximately 980
GB of data is screened out during backup, and only around 3 GB data is sent to the Avamar
server. While in the case of databases, file cache is 0% effective in reducing client load and
around 30 GB of data is sent to the Avamar server.
By default, file cache consume up to 1/8th of the physical RAM on the Avamar client. For
example, if the client has 4 GB of RAM, the file cache will be limited to 4 GB/8, or 512 MB
maximum. The file cache doubles in size each time it needs to grow. The current file cache
sizes are 5.5 MB, 11 MB, 22 MB, 44 MB, 88 MB, 176 MB, 352 MB, 704 MB, and 1,408 MB.
The file cache includes two 20-byte SHA-1 hash entries:
• the hash of the file attributes, which include file name, file path, modification time, file
size, owner, group, and permissions.
• the hash of the actual file content, independent of the file attributes.
File cache rule is: If the client comprises N million files, the file cache must be at least N million
files x 40MB/million files. Example: If a client has four million files, the file cache must be at least
160MB (4 million files x 40MB/million files). In other words, the file cache must be allowed to
grow to 176MB.
Hash cache consume up to 1/16th of the physical RAM on the Avamar client. Example: For
clients with 4 GB of RAM, the hash cache will be limited to 4 GB/16, or 256MB maximum. The
hash cache also doubles in size each time it needs to grow. The current hash cache sizes are
24MB, 48MB, 96MB, 192MB, 384MB, 768MB, and so forth.
2010 EMC Proven Professional Knowledge Sharing 20
So, in this example where a client has 4 GB of RAM, the maximum size of the hash cache will
be 192MB. The hash cache include only one SHA-1 hash per chunk, which is the hash of the
contents of the chunk.
Hash cache rule is: If the client comprises Y GB of database data, the hash cache must be at
least Y GB/average chunk size x 20MB/million chunks. Use 16kB as the average chunk size for
streaming Exchange and Oracle database backups. Use 24kB as the average chunk size for all
other backups, including backups of database dump files.
Example: If an Exchange database client has 250GB of database data, the hash cache must
accommodate 16 million chunks (250GB/16kB). At 20MB per million chunks, this means that the
hash cache must be at least 320MB.
The amount of memory consumed by the avtar process is generally in the range of 20 to 30 MB.
This amount depends on which operating system the client is running, and also fluctuates
during the backup depending on the structure of the files that are being backed up by avtar.
Since the file cache and hash cache can grow to maximum sizes of 1/8th and 1/16th of the total
RAM in the system, respectively, you can see that, for a client that has, for example, more than
1/2 GB of RAM, the file and hash caches contribute more to the overall memory use than the
rest of the avtar process. This is because both caches are read completely into memory at the
start of the avtar backup. Also, by default, the overall memory that client caches use is limited to
approximately 3/16th of the physical RAM on the Avamar client.
Changing Client Cache Size We can override the default limits on the size of the file and hash caches by using
the following two flags:
--filecachemax=VALUE --hashcachemax=VALUE Where VALUE is an amount in MB or a fraction (negative value = fraction of RAM).
As described previously, the default values are:
--filecachemax=-8 --hashcachemax=-16 For example, the following option will limit the file cache to 100 MB. Because the file cache
doubles in size every time it needs to grow, the file cache will actually grow to 88 MB if the
following option is used:
--filecachemax=100
2010 EMC Proven Professional Knowledge Sharing 21
To optimize performance, sometimes we need to increase the cache sizes from the default
values. These situations arise if there are either millions of small files or large databases. Millions of Small Files If the client has millions of small files, then we might need to increase
the file cache from the default size. The general rule is the client requires 512 MB of physical
RAM for every one million files on the Avamar client. If a client has one million files, a minimum
of 40MB is required just to store all the file hashes for a single backup, since there are two 20-
byte entries per file. Since the file hashes must be stored for several backups, more than this is
required. As discussed previously, the file cache doubles in size each time it needs to grow. The
current file cache sizes are: 5.5 MB, 11MB, 22MB, 44MB, 88MB, 176MB, 352MB, 704MB, and
1,408MB. This means growth to about 44MB occurs. By default, since 1/8th of the physical
RAM of 512MB is used, cache can grow to a limit of 64MB, which means that the default value
of 1/8th of RAM for the file cache is adequate. Large Databases If the client has a few large files, the default of 1/16th for the hash cache is
usually insufficient. For example, for a 240GB database, a cache of up to 10 million hashes is
required. Since each hash is 20 bytes, a hash cache that is at least 200MB is required, and that
can only store the hashes for a single backup. The hash cache also doubles in size each time it
needs to grow. The current hash cache sizes are: 24MB, 48MB, 96MB, 192MB, 384MB,
768MB, and 1,536MB. The next increment available is 384MB. Therefore, if this client has 4 GB
of RAM, the hash cache must grow to 1/8th of the RAM. If the default of 1/16th of the RAM is
used, then the hash cache will be limited to 192MB, resulting in an undersized hash cache. In
the case of databases, since very few files are backed up, the file cache will be considerably
smaller, so the net memory use is still about 1/8th to 3/16th of the RAM. Operational Best Practice: Never allow the total combined cache sizes to exceed 1/4 of the
total available physical RAM. INTEGRATION WITH TRADITIONAL BACKUP CIVILIZATION A growing number of organizations are considering the benefits of data deduplication to reduce
the cost of disk-based backup and recovery. A recent ESG Research Survey indicates that 11%
of respondents are currently using data deduplication and 32% plan on doing so. Until recently,
the only way to introduce data deduplication into an existing backup and recovery environment
was to swap tape or disk-based recovery hardware for an appliance that supports data
deduplication. While these hardware-based solutions can deduplicate backup data stored on
disk, they do nothing to reduce the amount of data flowing over the network and can create
islands of deduplication. ESG Lab has confirmed that EMC NetWorker can be easily deployed,
leveraging Avamar deduplication technology to create a global pool of deduplicated data that
2010 EMC Proven Professional Knowledge Sharing 22
drastically reduces disk and network requirements while preserving the familiar look and feel of
NetWorker.
The latest release of Avamar has been integrated with the NetWorker backup product. As a
result, NetWorker users can now take advantage of Avamar data deduplication using familiar
NetWorker interfaces, workflows, and backup policies. Also, it will provide centralized control of
traditional and next-generation backup.
NetWorker clients can be configured to take advantage of Avamar data deduplication to
dramatically decrease the amount of time, network bandwidth, and disk capacity required to
perform backup jobs. This is accomplished by installing a version 7.4 or later NetWorker agent
with built-in Avamar support on an application server. The agent performs source-side global
data deduplication before forwarding backup data to a NetWorker server and an Avamar Data
Store. Figure 11 shows a pictorial representation of how this is happening.
Figure 11 Avamar and Networker Integration Configuring NetWorker to deduplicate data can be performed in three simple steps:
• Select Avamar as a NetWorker managed application
• Create a deduplication node
• Discover Avamar deduplication node
Select the deduplication backup checkbox as shown in Figure 12.
.
2010 EMC Proven Professional Knowledge Sharing 23
Figure 12 Avamar and Networker Integration (Networker Console) ESG Lab used NetWorker with Avamar-enabled deduplication to back up a Windows virtual
server. Here are the results of the test.
• Backup took 5 minutes 31 seconds to back up 15 GB of data using an existing
NetWorker backup policy.
• Restore of a 10 MB log file took two seconds.
• NetWorker deduplication backup summary report yielded that the latest full backup of
the Windows virtual server protected 15,261 MB of data, yet only 127 MB of data was
transferred.
PROTECTION OF REMOTE OFFICES
Information critical to the success and efficiency of an organization is not just found in corporate
data centers; it also resides at remote and branch offices. Remote and branch offices (ROBOs)
contain a large amount of enterprise data, yet most still rely upon traditional tape-based backup
methods manned by non-IT staff. Tape backup, snapshots, mirroring, staging, and archiving all
significantly increase the amount of storage under management. Non-IT staff handling backups
and tapes increases the cost of data protection and the risk of data loss. In a distributed
environment, it is impractical to centralize backup operations with traditional backup
architectures and existing network capacity. Moving even daily incremental backup data sets
requires so much network bandwidth and time that even this simple process can become
prohibitively expensive and inefficient.
2010 EMC Proven Professional Knowledge Sharing 24
Avamar addresses this issue by improving backup processes in several ways. Replicating data
over a WAN instead of shipping tapes can be cost-prohibitive because of the significant amount
of data that needs to be transferred across the network. Avamar employs capacity reduction
technology and global data deduplication across sites and servers with a lightweight agent
running on the client systems it protects.
Avamar requires a one-time-only full backup where redundant data is eliminated and after that,
only new, unique sub-file data segments are backed up daily across the WAN to the data center
or DR site. The software identifies and filters repeated variable length data segments stored in
files within a single system and across multiple systems over time, capturing only net new
segments. In addition, data compression algorithms are applied to eliminate space and
redundant file patterns, further shrinking data volumes.
Avamar source-based global data deduplication significantly reduces the amount of data sent
over the WAN, enabling fast, daily full backups for remote office data. Backups are centrally
managed from a corporate data center using the intuitive Avamar Management console and
data can be encrypted for security.
Figure 13 illustrates how remote and branch offices using Avamar can reduce the amount of
data being transferred during backup (labeled ―Bandwidth Efficient�) by sending only unique
blocks of data to a centralized data center. At smaller remote offices, only Avamar software
agents are deployed on the systems, while larger remote offices and data centers typically
deploy a local Avamar server to improve recovery performance.
Figure 13 Protection of Remote Offices
2010 EMC Proven Professional Knowledge Sharing 25
Avamar also supports disaster recovery and offsite archival by replicating backup data over a
wide area network to a remote data center. Data deduplication and replication capabilities
ensure that only changes since the last backup, comprised of unique and compressed sub-file
variable length data segments, are replicated. This is extremely valuable, reducing the cost of
network bandwidth and improving the performance of backups and replication over the WAN.
There are 2 ways to back up clients in remote offices:
• Back up remote office clients to a small Avamar server located in the remote office
(remote Avamar backup server) and replicate data to a large centralized Avamar server
(centralized replication destination).
• Back up those clients directly to a large centralized Avamar server (centralized Avamar
backup server) and replicate data to another large centralized Avamar server
(centralized replication destination).
Which method is best depends upon the following factors:
• RTO (Recovery Time Objective): The first method is more reliable because restore can
be done directly from that server across the local area network to the client.
• Server Administration: The amount of administration and support required is roughly
proportional to the number of Avamar servers deployed in an environment. Accordingly,
option 2 is more reliable.
The tradeoff then becomes RTO versus the additional cost of deploying, managing, and
supporting multiple Avamar server instances.
According to the operational best practice, unless we cannot meet our restore time objectives,
architect the system so that clients back up directly to a large active centralized Avamar server
and then replicate data to another large, passive centralized Avamar server.
Best practices recommended for taking backup of remote clients:
• Copy the data to a portable storage device (such as a USB hard drive)
• Install the Avamar software on a client local to the Avamar server
• Attach the portable storage device to the local client
• Perform a one-time backup of all the data from the portable storage device
• Start to back up the remote client
This procedure ensures that when the initial remote client backup takes place, it will only need
to transfer data which has changed or has been added since the seeding process occurred.
2010 EMC Proven Professional Knowledge Sharing 26
ARCHITECTING AVAMAR TO ENSURE MAXIMUM SYSTEM AVAILABILITY In this section, we will consider the factors involved in architecting Avamar to ensure maximum
system availability. These factors include RAIN, RAID, replication, checkpoints, and backup of
remote offices. Of these factors, we have already discussed RAIN, replication, and how to
protect remote offices to ensure maximum system availability. We will discuss RAID and
checkpoints in this section.
RAID
An individual drive has a certain life expectancy before it fails, as measured by MTBF (Mean
time Before Failure). Since there are potentially hundreds or even thousands of drives in a disk
array, the probability of a drive failure increases significantly. As an example, if the MTBF of a
drive is 750,000 hours, and there are 100 drives in the array, then the MTBF of the array
becomes 750,000/100, or 7,500 hours. RAID (Redundant Array of Independent Disks) was
introduced to mitigate this problem.
RAID combines two or more disk drives in an array into a RAID set or a RAID group. The RAID
set appears to the host as a single disk drive. Properly implemented RAID sets provide:
• Higher data availability
• Improved I/O performance
• Streamlined management of storage devices
All standard Avamar server node configurations use RAID to protect the system from disk
failures. The Avamar server nodes are typically configured with SCSI hard drives in a RAID 5
configuration or SATA hard drives in a RAID 1 configuration.
Failed drives impact I/O performance and affect Avamar server performance and reliability.
Further, RAID rebuilds can significantly reduce the I/O performance, adversely impacting
performance and reliability of the Avamar server.
Operational best practices for implementing RAID:
• Protect the data on the Avamar server by using RAID to protect against individual hard
drive failures.
• Set the RAID rebuild to a relatively low priority.
• Always set up log scanning to monitor and report hardware issues; email home feature
can be set for this.
• For hardware purchased separately, the customer must regularly monitor and address
hardware issues promptly.
2010 EMC Proven Professional Knowledge Sharing 27
CHECKPOINTS A checkpoint is a snapshot of the Avamar server taken for the express purpose of facilitating
server rollbacks. If there are backups running, the server will suspend the backups, make the
system read-only, take the checkpoint, make the system read-write, and then resume the
backups. Checkpoints are taken to assist with disaster recovery and provide redundancy across
time. These are the consistent snapshots of the entire Avamar system that can be verified for
integrity. If an integrity check fails due to an inconsistency, and the inconsistency cannot be
fixed, the system can be quickly rolled back to a prior checkpoint. Checkpoints allow recovering
from operational issues and certain kinds of corruption by rolling back to the last validated
checkpoint. Although checkpoints are an effective way to revert the system to an earlier point in
time, checkpoints are like all other forms of redundancy and therefore, require disk space. The
more checkpoints we need to retain, the larger the checkpoint overhead. Operational best
practice is to leave the checkpoint retention policy at the default values, typically set to retain the
last two checkpoints and the last validated checkpoint. PROACTIVE STEPS TO MANAGE HEALTH AND CAPACITY Capacity is one of the most important aspects of administering an Avamar server. There are two
main components that define Avamar Server capacities:
• Storage Subsystem (GSAN) Capacity
• Operating System Capacity
GSAN capacity is the total amount of commonality factored data and RAIN parity data (net after
garbage collect) on each data partition of the server node. This is measured and reported by the
GSAN process. Operating System Capacity is the total amount of data in each data partition, as measured by
the operating system, which not only includes the size of all the stripes in the current working
directory, but also the checkpoint overhead and the amount of data stored on the data partition
by other activities.
Now we will describe how Avamar server will behave as it crosses various consumed storage
thresholds.
100% — The “Server Read-Only Limit.” When server utilization reaches 100% of total storage
capacity, it automatically becomes read-only. This is done to protect the integrity of the data
already stored on the server.
95% — The “Health Check Limit.” This is the amount of storage capacity that can be
consumed and still have a “healthy” server.
2010 EMC Proven Professional Knowledge Sharing 28
Although an Avamar server could be allowed to consume 100% of available storage capacity, it
is not a good practice to actually let that occur. Consuming all available storage can prevent
certain server maintenance activities from running, which might otherwise free additional
storage capacity for backups. For this reason, a second limit is established called the “health
check limit.” This health check limit is derived by subtracting some percentage of server storage
capacity from the server read-only limit. The default health check limit is 95%. We can
customize this setting according to a specific site but setting this limit higher than 95% is not
recommended. When server utilization reaches the health check limit, existing backups are
allowed to complete, but all new backup activity is suspended. Notifications are also sent in the
form of a pop-up alert each time we log into Avamar Administrator and a system event will
require acknowledgment before any future backup activity can resume.
80% — Capacity Warning Issued. When server utilization reaches 80%, a pop-up notification
will inform that the server has consumed 80% of its available storage capacity. Avamar
Enterprise Manager capacity state icons are yellow.
Icons that communicate capacity levels
The most significant daily operational impact of capacity is to backup performance and
hfscheck. Operational best practices recommendations:
• Understand how to monitor and manage the storage capacity of the Avamar server on a
daily basis.
• Limit storage capacity usage to 80% of the available "server utilization" capacity. Proactive steps to manage capacity:
When the GSAN capacity exceeds 80% of the final Read Only threshold, perform the following
operational best practices:
• Stop adding new clients to the system.
• Adjust retention policies to decrease the retention period of backups if possible and thus,
reduce capacity use.
2010 EMC Proven Professional Knowledge Sharing 29
• Check if backups are preventing a garbage collect operation from starting. If so, the
following error message is logged in the /usr/local/avamar/var/cron/gc.log file
MSG_ERR_BACKUPSINPROGRESS or garbage collection skipped because
backup is in progress.
• If possible, decrease the GSAN capacity by deleting or expiring backups and running
garbage collection. Note that deleting backups does not free space until garbage collect
has run several times. Garbage collect finds and deletes the unique data associated with
these backups.
• Replicate the data to another server temporarily, and then replicate the data back after
reinitializing the server. Because replication creates a logical copy of the data, this
compacts all the data onto fewer stripes.
• For multi-node Avamar servers utilizing RAIN, add nodes and rebalance the capacity.
DAILY SERVER MAINTENANCE:
Daily Avamar server maintenance comprises three essential activities:
• Checkpoint (discussed above)
• Garbage collection
• HFS check
Garbage Collection Garbage collection is the process of recovering unused space from
backups that have expired. The garbage collection process requires a quiet system in order to
run — garbage collection will not initiate if any backups are running.
HFS check A Hash File system (HFS) check is an internal operation that validates the integrity
of a specific checkpoint. Once a checkpoint has passed a HFS check, it can be considered
reliable enough to be used for a system rollback. In order for a HFS check to successfully start,
there must be two or fewer currently running backups per storage node.
Avamar servers define two daily “maintenance windows,” during which some or all of the
essential server maintenance activities are performed. The default start times for these
maintenance windows are:
• 6:00 A.M. (morning)
• 6:00 P.M. (evening)
2010 EMC Proven Professional Knowledge Sharing 30
The following table lists the essential server maintenance activities, i.e. checkpoint,
HFS check, and garbage collection, and summarizes each activity’s impact on
normal system operation.
CHECKPOINT GARBAGE COLLECTION
HFS CHECK
Runs in morning? Yes Yes Yes
Runs in evening? No No No
Typical run time 5 minutes -
1 hour
2 hours max Several hours
Starts if backups
are running?
Yes No Only if two or fewer
backups are running
on each node
Effect on running
backups
Temporarily
suspended
Not applicable Unaffected
Effect on new
backups
Can start Cannot start Can start
System state
during activity
Read-only Read-only Read-write
TAPE INTEGRATION
Avamar Tape Output option provides the ability to archive backup data stored within an Avamar
server to tape via a disk cache. This option utilizes a disk staging area (for example, a media
server with a disk pool or a file server) to temporarily stage Avamar backup data in un-
deduplicated format, which is then retrieved by any standard third-party tape-based backup
software solution to write the data to tape. The Avamar Tape Output option is configurable to
output to tape either all or a subset of the backup data stored within the Avamar server, and can
also be tuned based on the size of the disk staging area. A restore is performed by a restore
directly from tape to the appropriate client(s). The Avamar Tape Output option provides the
following advantages:
• A restore can be performed at the granularity of a single file.
• A restore can be performed directly from the tape software to a client without having to
restore via the Avamar server.
• ACL’s and alternative data streams are maintained on tape if the operating system of the
disk staging area system matches the operating system of the original client that was
backed up.
2010 EMC Proven Professional Knowledge Sharing 31
Avamar has introduced Avamar Data Transport, an Avamar server that runs as a virtual
machine in a VMware ESX Server (3.x) environment. Avamar Data Transport is a hardware and
software system that transports deduplicated data from an Avamar server to tape. Transport
nodes are sized at 1.0 TB nodes. A transport node functions as a target of replication from a
primary Avamar server which, after replication is complete, is moved to tape. Transport nodes
can also be recovered from tape, allowing backed up data to be accessed by Administrators or
users. Avamar Data Transport 1.0 supports a maximum of four transport nodes. Following are
the terms that will be useful for understanding its working:
Avamar Data Transport Framework The framework is a service-oriented architecture that
provides applications with common services such as communication, security, and event
logging.
Control Node The control node is a 64-bit Linux host where the Avamar Data Transport
application and the Avamar Data Transport Framework run. The Avamar Data Transport user
interface is accessed from a web server that is installed on this host. In addition, the database of
all transported files is often located on the control node. There is only one control node in an
Avamar Data Transport system.
Tape Backup Server Avamar Data Transport supports NetWorker or Symantec Net Backup
tape backup products. Both of these products have a server component that drive the tape
archiving process.
Transport Node Transport nodes are Avamar Virtual Edition servers that have been modified
specifically for use in the Avamar Data Transport system. They are the targets of replication
from Avamar servers, and are used as backup and restore target by the tape backup server.
The system consists of the following physical components:
• Physical Avamar server
• ESX server for hosting control and transport nodes
• Tape backup server configured with NetWorker or Symantec NetBackup
• Tape library, tape drives, and tapes
2010 EMC Proven Professional Knowledge Sharing 32
FIGURE 14 Avamar Data Transport Components
Before installing Avamar Data Transport, the components mentioned above must be running in
your network environment. After the pre-installation checks have been made, install and
configure the Avamar Data Transport components (as shown in Figure 14) on the appropriate
hosts in the system:
• Avamar Data Transport System Service (on the Avamar servers)
• Control node (on the ESX server)
• Transport nodes (on ESX server)
• Transport node services (on transport nodes)
• Avamar Data Transport groups or policies on the tape backup server
Installation and configuration of these components occurs in the following order, as shown in
Figure 14:
1. Control node
2. Avamar Data Transport System Service on the Avamar server (or servers)
3. Transport nodes
4. Tape backup server and tape backup clients
5. Transport node Services
After installing these components, Avamar Data Transport is configured to transport
deduplicated data to tape. This way, the Avamar Data Transport node is used for tape out
mechanism and ships the data backed up to tape.
2010 EMC Proven Professional Knowledge Sharing 33
WHAT’S NEW Avamar Desktop/Laptop introduction in Avamar 5.0 Recently EMC introduced Avamar 5.0, designed for backing up desktops and laptops. The
Avamar Desktop/Laptop features are designed to enhance the functionality of the Avamar client
for Windows and Mac desktops and laptops. Avamar Desktop/Laptop is based on the Avamar
client software. It adds enhanced features for an enterprise’s desktop and laptop computers.
These features include:
• Web browser UI that provides manual backup, restore by search, restore by browse, and
backup history.
• UI available in ten languages.
• Domain-oriented data security. Users are authenticated through your enterprise’s
Active Directory or Open LDAP-compliant directory service. Users can also be
authenticated using Avamar authentication.
• User-initiated backups. Users can manually initiate a backup from the client.
• Users can restore folders and files directly to the original location.
• Creation of a restore set by search or directory-tree browse. Users can use search or
can browse a backup directory-tree to create a set of folders and files to restore.
• Backup history. Users can view the folders and files backed up in their last 10 successful
backups.
• Deployment of Avamar clients using common Systems management tools. Avamar
Desktop/Laptop can be push installed on Windows and Mac desktop and laptop
computers using Systems management tools, such as Microsoft Systems Management
Server 2003 (SMS).
WHY AVAMAR?
• Solves the challenges associated with traditional backup by using patented global,
source data deduplication technology to identify redundant data segments at the source
before transfer across the network.
• Enables fast, reliable backup and recovery for VMware environments, remote offices,
and LAN/NAS servers across existing networks and infrastructure.
• Data encrypted in flight and at rest, thus adding security.
• RAID protection from disk failures.
• Supports SAN or internal disk storage.
• Scalable grid architecture providing scalability.
2010 EMC Proven Professional Knowledge Sharing 34
• Fault tolerance across nodes, thereby eliminating single points of failure and providing
high availability using RAIN technology.
• Eliminates manual recovery drill checks by ensuring that checks run daily in order to
ensure data integrity.
• New design and features for desktop and laptop solution.
• Centralized web-based management provides ease of management.
• Eliminates LAN backup bandwidth bottlenecks; reduces daily network bandwidth by up
to 500x using deduplication.
• Provides an alternative to archaic IT processes (shipping tapes for disaster recovery),
providing automated, encrypted remote copy over existing WANs, thereby mitigating risk
involved with shipping tapes.
• Integration with exiting backup infrastructure and software.
• Simple one-step recovery.
• Reduction of total back-end storage by up to 50x for cost-effective, long-term, disk-
based recovery.