IMA - VirtualizationAdmin.com

Download Report

Transcript IMA - VirtualizationAdmin.com

Independent Management
Architecture:
Terms and Concepts
Barry Flanagan, Southeast Systems Engineer
New Terms
• IMA: Independent Management Architecture
• Data Store: Central configuration database
• LHC: Local Host Cache (Persistent data cache that exists on
each server)
• Data Collector: Manages dynamic data and client
enumeration/resolution (replaces ICA Master Browser)
• Zone: Deliberate grouping of MetaFrame XP servers, each with
its own Data Collector
• CMC: Citrix Management Console (replaces MetaFrame 1.8
administration tools)
What is IMA? Why is it important?
IMA…
• Is a TCP-based, event driven messaging bus, used by
MetaFrame XP servers.
• Is a modular and easily extensible subsystem capable of
supporting current and future MetaFrame products and tools.
• Overcomes the scalability constraints of the MetaFrame 1.8
platform, allowing MetaFrame XP to scale environments to new
levels.
• Provides capability to administer any farm from a central tool
(CMC) that doesn’t have to run on a MetaFrame server.
Independent Management Architecture
Citrix Management
Console
MetaFrame XP
NT 4.0 TSE
MetaFrame XP
Windows 2000
Independent Management Architecture (IMA)
DB
Central Data Store
•SQL, Oracle, Access
Load
Management
Application
Packaging &
Delivery
System
Monitoring &
Analysis
MetaFrame Server Farms
MetaFrame 1.8:
• Server Farms in MetaFrame 1.8 are a collection of servers on a
given broadcast segment that are managed as a single unit.
• Server Farms in MetaFrame 1.8 may also be defined by sharing
a common ‘Application Set’.
MetaFrame XP:
• The Server Farm in MetaFrame XP defines the scope of
management as well as the ‘Application Set’.
• Server Farms in MetaFrame XP are designed to operate across
segments and are managed through the Citrix Management
Console.
MetaFrame 1.8/ICA Browser
MetaFrame 1.8/ICA Browser Attributes
• Server Farms cannot span segments.
• Each segment has ONE ICA Master Browser.
• ICA Master Browser stores dynamic data for the segment and
handles Enumeration/Resolution for ICA clients.
• Persistent data stored in registry (farm membership, licenses,
published applications, etc.).
ICA Master
Browser
MFAdmin,
PAM, etc.
Segment 1
10.1.1.x
Farm 1 (2, 3)
ICA Master
Browser
MFAdmin,
PAM, etc.
Segment 2
10.1.2.x
Farm 4 (5,
6)
MetaFrame 1.8/ICA Browser
MetaFrame 1.8/ICA Browser Attributes
• Persistent data read by ICA browser/PN Service at startup.
• Cross server configuration tools read/write to registry on all
servers.
• Servers communicate via UDP broadcasts, remote REG calls,
RPCs, etc.
ICA Master
Browser
MFAdmin,
PAM, etc.
Segment 1
10.1.1.x
Farm 1 (2, 3)
ICA Master
Browser
MFAdmin,
PAM, etc.
Segment 2
10.1.2.x
Farm 4 (5,
6)
MetaFrame XP/IMA
MetaFrame XP/IMA Attributes
• Server farms can span segments, can contain multiple zones.
• Each zone has ONE Data Collector.
• Data Collectors store dynamic data and handle
Enumeration/Resolution for ICA clients.
• Persistent farm data stored in shared, persistent Data Store.
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
MetaFrame XP/IMA
MetaFrame XP/IMA Attributes
• Persistent data read from DS at startup, cached in Local Host
Cache.
• Management tool communicates via IMA to Data Store and
member servers.
• Servers communicate via IMA (TCP).
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Data Store
Attributes of the MetaFrame XP Data Store
(DS)
• The DS is a repository (database) which contains persistent,
farm-wide data, such as member servers, licenses in farm, zone
configs, printers/drivers, published apps, load evaluators, trust
relationships, etc.
• Each MetaFrame XP farm shares one Data Store.
• All information in the DS is stored in an encrypted binary format
(except indexes).
• A farm can operate for 48 hours if DS is unavailable, then
licenses time out and no new users can connect.
• A DS can be an Access, MS SQL, or Oracle database.
• A DS can be configured for either ‘Direct’ or ‘Indirect’ access.
Data Store in ‘Direct’ Mode
Attributes of Direct Mode
• Uses Microsoft SQL 7/2000 or Oracle 7.3.4/8.0.6/8.1.6 database.
• Servers initialize directly from the DS via ODBC.
• Servers maintain an open connection to the database for
consistency checks.
LHC
LHC
LHC
LHC
LHC
LHC
DS
Data Store in ‘Indirect’ Mode
Attributes of Indirect Mode
• Uses JET 4.x, Microsoft SQL 7/2000 or Oracle 7.3.4/8.0.6/8.1.6
database.
• Member servers communicate via through ‘IMA host’ server to
read/write to data store.
• If using JET database, MF20.MDB lives on the ‘IMA host’ server.
DS
LHC
LHC
LHC
LHC
LHC
LHC
DC
IMA Host
(indirect mode)
Local Host Cache (LHC)
Attributes of the Local Host Cache
• A subset of the Data Store, stored on each individual server
(IMALHC.MDB).
• Contains basic info about servers in farm, pub. apps and
properties, trust relationships, server specific configs (product
code, SNMP settings, load evaluators, etc.).
• Used for initialization if DS is down.
• Used for ICA client application Enumeration.
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Data Collectors
Attributes of Data Collectors
• A DC stores dynamic information about a farm, such as servers
up/down, logons/logoffs, disconnect/reconnect, license in
use/released, server/application load, etc.
• There is a DC for each Zone.
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Data Collectors
Attributes of Data Collectors
• DC’s handle all ICA client Resolution activity, should handle all
Enumeration activity. ANY DC can Resolve ANY app for ANY
client (DC’s are peers in a multi-zone implementation).
• DC’s distribute most persistent data changes to member
servers for LHC update.
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Zones
Attributes of Zones
• Logical, centrally configurable grouping of MetaFrame XP
servers.
• Each Zone has one Data Collector (DC).
• Can span IP networks (LAN, WAN).
• Aren’t necessarily tied to an IP segment (only by default).
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Zones
Attributes of Zones
• Are useful for partitioning/controlling persistent data update
traffic and for distributing ICA client Enumeration/Resolution
traffic.
• A Zone can contain up to 256 hosts without a registry
modification.
• In most cases, fewer zones are better!
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Citrix Management Console (CMC)
Attributes of the CMC
• Central management tool where 98% of farm
configuration/maintenance occurs.
• Extensible framework that allows different tools to ‘snap in’.
• Doesn’t need to run on a MetaFrame server.
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
Citrix Management Console (CMC)
Attributes of the CMC
• Works through the IMA service (dest. port 2513) to access DS,
DC, and member servers.
• Should be run through a DC that has local access to the DS.
• Is the most read/write intensive usage of the DS.
Server
Farm
CMC
DC
DC
DS
Zone 2
LHC
LHC
LHC
LHC
LHC
LHC
Zone 1
IMA In Depth
Hardware and Software Configuration






Load up on processors and memory
Have home directories on separate server
Roaming profiles in multi-server enviroments
 Q161334-Guide to Windows NT 4.0 Profiles
and Policies
NTFS partitions only ( at least 4096 cluster )
Install only required network components and
protocols
Change drive letters at installation time only
Hardware and Software Configuration

For 4 and 8 processors systems, use one
controller for OS and one for applications and
temporary files.
 Dedicate a drive for page file for best
performance.
 Increase Maximum Registry Size to 100 MB.
 See MF Install and Tuning Tips for more info.
Selecting a Data Store
Direct Mode
IMA directly querying the database
• Microsoft SQL Server 7 or 2000
• Oracle 7.3.4, 8.0.6, or 8.1.6
Indirect Mode
IMA requesting another server to query the database
on its behalf
• Gathering its DS information indirectly from another
server who is accessing the DS directly
Data Store Info
Indirect Mode
Select Use
a local database as the data store to enable
Indirect mode to Access (Direct Mode is not available for
Access) on the first server installed. All subsequent servers
joining the farm must be installed with the Connect to a data
store set up locally on another server option.
First server will be Zone DC by default.
Server hosting the Access DS will be the only server to write
to the Access database.
Server hosting the DS in Access acts as proxy for all other
servers.
Overcomes the file locking and corruption problems common
with Access.
Data Store Info
Using Access
Approximately
20 MB of disk space should be available for
every 100 servers in the farm.
32 MB of additional RAM is recommended if the MetaFrame
XP server will also host connections.
Need MDAC 2.5 SP 1 installed on TSE. Stop TS Licensing
Service before Installing MDAC. Reboot.
 %ProgramFiles%\Citrix\Independent Management
Architecture\MF20.MDB ( System must have read/write
access)
The default user name/password is citrix/citrix. To change the
password on the database, use the dsmaint config
/pwd:newpassword command with the IMA service running.
Data Store Info
Using Access
Each
time the IMA service is stopped gracefully, the existing
mf20.mdb file is backed up, compacted, and copied as
mf20.unk. Each time the IMA service starts successfully, it
deletes any existing instance of mf20.bak and then renames the
mf20.unk file to mf20.bak. This file is used when the dsmaint
recover command is executed.
If the server runs out of disk space on the drive where the
mf20.mdb file is stored, the automatic backup stops
functioning. Always ensure there is enough disk space to hold
3 times the size of the mf20.mdb.
Perform backup of DS with DSMAINT BACKUP
Data Store Info
Using SQL
Approximately
20 MB of disk space for every 100 servers in
the farm. The disk space used may increase if there are a large
number of published applications in the farm.
The temp database should be set to Auto Grow on a partition
with at least 1 GB of free space (4 GB is recommended if it is a
large farm with multiple print drivers).
Verify that enough disk space exists on the server to support
growth of both the temp database and the farm database.
Use MDAC 2.5 SP1 on TSE. Do not use MDAC 2.6 with SQL
2000. Known bug.
Data Store Info
Using SQL
When
using Microsoft SQL Server in a replicated environment,
be sure to use the same user account on each Microsoft SQL
Server for the DS.
Each MetaFrame XP farm requires a dedicated database.
However, multiple databases may be running on a single
Microsoft SQL Server.
The MetaFrame XP farm should not be installed in a
database that is shared with any other client-server applications.
Databases should have the Truncate log on Checkpoint
option set to keep log space controlled.
Ensure DS is backed up whenever a change is made via CMC.
Data Store Info
Using SQL
For
high security environments, Citrix recommends using NT
Authentication only.
The account used for the DS connection should have
db_owner (DBO) rights on the database that is being used for
the DS.
If tighter security is required, after the initial installation of the
database as DBO, the user permissions may be modified to be
read/write only.
If installing more than 256 servers in a farm, increase number
of worker threads available for database.
Data Store Info
Using Oracle
Approximately
20 MB of disk space for every 100 servers in
the farm. The space used may increase if there are a large
number of published applications in the farm.
The Oracle Client (version 8.1.55 or 8.1.6) must be installed on
the terminal server prior to the installation of MetaFrame XP.
The 8.1.5 and 8.1.7 clients are not supported with MetaFrame
XP.
The server should be rebooted after installation of the Oracle
Client, or the MetaFrame XP installation fails to connect to the
DS.
Data Store Info
Using Oracle
Oracle8i version
8.1.6 or later is recommended. However,
Oracle7 (7.3.4) and Oracle8 (8.0.6) are supported for the
MetaFrame XP platform.
Creating a separate tablespace for the DS simplifies backup
and restoration operations.
A small amount of data is written to the system tablespace. If
experiencing installation problems, verify that the system
tablespace is not full.
Using Shared/MTS (Multi-Threaded Server) mode may reduce
the number of processes in farms over 200 servers. Consult the
Oracle documentation on configuring the database to run in
MTS mode.
Data Store Info
Using Oracle
Oracle
for Solaris supports Oracle authentication only.
Oracle user account must the the same for every server in the
farm because all servers share a common schema.
This account needs the following permissions:
Connect
Resource
Dedicating a server for Indirect Mode
May
be necessary when the following occurs:
Delays in using CMC
Increased IMA service start times due high CPU utilization on
server hosting DS.
Cut maximum users to one half to two thirds of full load to
improve performance.
Bandwidth Requirements
In
a single server configuration, a single server reads
approximately 275 KB of data from the DS. The amount of data
read is a function of the number of published applications in the
farm, the number of servers in the farm, and the number of
printers in the farm. The number of kilobytes read from the DS
during startup can be approximated by the following formula:
KB Read = 275 + 5*Srvs + 0.5*Apps + 92*PrintD
Where:
Srvs = Number of servers in the farm
Apps = Number of published applications in the farm
PrintD = Number of print drivers installed on the member
server
Data Store Info
High


Latency WAN Concerns
Without use of replicated databases, may create
situations where DS is locked for extensive periods
of time when performing maintenance
A high latency situation reads should not adversely
affect any local connections, but the remote site may
experience slow performance.
Replicated


Databases
Speed up performance if there is enough
MetaFrame servers to justify the cost
Database replication will consume bandwidth but is
controlled through the database chosen, not
MetaFrame
Data Store Info
Access
is best used for centralized farms.
Access supports only indirect mode for other servers, and as
such will have slower performance then a direct mode DS on
large farms.
Database replication is not supported with Access.
Databases supporting replication should be used when
deploying large farms across a WAN.
Server farms with over 100 servers should use SQL or Oracle
to remain at acceptable performance levels.
Data Store Info
Farms
using excessive printer drivers and scheduled
replication should use SQL or Oracle.
Farms that cycle boot large groups of servers simultaneously
should use SQL or Oracle in direct mode to minimize the IMA
service start time.
Both Microsoft SQL Server and Oracle are very similar in
performance. In the Citrix Test eLabs both database servers
performed similarly with large farms. The choice between the
two should be based on feature sets of the databases, in-house
expertise, management tools, and licensing costs rather than
performance numbers
Use Microsoft Clustering Services with SQL or Oracle
Parallel Server with Oracle for fault tolerance.
Data Store Info
DS
Query Interval
• Key: HKLM\Software\Citrix\IMA\DCNChangePollingInterval
• Default value: 600000 milliseconds REG_DWORD: 0x927C0
If
a member server is unable to contact the data store
for 48 hours, licensing will stop functioning on the
member server
CMC always connects directly to the DC
Change > 10K in size, all member servers in the
farm will be sent a change notification and query the
DS for the change
Data Distribution with Data Collectors





Server 1 writes information to the DS
Server 1 sends change notification to its zone
DC
Zone DC distributes change notification to all
member servers in its zone
Other zone DC’s receive notification and
distribute it to all member servers within their
respective zones
All member servers receive the notification and
update their LHC as requested
Data Distribution with Data Collectors

Inter-zone connection formula
• N * (N-1)/2, where N is the number of zones in the farm

IMA ping configuration parameter
• Key: HKLM\Software\Citrix\IMA\Runtime\KeepAliveInterval
• Default value: 60000 milliseconds REG_DWORD: 0xEA60

Zone DC synchronization parameter
• Key:HKLM\Software\Citrix\IMA\Runtime\Gateway\ValidationInterval
• Default value: 300000 milliseconds REG_DWORD: 0x493E0

Inter-zone connection formula
•
Key:HKLM\Software\Citrix\IMA\Runtime\MaxHostAddressCacheEntriesl
• Default Value: 256 Entries REG_DWORD: 0x100
Data Distribution with Data Collectors

Bandwidth requirements between zones
 Connect: ~3Kb
 Disconnect: ~2.25Kb
 Reconnect: ~2.91Kb
 Logoff: ~1.50Kb
 CMC: ~2.23
 Application Publishing: ~9.07
Data Collector Elections

Each zone is responsible for electing its own data
collector (DC). By default, the first server in the
farm becomes the DC and is set to Most Preferred.
If the setting is changed from Most Preferred,
another election will take place. DC elections are
won based on the following criteria:
1. Highest Master Version Number (1 for all
MetaFrame XP 1.0 servers)
2. Lowest Master Ranking (1=Most Preferred –
4=Not Preferred)
3. Highest Host ID (0-65536 randomly assigned at
installation)
Data Collector Elections







To view server’s ranking, use Queryhr ( copy from
support\debug\i386 on CD
DC elections are triggered in the following
situations:
A member server loses contact with the DC.
The DC goes offline.
A farm server is brought online.
The querydc -e command is executed to force
an election.
Zone configurations are changed (i.e. zone
name, election preference, adding or removing
servers)
Data Collector Elections







When a new DC is elected, all servers in the zone send a
complete update to the new DC. The following formula
can be used to approximate the amount of data in bytes
sent by all servers in the zone to the new zone DC:
Bytes = (11000 + (1000 * Con) + (600 * Discon) + (350
* Apps)) * (Srvs - 1)
Where:
Con = Average number of connected sessions per server
Discon = Average number of disconnected sessions per
server
Apps = Number of published applications in the farm
Srvs = Number of servers in the zone
Local Host Cache
Attributes of the Local Host Cache
A subset of the Data Store, stored on each
individual server (IMALHC.MDB).

Contains basic info about servers in farm,
pub. apps and properties, trust relationships,
server specific configs (product code, SNMP
settings, load evaluators, etc.).


Used for initialization if DS is down.
Used for ICA client application
Enumeration.

Local Host Cache
On
the first startup of the member server, the LHC is populated
with a subset of information from the DS. From then on, the
IMA service is responsible for keeping the LHC synchronized
with the DS. The IMA service performs this task through change
notifications and periodic polling of the DS.
In
the event the DS is unreachable, the LHC contains enough
information about the farm to allow normal operations for up to
48 hours.
During this “grace” period, the server continues to service
requests while the IMA service attempts to connect to the DS
periodically (based on the DS query interval as described in the
Data Store Activity section of the IMA Architecture chapter of
this document). If the DS is unreachable for 48 hours, the
licensing subsystem fails to verify licensing and the server stops
taking incoming connections.

Local Host Cache
Because
the LHC holds a copy of the published applications
and NT trust relationships, ICA Client application
enumeration requests can be resolved locally by the LHC.
The member server must still contact the zone DC for LM
resolutions.
If
the IMA service is currently running, but information in
the CMC appears to be incorrect, a refresh of the LHC can
be manually forced by executing dsmaint refreshlhc from
the command prompt of the affected server.
Local Host Cache
If the IMA service does not start, it may be caused by
a corrupt LHC.
.
1. Verify the DS is available before continuing because
this procedure causes the LHC to be reloaded directly
from the DS.
2. Stop the IMA service on the MetaFrame server.
3. Launch the ODBC Data Source Administrator. On
Windows 2000, choose Control Panel |
Administrative Tools | Data Sources (ODBC). On
TSE choose Control Panel | ODBC Data Sources.
4. Select the File DSN tab.
Local Host Cache
5. Open the imalhc.dsn file located in
%ProgramFiles%\Citrix\IndependentManagement
Architecture by default.
6. Once that file is selected, click on Create from the
ODBC Setup screen.
7. Enter in any name besides imalhc for the new LHC
database. Optionally, rename the old imalhc and reuse
the name.
8. Exit the ODBC Data Source Administrator.
Local Host Cache
8. Exit the ODBC Data Source Administrator.
9. Modify the following registry value:
Key: HKEY_LOCAL_MACHINE\SOFTWARE\Citrix\
IMA\RUNTIME
Value: PSRequired REG_DWORD: 0x1
10. Restart the IMA service.
Note: The DS server must be available for this
procedure to work. If the DS is not available, the IMA
service fails to start until the DS is available.