Capacity planning and system design determine several
important characteristics of a cc:Mail system. Capacity issues affect
the performance, reliability and maintanability of the system. System
design affects performance, maintainability and the amount of administrative
overhead for maintenance. These topics are fairly advanced but since
an ounce of prevention is worth a pound of cure, I wanted to cover them
before jumping into the how-to aspects of setting up a cc:Mail system.
For the purposes of this chapter, an understanding
of OSI model networks and cc:Mail system architecture is assumed. If
you are new to cc:Mail, you may want to read the capacity planning section
and return to this chapter again later. Issues internal to the cc:Mail
system, such as the numbers of post offices and cc:Mail Routers, are
discussed in the system design section of this chapter. Server disk
space requirements are discussed in the capacity planning section. Implementation
planning is covered in Appendix B.
Capacity Planning
More than client/server systems, the performance and
reliability of cc:Mail systems depends on the stability and performance
of the network environment. As a result, some capacity planning issues
for cc:Mail are really network design issues while others are cc:Mail-specific.
Most of the literature pertaining to cc:Mail does not discuss network
design issues because in stable network environments with low and moderate
loads, cc:Mail systems tend to be very reliable.
Capacity Planning for Network Design Aspects of cc:Mail
The network design aspects of capacity planning involve
the anticipated load on file servers and networks as well as file server
storage requirements. Ideally, networks and file servers with capacities
appropriate for immediate and near-term loads should be in place before
a new cc:Mail system is rolled out. However, anticipating the load on
networks and file servers produced by a cc:Mail system is extremely
difficult. In fact, it is not possible to accurately predict the load
that a cc:Mail system will produce because there are too many variables
and not all of them can be known in advance. Nevertheless, because network
design issues are critical to cc:Mail, I have listed some of the key
issues below. The issues can be broken down into local area network
issues and wide area network issues.
Key LAN capacity planning issues include:
1. File server configuration and utilization
2. File server storage requirements
3. Workstation CPU, speed, memory and storage requirements
4. The capacities and utilizations of LAN segments
Key WAN capacity planning issues include:
1. Network protocols used for cc:Mail data communications
2. Configuration of cc:Mail Routers and e-mail gateways
5. The capacities and utilizations of WAN segments
Think of these issues as things you should consider
before implementing a cc:Mail system. Although the load on networks
and file servers that a cc:Mail system will produce cannot be accurately
predicted, capacity planning for cc:Mail always begins with a knowledge
of current utilizations and capacities.
Known Variables: The Ideal World
In an ideal world, you would be able to determine the
load on networks and file servers that a cc:Mail system will produce.
Although not all of the variables can be known before the system is
in full production, some of them typically are known ahead of time.
Commonly known LAN variables include:
1. The number of users
2. Workstation configuration and available storage capacities
3. Current file server configuration and utilization
4. Available file server storage capacities
5. The capacities and utilizations of LAN segments
6. Anticipated one-year growthCommonly known WAN variables include:1.
Network protocols used for cc:Mail data communications
2. Configuration of cc:Mail Routers and e-mail gateways
3. The capacities and utilizations of WAN segments
4. Anticipated one-year growth
For some network applications these factors would be
adequate to predict network load. For example, to implement software
to make customer records available to customer service personnel when
customers called, capacity issues might be broken down as follows:
1. Average number of calls handled during each hour
of the business day
2. Average network bandwidth utilization for each hour of the business
day
3. Average size of customer records
4. The amount of network data transmission overhead per record
5. Anticipated one to two year growth in call volume
A test system could be used to estimate the amount
of network bandwidth that the application would consume. Network analysis
tools such as a Network General Sniffer, HP OpenView or CoroNet Super
Monitor could be used to determine the total amount of data transmitted
over the network for each one-record transaction and the amount of time
required for each transaction. Combined with the call volume, an estimate
of the total amount of data transmitted over the network during each
hour of the business day could be calculated. The result could be compared
to the known network capacity and current bandwidth utilization.
In other words, it wouldn’t difficult to tell with
reasonable accuracy whether implementing such as system would overload
the network during peak network utilization and call volume hours. Redesigning
or upgrading the network could become a prerequisite for the project.
Unfortunately, not everything is as neat and tidy as predictable call
volume and customer records of known sizes.
Unknown Variables: The Real World
Although no one wants to implement a network application
on a production network without knowing what will happen, e-mail is
something that the users themselves have a certain amount of control
over. Users control the e-mail system through the way that they use
it. If you have a site with 500 users, it makes a great deal of difference
whether each user sends and receives an average of 3 to 5 messages a
day or 25 to 30 messages. It also matters whether the messages are addressed
to users at the same post office or other post offices. It matters much
more whether the messages are simple text or contain files, and whether
the files are documents, fax images or QuickTime movies.
Variables that are not known before a cc:Mail system
is implemented include:
1. Number of messages sent and received by each user
2. Contents and corresponding sizes of messages
3. Amount of inter-post-office message traffic
4. Which messages are retained by recipients and for how long
New cc:Mail administrators usually ask what the average
values for these variables are. Unfortunately, averages are not known.
Average values cannot be known without conducting a rather sophisticated
study involving sites of various sizes in different industries with
different system designs. What is much more important, however, is that
any averages that could be known would not be very meaningful for a
new installation.
A group of 200 users on one post office at a law firm
might e-mail 500 or more 100K to 250K documents every day. If the users
never deleted any messages, which is likely in a legal environment,
the post office could grow as much 90MB per day. If, on the other hand,
they sent 1,000 1K text messages per day and half were deleted after
being read on the same day, the post offices might grow by less than
500K per day -- less if additional messages older than one day were
deleted each day.
Capacity Planning: Facing the Unknown
In practice, cc:Mail administrators deal with the network
design aspects of cc:Mail in one of three ways:
1. Make an educated guess
2. Base expectations on a similar site in the same organization
3. Over-engineer
Other than server storage and workstation requirements,
most smaller cc:Mail systems are implemented without a detailed review
of capacity issues. Most of the time there are no negative consequences.
A properly engineered 16Mbps Token Ring network with an average 20%
utilization during peak hours is probably a hospitable environment for
a cc:Mail system. A 66MHz 486 NetWare 3.11 file server with 32MB of
RAM less than 100 DOS and Windows users can almost certainly host a
100-user cc:Mail post office as well. In low utilization environments,
making educated guesses about capacity issues is fairly safe.
Internal comparisons, while perhaps better than averages,
may not be valid for the same reasons that averages are not meaningful.
Users in an advertising department may e-mail large graphics files to
one another while accountants may e-mail smaller spreadsheet files.
A valid internal comparison requires finding a similar group of users.
However, even when this is possible, there is no guarantee that the
new user group will use e-mail in the same way as the reference group.
I recommend an over-engineering design philosophy.
Over-engineering means deliberately implementing systems the capacities
of which exceed any reasonably foreseeable near-term load.
Over-engineering and Worst Case Scenarios
On its face, over-engineering faces the same problems
as capacity planning in general because maximum loads are unknown. Nevertheless,
theoretical maximums can be determined for worst-case scenarios. For
example, if 33% of 350 users on one post office each sent messages averaging
25K simultaneously, this would mean approximately 158 workstations pouring
about 4MB of data into the same file server. To make this work, however,
you still have to guess how users will use the system. If, in the same
example, 5% of users sent messages with 1MB file attachments simultaneously,
this would mean approximately 18 workstations pouring about 18MB of
data onto the network.
The short answer is that over-engineering also involves
guesswork. Using the maximum supported file server volume size for cc:Mail
post offices, for example, means guessing that post offices are very
unlikely to exceed this size.
Over-Engineering for Mission-Critical Applications
Over-engineering is the best insurance money can buy
when it comes to mission-critical applications. When a mission critical
system fails, there is a loss of productivity. Employees get less work
done in the same amount of time, although they are paid the same. If
a critical system fails and 200 employees lose 10 minutes of work each,
approximately 33 person hours of labor are instantly destroyed. If the
average employee earns $25.00 per hour, $825.00 are lost the second
the failure occurs. Losses do not stop there. Employees are not paid
to wait for information systems to come back on line so they can work.
In most organizations, networks and e-mail are mission
critical systems. If a post office goes down because a server was configured
with 32MB of RAM instead of 128MB; or because one less $3,000.00 network
router was used in the network design, the small savings on paper can
rapidly translate into many times as much in hidden losses. From this
perspective, over-engineering for cc:Mail system is a financially sound
strategy.
Capacity Planning and Server Storage Requirements
File server volume sizes, the number of post offices
and the number of users per post office, are closely related. The value
of each of these variables depends both on the other two and
on how much message data each user will have. The amount of message
data each user will have depends on other variables such as the number
of messages, the sizes of messages, and the length of time that messages
are retained. The amount of server disk space required for a given cc:Mail
post office depends on several variables, including:
1. The number of local users in the post office
2. Number of messages sent and received by users
3. Contents and corresponding sizes of messages
4. Which messages are retained by recipients and for how long
5. Whether messages are routed through the post office
6. Prior to Release 6, how often maintenance is performed
For versions prior to Release 6, Lotus recommends 100
users per post office and 10MB of server disk space per user. The absolutely
arbitrary 10MB per user number has is a good rule of thumb. For versions
prior to Release 6, the maximum server volume size supported by the
cc:Mail maintenance utilities was 2GB. For Release 6 and later, the
maximum supported volume size doubled from 2GB to 4GB.
If messages exchanged between other post offices are
routed through a post office, server disk space requirements will be
increased. For versions prior to Release 6, running the maintenance
utilities can significantly reduce post office size. In versions prior
to Release 6, free spaces in the post office caused by message deletions
were not reused efficiently.
Unless the cc:Mail system has less than 100 users,
I recommend always using the maximum supported volume size and
for versions prior to Release 6, you should reserve at least one additional
2GB volume for maintenance purposes.
Pushing Back User Control of File Server Disk Space
Although there is no mechanism within cc:Mail to limit
the amount of disk space allocated to each user, as an administrator,
you can push back user control by establishing policies regarding what
can be sent through the cc:Mail system and how long messages can be
retained. Nonetheless, you cannot control how many messages users will
send and receive, when they will send them, and, beyond a point, how
large messages can be.Here are some of things you can do to keep server
disk space utilization under control:
1. Establish policies regarding what types of files
can be sent through the cc:Mail system
2. Institute cc:Mail Router Call List message size limits to control
inter-post office message sizes
3. Establish message retention policies and procedures for users
4. Selectively purge agreed upon messages, such as message log, trash
folder and inbox messages, from user mailboxes using administrative
tools
5. Monitor user mailbox sizes using administrative tools and work
with users to remove unnecessary data from the post office
Capacity Planning for the cc:Mail System
Within a cc:Mail system, capacity planning means more
than determining an adequate number of cc:Mail Routers and e-mail gateways.
Ideally, the system would be designed to support specific service levels.
Installation requirements for cc:Mail Routers and e-mail
gateways as well as for user workstations are straightforward. The numbers
of users, post offices, cc:Mail Routers and e-mail gateways that will
be part of the cc:Mail system are part of the system design. Performance,
however, is difficult to predict. If you wanted to determine the number
of cc:Mail Routers that you would need based on an arbitrary end-to-end
delivery time for messages, you would have to know how many messages
users will send, when they will send them, and what sizes the messages
will be.
Grappling with cc:Mail system performance, it is possible
to over-engineer the cc:Mail system itself but I don’t generally recommend
this. If there is a need for high-speed message delivery throughout
the cc:Mail system, a special message routing system can be put in place
only for urgent messages.
Breakdown of cc:Mail System Performance
Performance can be defined in terms of user mail application
response times at the front-end and end-to-end message delivery times
at the back-end. Quick response at the desktop depends on several factors.
I have listed the main factors below, ranked according to overall impact.
The factors that are most significant, however, are different at different
times. A slow server, for example, may be the most significant factor
overall, but running an on-line reclaim (Release 6 and later) will degrade
performance temporarily but to a greater extent.
User mail application response times depend on these
factors:
1. File server and network performance
2. Workstation CPU class, amount of RAM, and speed
3. The number of concurrent users, cc:Mail Routers, and so forth accessing
the post office
4. For versions prior to Release 6, amount of message traffic and
frequency post office maintenance
5. For Release 6 and later, whether maintenance is being performed
on-line.
The performance of cc:Mail Routers and e-mail gateways
is influenced by all of the factors that affect user mail applications.
Within the cc:Mail system, end-to-end delivery times generally depend
on these factors:
1. Number of post offices and cc:Mail Routers
2. cc:Mail message routing topology and the number of hops
3. cc:Mail Router Call List configurations
4. Amount of message traffic
5. Type and speed of data communications equipment used for cc:Mail
Routers
For cc:Mail Routers and e-mail gateways use 33MHz or
faster 80386-based IBM or compatible machines. If you are running cc:Mail
Multisession Router under OS/2, you’ll need at least 16MB of RAM or
more. However, DOS-based back-end components, such as cc:Mail Router
for DOS, use DOS conventional memory (the first 640K within the first
1MB or RAM)
TIP: For DOS machines running e-mail gateways that
use a local hard drive for transient data, use 4MB of RAM and run SMARTDRV.EXE
for local disk caching. If you need to load multiple
Implementation Planning
Administration
No one who is technical is going to want to just maintain
post offices. After a while of maintenance including the normal adding/deleting,
a technical person will become bored with just cc:Mail administration.
Generally as your users become more familiar with cc:Mail,
they require less help. If they are your more sophisticated users and
want to take advantage of all the features and have numerous options
(aliases, bankshot routing, Internet addressing, and anything else they
can think of), then it can be a full time job even with 400 users. So
it depends on how demanding your users are regardless of just size.
I would always recommend having at least a part time
back-up. Usually a technical person who acts as a network administrator
can do this. Assuming that this person will only be called on occasionally
and not for everything mail related.
Making improvements to the mail system takes time,
and maintenance takes more time. When DB8 comes out, it will take less
time.
We are a very large site. We have about 5 persons just
for cc:Mail related work. It does involve some network administrator
work, but only to support the mail. We are constantly making changes
both to the directory and to the system to improve it. We have very
demanding users that want to do everything. The through-put is very
high as well. They want the messages to get around the country fast.
This takes up a lot of time.
We have over 1,500 users on 6 post offices and we leave
all the cc:Mail administrator to one person. And not only that, it isn't
his only job. The job has traded hands a bunch of times, but e-mail
administration is pretty easy once all the batch files are written to
do the maintenance.
Up until now, we have really had two cc:Mail administrators
- one was the network supervisor who had responsibility for the entire
network including installation/setting up of cc:Mail post offices, router,
gateway - the more, what I consider, technical side of the administration.
The other responsibilities (my job) include user support (425 users,
6 post offices). I add/change/remove users, unlock accounts, train,
and install at the user's workstation (including mobile setups) and
do all the support. I'm the user contact person and only call in the
network person when needed. Our network supervisor left and we're kind
of in limbo now, although we do have two other network people that have
been able to take over and assist me as needed. Fortunately, we haven't
had the need to setup any new post offices or anything.
My question is how do other cc:Mail sites handle their
administration - do you have one person to do it all? Or do you have
a help desk function or what? Any information would be helpful to me.
Implementation Planning
Number of users
Number of post offices
Number of users per post office
System Design
Number of users
Number of post offices
Number of users per post office
User Naming Conventions
Post Office Naming Conventions
Selecting a cc:Mail routing topology
Directory propagation topologies
Dial-up cc:Mail and dial-up network access
Choosing cc:Mail Router data communications transports
Physical location of cc:Mail Routers and e-mail gateways
Bridging of LANs and network operating systems
Post Office Naming Conventions
You can save yourself some headache later if you take
a minute to create a scaleable post office naming convention. Post offices
are usually named in straightforward ways easily understood by users
and administrators. A telecommunications company called Two Cans and
a String, Inc., for example, might name the first post office TCSI,
TCSI-01, or simply TwoCans. Other post offices might be named according
to department or location, such as TCSI-SALES or TCSI-BOSTON.
A larger organization may have one or more post offices
used for message routing or communication with external e-mail systems.
These ‘hub’ post offices usually indicate the company name and geographic
location.
Although a post office name could contain characters
or strings of characters such as ?, $, "",
$_$P$G, or %1, these characters cannot easily be used
in OS/2 command files, DOS batch files or UNIX shell scripts. Some of
these characters cannot be used on a command line because they have
special meanings for the command interpreter or shell (COMMAND.COM for
OS/2 and DOS and usually /bin/sh or /bin/csh in UNIX). Characters such
as @, }. (, or [ can be used but these characters can be problematic
when exchanging mail with other types of e-mail systems that assign
special meanings to these characters. As a rule of thumb, only numbers
and letters should be used in post office names. The dash (-), underscore
(_), and tilde (~) characters, however, are often used. Spaces on the
other hand are almost never used and I do not recommend using them.
Ideally, post office naming conventions should be simple
and understandable. On the other hand, if every organization used names
like ACCOUNTING and E-MAIL it would be very difficult to connect cc:Mail
systems across organizations. Post office names should be as unique
as possible so long as they remain meaningful. From an administrative
point of view consistency and meaningfulness are paramount while for
users names like SALES and MANUFACTURING are preferable. Use your judgment.
User Naming Conventions
I recommend entering names last name first as in "Smith,
John." Enter the last name, a comma, then a space and then the
first name and press the Enter key. Other naming conventions such as
first-name-last-name ("John Smith"), initials ("JS")
or first initial and the first seven characters of the last name ( "jsmith")
are allowed. As with post office names, you should avoid using characters
other than alphanumeric characters.
- Number of cc:Mail Routers
- Physical location of cc:Mail Routers and e-mail
gateways
- Selecting a cc:Mail routing topology
- Choosing cc:Mail Router data communications transports
- E-mail bridging of LANs and network operating systems
- Dial-up cc:Mail vs. dial-up network access
The cc:Mail Mobile products can communicate with the
cc:Mail Router using telephone modems or LAN protocols such as TCP/IP
or IPX. A Router-to-Mobile connection, however, never involves a user
mail application accessing the post office files over a LAN.
Most cc:Mail users access a post office over a LAN,
as opposed to using cc:Mail Mobile. The cc:Mail Mobile product can use
a modem or a LAN protocols to communicate with a cc:Mail Router but
the regular, LAN-based user mail applications, like cc:Mail for Windows
cannot. The LAN-based user mail applications do not directly use LAN
protocols to access the post office files. File sharing services are
provided by other software that does not directly interface with the
user mail applications. Although, all network operating systems use
LAN protocols, it would not be correct to say that cc:Mail uses a particular
protocol.
Although cc:Mail Router and Mobile can interface directly
with LAN protocols like Novell’s SPX, it’s important to understand that
the LAN-based user mail applications do not. This can be confusing because
the cc:Mail Mobile products can function both in file sharing mode and
in Mobile mode. When cc:Mail Mobile is run in file sharing mode it uses
the same mechanisms used by user mail applications such as cc:Mail for
OS/2 Workplace Shell.
The ability of the cc:Mail Mobile products to run in
both a file sharing mode and in a Mobile mode sometimes results in a
misunderstanding of how the cc:Mail system works. As a result, administrators
sometimes mistake the protocols used by their network operating systems
as the mechanism by which the user mail applications access the post
office files. For example, when file sharing services are provided by
the Microsoft Lan Manager network operating system, you might think
that that "cc:Mail uses the NetBEUI LAN protocol." While this
is true in a sense, the user mail applications function independently
of underlying network protocols. They rely on file sharing services
provided by other software.
Remember that the cc:Mail post office is not a server
or server processes in a client/server system but merely a set of shared
files. The user mail applications read and write to the post office
files using the same higher-level mechanisms regardless of what network
operating system software and underlying protocols allow the post office
files to be shared.
Messaging with foreign and legacy e-mail systems
The comment field is used by the LMX (formerly SoftSwitch
Central) e-mail gateway and by Lotus Organizer. Organizer right justifies
user information leaving the left porting of the comment field free
for normal use. The LMX gateway places a mainframe DGN.DEN address in
the comment field. The DGN.DEN address is left justified.
Directory synchronization with foreign and legacy
e-mail systems
A list of user names from a mainframe host or other
system can be used as the basis for constructing a cc:Mail Import file
to populate your cc:Mail directory. The cc:Mail Import file format is
a simple ASCII text file format defined in the cc:Mail Import/Export
manual. A list of user names, one per line, can be converted into a
cc:Mail Import file in a variety of ways, such as using a UNIX shell
script or a small program.
If you are using a version of cc:Mail prior to Release
6 and you are using Novell NetWare or Banyan VINES you can populate
your cc:Mail directory using the cc:Mail Directory Services program
(refer to the documentation that came with your cc:Mail products for
more information).
Number of User Per Post Office
In practice, many implementations had more users per
post office with no negative consequences. What’s good about the 100
user number is that smaller post offices take less time to maintain
and fewer users are affected if the post office is shut down. However,
in larger systems, the number of 100-user post offices can become unmanageable.
For versions prior to Release 6, the maximum server
volume size supported by the cc:Mail maintenance utilities was 2GB.
This means that with a server disk space allocation of 10MB per user,
there could be approximately 210 users per post office. Since messages
temporarily accumulate in queues for other post offices and because
free spaces were not 100% reusable between maintenance cycles, 200 users
per post office is a good rule of thumb.
For Release 6, the maximum volume size doubled from
2GB to 4GB. This means that, without multiple message stores per post
office, a reasonable maximum number of users per post office is still
200.
Next
Chapter...
Back
to Contents
©1996, 1997 by Global System Services Corporation (GSS)
Portions of this material are ©1995 by Ron Herardian
DISSEMINATION, DISTRIBUTION OR COPY OF THIS INFORMATION
IS STRICTLY PROHIBITED. NO PART OF THE INFORMATION CONTAINED HEREIN
MAY BE REPRODUCED IN ANY FORM BY ANY ELECTRONIC OR MECHANICAL MEANS,
INCLUDING PHOTOCOPYING, RECORDING, OR INFORMATION STORAGE AND RETRIEVAL,
WITHOUT THE WRITTEN PERMISSION OF GLOBAL SYSTEM SERVICES CORPORATION.