docs/diploma

annotate thesis/tex/3-MailTransferAgents.tex @ 188:afb72fb64962

worked on discussion on architecture
author meillo@marmaro.de
date Mon, 29 Dec 2008 21:43:52 +0100
parents 5f7beb2142d6
children ee687abf386c
rev   line source
meillo@89 1 \chapter{Mail transfer agents}
meillo@89 2
meillo@117 3 After having analyzed the market for electronic mail and trends for it in the last chapter, this chapter takes a look at \mta{}s, the intelligent most important parts of the email infrastructure. \MTA{}s will be grouped by similarities, and the four most popular \freesw\ \mta{}s, will be presented to the reader in a short overview and with the most important facts. At the end of this chapter a comparison of these programs in several disciplines will be done.
meillo@89 4
meillo@117 5
meillo@89 6
meillo@89 7
meillo@120 8 \section{Types of MTAs}
meillo@117 9 ``Mail transfer agent'' is a term covering a variety of programs. One thing is common to them: they transfer email from one machine to another.
meillo@89 10
meillo@143 11 This is how Bryan \person{Costales} defines a \mta\ in \cite{costales97}:
meillo@117 12 \begin{quote}
meillo@117 13 A mail transfer agent (MTA) is a highly specialized program that delivers mail and transports it between machines, like the post office.
meillo@117 14 \end{quote}
meillo@117 15 \name{The Free Dictionary} is a bit more concrete on the term: \citeweb{website:thefreedictionary}
meillo@117 16 \begin{quote}
meillo@117 17 Message Transfer Agent - (MTA, Mail Transfer Agent): Any program responsible for delivering e-mail messages. Upon receiving a message from a Mail User Agent or another MTA, [...] it [...] delivers it to any local addressees and/or forwards it to other remote MTAs (routing) for delivery to remote recipients.
meillo@117 18 \end{quote}
meillo@89 19
meillo@117 20 Common to all \MTA{}s is the transfer of mail to other machines; this is the actual job. Besides this similarity, \MTA{}s can be very different. Some of them have \NAME{POP3} and/or \NAME{IMAP} servers included. Some can fetch mails through these protocols. Others have have all features you can think of. And maybe there are some that do nothing else but transporting email.
meillo@89 21
meillo@117 22 Following is a classification of \mta{}s into groups of similar programs, regarding what is viewable from the outside.
meillo@117 23
meillo@117 24
meillo@120 25 \subsubsection*{Relay-only MTAs}
meillo@89 26 \label{subsec:relay-only}
meillo@89 27 This is the most simple kind of \MTA. It transfers mail only to defined \name{smart hosts}\footnote{\name{smart host}s are \MTA{}s that receives email and route it to the actual destination}. \name{Relay-only} \MTA{}s do not receive mail from outside the system, and they do not deliver locally.
meillo@89 28
meillo@89 29 Most \MTA{}s can be configured to act as such a \name{forwarder}. But this is usually an additional functionality.
meillo@89 30
meillo@89 31 One would use such a program to give a system the possibility to send mail, without the need to do lots of configuration. In a local network, usually the clients are set up with a \name{relay-only} \MTA, while there is one \name{mail server} that acts as a \name{smart host}. The ``dumb'' clients send mail to this one \name{mail server} which does all the work.
meillo@89 32
meillo@89 33 Examples for that group are: \name{nullmailer}, \name{ssmtp} and \name{esmtp}.
meillo@89 34
meillo@89 35
meillo@117 36 \subsubsection*{Groupware}
meillo@124 37 Normally the term ``groupware'' does not mean one single program, but a suite of programs. They build a framework which is then populated with various modules that provide actual functionality. Modules for mail transfer, file storage, calendars, resource management, instant messaging, etc., are commonly available.
meillo@89 38
meillo@148 39 These program suites are used if the main work to do is providing integrated communication facilities and team working support for a group of people. Mail transfer is only part of the problem to solve. The most common scenario are companies. They have \name{groupware} running to provide adequate services for their teams to work efficiently. But one may use \name{groupware} on the home server for his family members also.
meillo@89 40
meillo@89 41 Examples are: \name{Lotus Notes}, \name{Microsoft Exchange}, \name{OpenGroupware.org} and \name{eGroupWare}.
meillo@89 42
meillo@89 43
meillo@120 44 \subsubsection*{``Real'' MTAs}
meillo@117 45 There is a third type of \mta{}s in between the minimalistic \name{relay-only} \MTA{}s and the bloated \name{groupware}. Those programs may be named ``real \MTA{}s'', or ``proper \MTA{}s'', though there is no common name. They are what is meant with the term ``\mta''---programs that transfer mail between hosts.
meillo@89 46
meillo@132 47 Common to them is their focus on transferring email, while being able to act as \name{smart host}. Their variety ranges from ones mostly restricted to mail transfer (\qmail) to others already having interfaces for adding further mail processing modules (\postfix). They cover everything in between the other two groups. %FIXME: are postfix and qmail good examples?
meillo@89 48
meillo@117 49 This group is of importance in this document. All programs selected for the comparison in the following section are ``real \MTA{}s''. \masqmail\ is one too.
meillo@89 50
meillo@89 51
meillo@117 52 \subsubsection*{Other segmenting}
meillo@124 53 \name{Mail transfer agents} can also be split in other ways.
meillo@117 54 \begin{itemize}
meillo@117 55 \item
meillo@124 56 Due to \sendmail's significance---described in section \ref{sec:sendmail}---compatibility interfaces for \sendmail\ are of importance for \unix\ \MTA{}s. Being not \emph{sendmail-compatible} does not need to matter for some fields of action, but makes the program ineligible for serving as a general purpose \MTA\ on \unix\ systems. Hence being sendmail-compatible is a major property of a \mta. %todo: how many MTAs are sendmail-compatible?
meillo@124 57 \MTA{}s not having a \emph{sendmail-compatible} interface or not offering it as a compatibility add-on, will not be covered here. One example for such a program is \name{Apache James}. %FIXME: check if correct
meillo@89 58
meillo@117 59 \item
meillo@117 60 Another separation can be done between \freesw\ programs and proprietary software. Many of the \MTA{}s for \unix\ systems are \freesw. Only these are regarded in the following sections, because comparing \freesw\ with proprietary or commercial software is not what typical users of programs like \masqmail\ do. %fixme: what are typical users?
meillo@117 61 Comparison with those non-free programs may be a point for large \freesw\ projects, trying to step into the business world. Small projects, mostly used by individuals at home, %fixme: is this the right target field? see chap02
meillo@119 62 need to be compared against other projects of similar shape. The document should be seen from \masqmail's point of view---an \MTA\ for a \unix\ system on home servers, workstations, or maybe embedded platforms---so non-free software is out of the way.
meillo@117 63 \end{itemize}
meillo@89 64
meillo@89 65
meillo@89 66
meillo@89 67
meillo@120 68 \section{Popular MTAs}
meillo@89 69
meillo@130 70 This section introduces a selection of popular \MTA{}s; they are the most likely substitutes for \masqmail. All are \emph{sendmail-compatible} ``smart'' \freesw\ \MTA{}s that focus on mail transfer, as is \masqmail.
meillo@89 71
meillo@143 72 The programs chosen are: \sendmail, \exim, \qmail, and \postfix. They are the most important representatives of the regarded group. Although \MTA\ statistics are rare, differ, and good data is hard to collect, these programs tend to stay near the top.
meillo@117 73
meillo@145 74
meillo@145 75 \subsection{Market share analysis}
meillo@145 76
meillo@155 77 Table \ref{tab:mta-market-share} shows the most used \MTA{}s determined by three different statistics. The first was done by Daniel~J.\ \person{Bernstein} (the author of \qmail) in 2001 \cite{bernstein01}. The second is by \person{Simpson} and \person{Bekman} in 2007 and was published by \name{O'ReillyNet} \cite{simpson07}. And the third is from \name{MailRadar.com} with unknown date\footnote{The footer of the website shows ``Copyright 2007'' but more likely does this refer to the whole website.} \citeweb{mailradar:mta-stats}.
meillo@117 78
meillo@130 79 \begin{table}
meillo@130 80 \begin{center}
meillo@130 81 \input{input/mta-market-share.tex}
meillo@130 82 \end{center}
meillo@130 83 \caption{Market share of \MTA{}s}
meillo@130 84 \label{tab:mta-market-share}
meillo@130 85 \end{table}
meillo@89 86
meillo@145 87 All surveys show high market shares for the four \MTA{}s: \sendmail, \exim, \qmail, and \postfix. Only the \name{Microsoft} mail server software and \name{IMail} have comparable large shares. Other \freesw\ \mta{}s are: \name{smail}, \name{zmailer}, \name{MMDF}, and \name{courier-mta}. They all are less important and rarely used.
meillo@130 88
meillo@145 89 The three surveys base on different data. \person{Bernstein} took 1,000,000 randomly chosen \NAME{IP} addresses, containing 39,206 valid hosts; 958 of them accepted \NAME{SMTP} connections. The \person{Simpson} and \person{Bekman} survey used only domains owned by companies; in total 400,000 hosts. \name{MailRadar} scanned 2,818,895 servers, leading to 59,209 accepted connections.
meillo@130 90
meillo@145 91 All surveys show \sendmail\ to be the most popular \MTA. \postfix, \qmail, and \exim\ are among the best seven in each. \exim\ has lightly smaller shares than the other two. The four together share more than half of the market according to \person{Bernstein} and the \name{MailRadar} statistics. \person{Simpson} and \person{Bekman} have their share to be somewhere between a third and the half. The reason for this uncertainty follows.
meillo@143 92
meillo@145 93 Remarkable are the 22 percent of ``mail security layers''\footnote{\name{mail security layers} are software guards between the network and the \mta\ that filter unwanted mail before it reaches the \MTA. This increases security by filtering malicious content and by blocking attacks against the \MTA. It also decreases the work load of the guarded \MTA.} in the \name{O'Reilly} survey. This large size of this share may be a result of only regarding business mail servers. The problem concerning the survey is the deguise of the \mta\ working behind the security layer. It seems wrong to assume equal shares for the \MTA{}s behind the guards as for the unguarded \MTA{}s, because they will be more often used to guard weak \MTA{}s, as strong ones do not need them so much. This needs to be kept in mind when using the \name{O'Reilly} survey.
meillo@145 94
meillo@145 95 It seems quite sure that the \name{MailRadar} statistics were published after 2001, caused by the \sendmail\ and \postfix\ shares. But to decide whether before or after the one from \name{O'Reilly} would be just a guess.
meillo@145 96
meillo@145 97
meillo@145 98 \subsection{The four major Free Software MTAs}
meillo@143 99
meillo@155 100 Now follows a small introduction to the five programs chosen for comparison, except \masqmail\ which already was introduced in chapter \ref{chap:introduction}. Longer introductions, including analysis and comparison, were written by Jonathan de \person{Boyne Pollard} \cite{jdebp}.
meillo@89 101
meillo@117 102
meillo@117 103
meillo@120 104 \subsubsection*{sendmail}
meillo@89 105 \label{sec:sendmail}
meillo@117 106 \sendmail\ is the most popular \mta, since it was one of the first and was shipped as default \MTA{}s by many vendors of \unix\ systems. %fixme: ref
meillo@89 107
meillo@143 108 The program was written by Eric \person{Allman} as the successor of his program \name{delivermail}. \sendmail\ was first released with \NAME{BSD} 4.1c in 1983. Allman was not the only one working on the program. Other people developed own versions of it and a variety of flavors came up, especially in the late eighties when Allman was inactive. %fixme: ref
meillo@89 109
meillo@124 110 \sendmail\ is focused on transferring mails between different protocols and networks, this lead to a very flexible (though complex) configuration.
meillo@89 111
meillo@89 112 The latest version is 8.14.3 from May 2008. The program is distributed under the \name{Sendmail License} as both, \freesw\ and proprietary software of \name{Sendmail, Inc.}.
meillo@89 113
meillo@128 114 Further development will go into the project \name{MeTA1} (the former name was \name{sendmail X}) which succeeds \sendmail.
meillo@89 115
meillo@130 116 More information can be found on the \sendmail\ homepage \citeweb{sendmail:homepage}.
meillo@89 117
meillo@89 118
meillo@117 119
meillo@120 120 \subsubsection*{exim}
meillo@117 121 \label{sec:exim}
meillo@143 122 \exim\ was started in 1995 by Philip \person{Hazel} at the \name{University of Cambridge}. It is forked of \name{smail-3}, and inherited the monolithic architecture, similar to \sendmail's. But having no separation of the individual components of the system, like \qmail\ and \postfix\ have, did not hurt. Its security is comparably good. %fixme: ref
meillo@117 123
meillo@132 124 \exim\ is highly configurable, especially in the field of mail policies. This makes it easy to specify how mail is routed through the system and who is allowed to send email to whom. Also interfaces for integration of virus and spam check programs are provided by design. %fixme: ref
meillo@117 125
meillo@117 126 The program is \freesw, released under the \GPL. The latest stable version is 4.69 from December 2007.
meillo@117 127
meillo@132 128 One finds \exim\ on its homepage \citeweb{exim:homepage}.
meillo@117 129
meillo@117 130
meillo@117 131
meillo@120 132 \subsubsection*{qmail}
meillo@89 133 \label{sec:qmail}
meillo@132 134 \qmail\ is seen by its community as ``a modern SMTP server which makes sendmail obsolete''.%fixme: ref
meillo@143 135 It was written by Daniel~J.\ \person{Bernstein} starting in 1995. His primary goal was to create a secure \MTA\ to replace the popular, but vulnerable, \sendmail. %fixme: ref
meillo@89 136
meillo@132 137 \qmail\ first introduced many innovative concepts in \mta\ design and is generally seen as the first security-aware \MTA\ developed. %fixme:ref
meillo@117 138 %fixme: what about mmdf?
meillo@89 139
meillo@132 140 Since November 2007, \qmail\ is released in the \name{public domain} which makes it \freesw. The latest release is 1.03 from July 1998.
meillo@89 141
meillo@155 142 The programs homepages are \citeweb{qmail:homepage1} and \citeweb{qmail:homepage2}. Further information about \qmail\ is available with Dave \person{Sill}'s ``Life with qmail'' \cite{lifewithqmail}.
meillo@89 143
meillo@89 144
meillo@117 145
meillo@120 146 \subsubsection*{postfix}
meillo@89 147 \label{sec:postfix}
meillo@143 148 The \postfix\ project was started in 1999 at \name{IBM research}, then called \name{VMailer} or \name{IBM Secure Mailer}. Wietse \person{Venema}'s program ``attempts to be fast, easy to administer, and secure. The outside has a definite Sendmail-ish flavor, but the inside is completely different.''\citeweb{postfix:homepage} In fact, \postfix\ was mainly designed after qmail's architecture to gain security. But in contrast to \qmail\ it aims much more on being fast and full-featured.
meillo@89 149
meillo@132 150 Today \postfix\ is taken by many \unix\ systems and \gnulinux\ distributions as default \MTA.
meillo@89 151
meillo@132 152 The latest stable version is numbered 2.5.5 from August 2008. \postfix\ is covered by the \name{IBM Public License 1.0} which is a \freesw\ license.
meillo@89 153
meillo@132 154 Additional information can be retrieved from the program's homepage \citeweb{postfix:homepage}.
meillo@89 155
meillo@89 156
meillo@89 157
meillo@89 158
meillo@89 159
meillo@89 160
meillo@120 161 \section{Comparison of MTAs}
meillo@89 162
meillo@155 163 This section does not try to provide an overall \MTA\ comparison, because this is already done by others. Remarkable comparisons are the one by Dan \person{Shearer} \cite{shearer06} and a discussion on the mailing list \name{plug@lists.q-linux.com} \cite{plug:mtas}. Tabulary overviews may be found at \citeweb{mailsoftware42}, \citeweb{wikipedia:comparison-of-mail-servers}, and \cite[section 1.9]{lifewithqmail}.
meillo@89 164
meillo@128 165 Here provided is an overview on a selection of important properties, covering the four previously introduced programs. The data comes from the above stated sources and is collected in table \ref{tab:mta-comparison}.
meillo@126 166
meillo@117 167 \begin{table}
meillo@126 168 \begin{center}
meillo@126 169 \input{input/mta-comparison.tex}
meillo@126 170 \end{center}
meillo@126 171 \caption{Comparison of MTAs}
meillo@126 172 \label{tab:mta-comparison}
meillo@117 173 \end{table}
meillo@89 174
meillo@89 175
meillo@132 176 \subsection{Architecture}
meillo@89 177
meillo@132 178 Architecture is most important when comparing \MTA{}s. Many other properties of a program depend on its architecture. %fixme: add ref?
meillo@155 179 Munawar \person{Hafiz} \cite{hafiz05} discusses in detail on \mta\ architecture, comparing \sendmail, \qmail, \postfix, and \name{sendmail X}. Jonathan de \person{Boyne Pollard}'s \MTA\ review \cite{jdebp} is a source too.
meillo@89 180
meillo@132 181 Two different architecture types show off: monolithic and modular \mta{}s.
meillo@130 182
meillo@132 183 Monolithic \MTA{}s are \sendmail, \name{smail}, \exim, and \masqmail. They all consist of one single \emph{setuid root}\footnote{\emph{setuid root} lets a program run with the rights of its owner, here root. This is considered a security risk.} binary which does all the work.
meillo@130 184
meillo@132 185 Modular \MTA{}s are \NAME{MMDF}, \qmail, \postfix, and \name{MeTA1}. They consist of several programs, each doing a part of the overall job. The different programs run with the least permissions the need, and \emph{setuid root} needs not to be used.
meillo@130 186
meillo@155 187 The architecture does not directly define the program's security, but ``[t]he goal of making a software secure can be better achieved by making the design simple and easier to understand and verify''\cite[chapter 6]{hafiz05}. \exim, though being monolithic, has a fairly clean security record. But it is very hard to keep the security up, as the program growth. Wietse \person{Venema} (the author of \postfix) says, the architecture enabled \postfix\ to grow without running into security problems. \cite[page 13]{venema:postfix-growth}
meillo@130 188
meillo@140 189 The modular design, with each sub-program doing one part of the overall job, is applied \name{Unix Philosophy}. The Unix Philosophy \cite{gancarz95} demands ``small is beautiful'' and ``make each program do one thing well''. Monolithic \MTA{}s fail here.
meillo@130 190
meillo@132 191 Today modular \mta\ architectures are the state-of-the-art.
meillo@89 192
meillo@89 193
meillo@89 194
meillo@132 195 \subsection{With focus on the future}
meillo@89 196
meillo@132 197 Section \ref{sec:what-will-be-important} tried to figure out the importances for future \MTA{}s. The four programs are compared on these (possible) future requirements now.
meillo@89 198
meillo@140 199 The first trend was provider independence, requiring easy configuration. \postfix\ seems to do best here. It used primary two configuration files (\path{master.cf} and \path{main.cf}) which are easy to manage. \sendmail\ appears to have a bad position. Its configuration file \path{sendmail.cf} is very complex, including Turing-completeness, thus it needs simplification wrappers around it to provide easier configuration. There exist the \name{m4} macros to generate \path{sendmail.cf}, but adjusting the generated result by hand seems to be nessesary for non-trivial configurations. \qmail's configuration files are simple, but the whole system is complex to set up; it requires various system users and is hardly usable without applying several patches to add basic functionality. \name{netqmail} is the community effort to help here. \exim\ has only one single configuration file (\path{exim.conf}), but it suffers most from its flexibility, like \sendmail. Flexibility and easy configuration are contrary.
meillo@89 200
meillo@132 201 As second trend, the decreasing nessesarity for high performance was identified. This goes along with the move of \MTA{}s from service providers to home servers. \postfix\ focuses much on performance, this might not be an important point then. Of course there still will be the need for high performance \MTA{}s, but a growing share of the market will not require high performance. Performance is related to simplicity, which effects security. Increasing performance does in most times decrease the other two. Simple \mta{}s not aiming for highest performance are what is needed in future. The simple of \qmail, still being fast enough, seems to be a good example.
meillo@126 202
meillo@132 203 The third trend, even more security awareness, is addressed by each of the four programs. It seems as if all widely used \mta{}s provide good security nowadays---even \sendmail\ can be considered secure today. %fixme:ref
meillo@132 204 The modular architecture, used by \qmail\ and \postfix, is generally seen to be conceptually more secure, however.%fixme: ref
meillo@132 205 \sendmail's creators have started \name{MeTA1}, a modular \MTA\ merging the best of \qmail\ and \postfix, to replace the old \sendmail. It will be interesting to watch \exim's future---will it become modular too?
meillo@126 206
meillo@126 207
meillo@93 208
meillo@89 209
meillo@89 210
meillo@117 211
meillo@117 212
meillo@132 213 %todo: my own poll (?)
meillo@117 214
meillo@117 215
meillo@132 216 %<< complexity >> << security >> << simplicity of configuration and administration >> << flexibility of configuration and administration >> << code size >> << code quality >> << documentation (amount and quality) >> << community (amount and quality) >> << used it myself >> << had problems with it >>
meillo@117 217
meillo@117 218
meillo@132 219 %<< quality criteria >> << standards of any kind >> << how to compare? >> << (bewertungsmatrix) objectivity >> << how many criteria for ``good''? >>
meillo@133 220