3proxy/doc/html/highload.html

<h3>Optimizing 3proxy for high load</h3>
<p>Precaution 1: 3proxy was not initially developed for high load and is positioned as a SOHO product, the main reason is "one connection - one thread" model 3proxy uses. 3proxy is known to work with above 200,000 connections under proper configuration, but use it in production environment under high loads at your own risk and do not expect too much.
<p>Precaution 2: This documentation is incomplete and is not sufficient. High loads may require very specific system tuning including, but not limited to specific or cusomized kernels, builds, settings, sysctls, options, etc. All this is not covered by this documentation.
<h4>Configuring 'maxconn'</h4>
A number of simulatineous connections per service is limited by 'maxconn' option.
Default maxconn value since 3proxy 0.8 is 500. You may want to set 'maxconn'
to higher value. Under this configuration:
<pre>
maxconn 1000
proxy -p3129
proxy -p3128
socks
</pre>
maxconn for every service is 1000, and there are 3 services running
(2 proxy and 1 socks), so, for all services there can be up to 3000
simulatineous connections to 3proxy.
<p>Avoid setting 'maxconn' to arbitrary high value, it should be carefully
choosen to protect system and proxy from resources exhaution. Setting maxconn
above resources available can lead to denial of service conditions.
<h4>Understanding resources requirements</h4>
Each running service require:
<ul>
<li>1*thread (process)
<li>1*socket (file descriptor)
<li>1 stack memory segment + some heap memory, ~64K-128K depending on the system
</ul>
Each connected client require:
<ul>
<li>1*thread (process)
<li>2*socket (file descriptor). For FTP 4 sockets are required. 
<br>Under linux since 0.9 splice() is used. It much more effective, but requires
<br>2*socket (file descriptor) + 2*pipe (file descriptors) = 4 file descriptors.
<br>For FTP 4 sockets and 2 pipes are required with splice().
<br>1 additional socket (file descriptor) during name resolution for non-cached names
<br>1 additional socket during authentication or logging for RADIUS authentication or logging.
<li>1*ephemeral port (3*ephemeral ports for FTP connection).
<li>1 stack memory segment of ~32K-128K depending on the system + at least 16K and up to few MB (for 'proxy' and 'ftppr') of heap memory. If you are short of memory, prefer 'socks' to 'proxy' and 'ftppr'.
<li>a lot of system buffers, specially in the case of slow network connections.
</ul>
Also, additional resources like system buffers are required for network activity.
<h4>Setting ulimits</h4>
Hard and soft ulimits must be set above calculated requirements. Validate
ulimits match your expectation, especially if you run 3proxy under dedicated account
by adding e.g.
<pre>
system "ulimit -Ha >>/tmp/3proxy.ulim.hard"
system "ulimit -Sa >>/tmp/3proxy.ulim.soft"
</pre>
in the beginning (before first service started) and the end of config file.
Make both hard restart (that is kill and start 3proxy process) and soft restart
by sending SIGUSR1 to 3proxy process, check ulimits recorded to files match your
expecation.
<h4>Extending system limitation</h4>
Check manuals / documentation for your system limitations. You may need to change
sysctls or even rebuild the kernel from source.
To help with system-dependant settings, since 0.9-devel 3proxy supports different
socket options which can be set via -ol option for listening socket, -oc for proxy-to-client
socket and -os for proxy-to-server socket. Example:
<pre>
proxy -olSO_REUSEADDR,SO_REUSEPORT -ocTCP_TIMESTAMPS,TCP_NODELAY -osTCP_NODELAY
</pre>
available options are system dependant.
<h4>Extending ephemeral port range</h4>
Check ephemeral port range for your system and extend it to the number of the
ports required.
Ephimeral range is always limited to maximum number of ports (64K). To extend the
number of outgoing connections above this limit, extending ephemeral port range
is not enough, you need additional actions:
<ol>
<li> Configure multiple outgoing IPs
<li> Make sure 3proxy is configured to use different outgoing IP by either setting
external IP via RADIUS
<pre>
radius secret 1.2.3.4
auth radius
proxy
</pre>
or by using multiple services with different external
interfaces, example:
<pre>
allow user1,user11,user111
proxy -p1001 -e1.1.1.1
flush
allow user2,user22,user222
proxy -p1001 -e1.1.1.2
flush
allow user3,user33,user333
proxy -p1001 -e1.1.1.3
flush
allow user4,user44,user444
proxy -p1001 -e1.1.1.4
flush
</pre>
or via "parent extip" rotation,
e.g.
<pre>
allow user1,user11,user111
parent 1000 extip 1.1.1.1 0
allow user2,user22,user222
parent 1000 extip 1.1.1.2 0
allow user3,user33,user333
parent 1000 extip 1.1.1.3 0
allow user4,user44,user444
parent 1000 extip 1.1.1.4 0
proxy
</pre>
or
<pre>
allow *
parent 250 extip 1.1.1.1 0
parent 250 extip 1.1.1.2 0
parent 250 extip 1.1.1.3 0
parent 250 extip 1.1.1.4 0
socks
</pre>
<pre>
</pre>
Under latest Linux version you can also start multiple services with different
external addresses on the single port with SO_REUSEPORT on listening socket to
evenly distribute incoming connections between outgoing interfaces:
<pre>
socks -olSO_REUSEPORT -p3128 -e 1.1.1.1
socks -olSO_REUSEPORT -p3128 -e 1.1.1.2
socks -olSO_REUSEPORT -p3128 -e 1.1.1.3
socks -olSO_REUSEPORT -p3128 -e 1.1.1.4
</pre>
for Web browsing last two examples are not recommended, because same client can get
different external address for different requests, you should choose external
interface with user-based rules instead.
<li> You may need additional system dependant actions to use same port on different IPs,
usually by adding SO_REUSEADDR (SO_PORT_SCALABILITY for Windows) socket option to
external socket. This option can be set (since 0.9 devel) with -os option:
<pre>
proxy -p3128 -e1.2.3.4 -osSO_REUSEADDR
</pre>
Behavior for SO_REUSEADDR and SO_REUSEPOR is different between different system,
even between different kernel versions and can lead to unexpected results.
Specifics is described <a href="https://stackoverflow.com/questions/14388706/socket-options-so-reuseaddr-and-so-reuseport-how-do-they-differ-do-they-mean-t">here</a>.
Use this options only if actually required and if you fully understand possible
consiquences. E.g. SO_REUSEPORT can help to establish more connections than the
number of the client port available, but it can also lead to situation connections
are randomely fail due to ip+port pairs collision if remote or local system 
doesn't support this trick.
</ol>
<h4>Setting stacksize</h4>
'stacksize' is a size added to all stack allocations and can be both positive and
negative. Stack is required in functions call. 3proxy itself doesn't require large
stack, but it can be required if some
purely-written libc, 3rd party libraries or system functions called. There is known\
dirty code in Unix ODBC
implementations, build-in DNS resolvers, especially in the case of IPv6 and large
number of interfaces. Under most 64-bit system extending stacksize will lead
to additional memory space usage, but do not require actual commited memory,
so you can inrease stacksize to relatively large value (e.g. 1024000) without
the need to add additional phisical memory,
but it's system/libc dependant and requires additional testing under your
installation. Don't forget about memory related ulimts.
<p>For 32-bit systems address space can be a bottlneck you should consider. If
you're short of address space you can try to use negative stack size.
<h4>Known system issues</h4>
There are known race condition issues in Linux / glibc resolver. The probability
of race condition arises under configuration with IPv6, large number of interfaces
or IP addresses or resolvers configured. In this case, install local recursor and
use 3proxy built-in resolver (nserver / nscache / nscache6).
<h4>Do not use public resolvers</h4>
Public resolvers like ones from Google have ratelimits. For large number of
requests install local caching recursor (ISC bind named, PowerDNS recursor, etc).
<h4>Avoid large lists</h4>
Currently, 3proxy is not optimized to use large ACLs, user lists, etc. All lists
are processed lineary. In devel version you can use RADIUS authentication to avoid
user lists and ACLs in 3proxy itself. Also, RADIUS allows to easily set outgoing IP
on per-user basis or more sophisicated logics.
RADIUS is a new beta feature, test it before using in production.
<h4>Avoid changing configuration too often</h4>
Every configuration reload requires additional resources. Do not do frequent
changes, like users addition/deletaion via connfiguration, use alternative
authentication methods instead, like RADIUS.
<h4>Consider using 'noforce'</h4>
'force' behaviour (default) re-authenticates all connections after
configuration reload, it may be resource consuming on large number of
connections. Consider adding 'noforce' command before services started
to prevent connections reauthentication.
<h4>Do not monitor configuration files directly</h4>
Using configuration file directly in 'monitor' can lead to race condition where
configuration is reloaded while file is being written.
To avoid race conditions:
<ol>
<li> Update config files only if there is no lock file
<li> Create lock file then 3proxy configuration is updated, e.g. with
"touch /some/path/3proxy/3proxy.lck". If you generate config files
asynchronously, e.g. by user's request via web, you should consider
implementing existance checking and file creation as atomic operation.
<li>add
<pre>
system "rm /some/path/3proxy/3proxy.lck"
</pre>
at the end of config file to remove it after configuration is successfully loaded
<li> Use a dedicated version file to monitor, e.g.
<pre>
monitor "/some/path/3proxy/3proxy.ver"
</pre>
<li> After config is updated, change version file for 3proxy to reload configuration,
e.g. with "touch /some/path/3proxy/3proxy.ver".
</ol>
<h4>Use TCP_NODELAY to speed-up connections with small amount of data</h4>
If most requests require exchange with a small amount of data in a both ways
without the need for bandwidth, e.g. messengers or small web request,
you can eliminate Nagle's algorithm delay with TCP_NODELAY flag. Usage example:
<pre>
proxy -osTCP_NODELAY -ocTCP_NODELAY
</pre>
sets TCP_NODELAY for client (oc) and server (os) connections.
<p>Do not use TCP_NODELAY on slow connections with high delays and then
connection bandwidth is a bottleneck.
<h4>Use slice to speedup large data amount transfers</h4>
slice() allows to copy data between connections without copying to process
addres space. It can speedup proxy on high bandwidth connections, if most
connections require large data transfers. "-s" allows slice usage. Example:
<pre>
proxy -s
</pre>
Slice is only available in Linux and is currently beta option available in
devel version. Do not use it in production without testing. Slice requires
more system buffers, but reduces process memory usage.
Do not use slice if there is a lot of short-living connections with no bandwidth
requirements.
<p>Use slice only on high-speed connections (e.g. 10GBE), if processor or
bus are bottlenecks.
<p>TCP_NODELAY and slice are not contrary to each over and can be combined on
high-speed connections.