3xsd is a native epoll server serving TCP/UDP connections, a high performance static web server, a failover dns server, a http-based distributed file server, a load-balance proxy-cache server, and a 'warp drive' server. Written in python, take the full power of multi-cores.
supporting: static files, event driven(epoll), using mmap & sendfile to send files,
in-mem xcache, transparent gzip content transfer with fixed length(small file) &
chunked(large file), persistent storage of gzip files,
partial support of WebDAV(PUT/DELETE), pipelining support
supporting: only A record resolution, domainname failover(refer to conf file),
ip icmp probe & hide when fail, round robbin ip resolving
global DNS Left-Right Range Resolve(LRRR)(experimental)
supporting: load balance backend servers, in-mem file caching &
persistent cache file storage
supporting: mass unlimitted file storage, easy to expand,
O(1) location algorithm, non-centralized, can work with standard web server(WebDAV)
in proxy mode, file redundancy, file persistent caching
supporting: data tunneling over UDT and tun,
better congestion control than TCP/UDP over wan link,
better thoughput(above 80%) over wan link, refer to this report:
http://www.c-s-a.org.cn/ch/reader/create_pdf.aspx?file_no=20091035
tunnel ip/mtu/txqueuelen/route define, auto create/recreate/destroy
encrypt packages through AES-128-ECB/CBC/CFB/CTR and Blowfish-CBC/CFB/CTR
tunnel on-the-fly compress with zlib/lzo, tunnel data relaying
route metric, routing data through different path, depending on tunnel rtt(choose the best one)
More to find in .conf file.
Small file under 1KB single process test(full in-mem), contrast with nginx configuring accept_mutex off, 80% performance. Multi processes test, with reuse_port enabling kernel, 95% performance of nginx(and beyond, may be 105% or more, based on process number, I tested 2-4). The tests above is not quite strict， but I just want to say that it's fast enough.
And with pipelining enabled, 3wsd will perform better with 3-4 requests/send(5%-10% performance increase), 2 requests/send have the same speed with non-piplining.
About 80% performance of 3wsd.
Fast enough...about 2800-3000 queries/s per processes, with 1GHz bcm2709 4-cores ARMv7 cpu testing, better when multi-processes with reuse_port enabling kernel.
Same with 3zsd.
Early testing indicated that: UDT tunnel(no encrypt) performing 50%-60% speed of direct TCP connection with ZetaTCP, and package lost rate remaining below 0.6%, while direct connection has 1.4%-3%. (Test CN-US WAN link with 150ms-280ms latency, through the always-jammed CUCN submarine cable) However, UDT tunnel beats normal TCP connection without ZetaTCP, with 50% - 4 times (commonly 1-2 times) outperforming.(v)(Test link like above)
Update: And an encrypted UDT tunnel with AES-CBC/CFB will has 50% performance decrease (because the method itself processes doubled size of data, and extra iv/padding data transfer). Now with a Blowfish-CTR method, tunnel data transfer performance is closed to raw non-encrypt tunnel. I believe that with a intel AES-NI supported CPU(like XEON E3-1240/1270), AES-128-CTR can also do it.
There are at lease two ways to increase the performance of 3xsd:
1.Install Cython, and rename _3xsd.py to _3xsd.pyx, run it.
Cython will compile _3xsd.py lib into a _3xsd.so file, using static type
declarations. This can gain about 5%-6% performance increasement.
2.Use PyPy.This can gain about 10%-15% performance increasement(or more).
CentOS 6/7 with python 2.6/2.7, Debian 6/7. Python 2.7 recommended.
Doing this before running the program(minimal requirement):
yum install python-gevent pysendfile python-setproctitle python-psutil python-pip
(python-pip is optional if install dpkt)
Dpkt and geoip2 module are also needed when running 3nsd DNS server, pip install it.
If you want to use 3wdd, python-pytun, pyudt4, pycrypto, python-lzo are also needed.
yum install python-crypto2.6 python-lzo (for centos6)
yum install python2-crypto (for centos7)
will quickly install pycrypto(probably do some 'linking' works) and lzo. The other two depended on pip install.
Probably you need this easy-install.pth file in python's site-packages dir:
import sys; sys.__plen = len(sys.path)
./pycrypto-2.6.1-py2.6-linux-x86_64.egg
./pyudt4-0.6.0-py2.6-linux-x86_64.egg
import sys; new=sys.path[sys.__plen:]; del sys.path[sys.__plen:]; p=getattr(sys,'__egginsert',0); sys.path[p:p]=new; sys.__egginsert = p+len(new)
I provide pre-compiled package pyudt_tun-centos6-x86_64.tar.gz and pyudt_tun_lzo-centos7-x86_64.tar.gz to simplify the installation procedure of pyudt4 & python-pytun.
Be aware of pyudt4 having some bugs, you'd better download it's source code of epoll-fixes branch and apply the patch I offered. See changelog.txt v0.0.20 2016.03.07 fixed section for detail. (Already included in pyudt_tun-centos6-x86_64.tar.gz and pyudt_tun_lzo-centos7-x86_64.tar.gz)
Or, of cause you can let pip do it all for you(not including patching pyudt4):
pip install 3xsd
In a debian, you can use apt-get to install python-pip(pip) or python-setuptools(easy_install), then to install the packages following.
Python Packages(Modules) version reference:
gevent==0.13.8(1.0.1, 1.1)
greenlet==0.4.2
pysendfile==2.0.1
setproctitle==1.0.1
psutil==0.6.1
dpkt==1.6(1.8.6)
python-pytun==2.2.1
pyudt4==0.6.0(epoll-fixes branch)
pycrypto==2.6.1
python-lzo==1.8
System libs version reference:
libevent-1.4.13-4(not actually used, just needed for gevent to function)
udt-4.11-6
lzo-2.03-3.1
To install a module of specific version(like gevent 0.13.8), you can:
pip install gevent==0.13.8
This will install the latest version of gevent(pypy will need it):
pip install git+git://github.com/surfly/gevent.git#egg=gevent