根据之前MQ的性能测试经验,撰写了Pushlet(高性能,分布式消息推送中间件)的测试计划。这里放出来,希望能给大家一些参考,同时也希望大家多提意见~
- Summary
- TestMethodology
2.1 Test Conditions
2.2 Test Scenario’s
2.3 Test Duration
2.4 Environment
2.5 Measurement
2.6 Topology
PerformanceResults
3.1 Connection
3.2 Non-Reliable P2P
3.3 Reliable P2P
3.4 Non-Reliable Pub-Sub
3.5 Reliable Pub-sub
References
- Summary
Pushlet是一个基于socket.io协议的高性能,分布式消息推送服务Suite,为web,无线应用(Android,IOS)提供了一种简单统一的Push机制。
针对Pushlet这种能够承载C500k并发连接,并能实时可靠推送消息的特性,我们重点考察如下几个应用指标:并发连接总数,连接稳定性,TPS(send/get),推送可靠性,推送实时性。同时考察的系统指标有:CPU,Mem,Network,minor/
full GC频度与时间。
- TestMethodology
- 1 TestConditions
- Client-> socket -> connection
- Pushlet连接分两阶段,所有结果均建立在连接完全建立后。
- 每种测试场景运行多次(3~5),达到统计平均效果
- 测试期间,独占系统共享资源(CPU,MEM etc.)
- 可靠消息测试需要分standalone和cluster两种场景
- 所有测试场景均要考虑warmup和cooldown
- 2 TestScenario’s
场景1:并发连接总数&连接稳定性
按照一定的rampup频率(可设定),脚本控制施压机发起连接请求(单网卡连接大致在64511),服务端观测tcp连接状态(重点关注SYN_RECV,ESTABLISHED,TIME_WAIT,CLOSE_WAIT状态),同时观察各个CPU(利用率是否均衡),JVM内存情况(新老生代回收频率,回收时间)。找到系统稳定连接总数(必要时校准kernel,JVM参数)。
场景2:点对点消息推送
选取10个左右的点对点通道,消息体(50byte,150byte,255byte,1k,5k)持续运行30分钟,测算推送TPS,推送成功率,推送时延。
场景3:点对点可靠消息推送
同上,选取10个左右的点对点通道,消息体(50byte,150byte,255byte,1k,5k)持续运行30分钟,测算推送TPS,推送成功率,推送时延。这里需要考虑测试离线消息推送,在线消息接受的场景。
场景4:广播消息推送
选取10个左右的广播通道(订阅方个数分别为5,10,15),消息体(50byte,150byte,255byte,1k,5k)持续运行30分钟,测算推送TPS,推送成功率,推送时延。
场景5:可靠广播消息推送
暂不支持
2.3 TestDuration
所有的测试场景需要持续10分钟以上,并伴有一定时长的warmup和cooldown。
2.4 Environment
- Serversystem
RedHat Enterprise Linux Server release 5.7 (Tikanga)
2.6.18-308.4.2.ali1012.el5xen
4CPU Intel(R)Xeon(R)L5630@2.13GHz
x86_648G RAM 6T disk
- Clientsystem
RedHat Enterprise Linux Server release 5.7 (Tikanga)
2.6.18-308.4.2.ali1012.el5xen
4CPU Intel(R)Xeon(R)L5630@2.13GHz
x86_648G RAM 6T disk
- NetworkSettings
分两种情况:办公网络,公网
网速1GBPS
- Serverkernel(sudo vi /etc/sysctl.conf ... sudo /sbin/sysctl -p,另外也可以通过echo,cat形式验证,如echo 1 >cat /proc/sys/net/ipv4/tcp_tw_reuse cat
/proc/sys/net/ipv4/tcp_tw_reuse)
fs.file-max =1000000
ulimit -n 1000000
net.ipv4.tcp_max_syn_backlog=3240000
net.core.netdev_max_backlog=3240000
net.core.somaxconn= 3240000
net.ipv4.ip_local_port_range= 1024 65535
net.ipv4.tcp_wmem= 4096 65536 524288
net.core.wmem_max= 1048576
net.ipv4.tcp_rmem= 4096 87380 524288
net.core.rmem_max= 1048576
net.ipv4.tcp_max_tw_buckets= 6000
net.ipv4.tcp_tw_reuse= 1
net.ipv4.tcp_syncookies= 1
net.ipv4.tcp_max_orphans= 262144
这里简单说明一下TCP的连接队列和半连接队列。半连接队列,收到syn包但是还没有收到ack包,半连接队列长度由min(min(somaxconn,listen的backlog参数),tcp_max_syn_backlog)决定,再取大于这个值的2的n次幂的值;
连接队列,已完成3次握手,但是由于应用程序accept线程(boss线程)忙,尚未accept。队列长度由min(somaxconn,应用socket的listen方法的backlog参数)决定。更详细说明,请参看:
https://gist.github.com/vongosling/9929680
补充说明:
The current TCP/IP parameters can be edited without the need for reboot in the following locations:
/proc/sys/net/core/
rmem_default = Default Receive Window
rmem_max = Maximum Receive Window
wmem_default = Default Send Window
wmem_max = Maximum Send Window
/proc/sys/net/ipv4/
You’ll find timestamps, window scalling, selective acknowledgements, etc.Keep in mind the values in /proc will be reset upon reboot.You still need to add the code in /etc/rc.local or /etc/boot.local in order to have the changes
applied at boot time as described above.
- Software
JavaHotSpot(TM) 64-Bit Server VM (20.0-b11, mixed mode) 1.6.0_25
JavaHotSpot(TM) 64-Bit Server VM (20.0-b11, mixed mode) 1.7.0_51
Jdk6 jvm options
Xmx6g –Xms6g-Xmn256m
-XX:PermSize=128m-XX:MaxPermSize=256m
-Xss256k
-XX:+DisableExplicitGC
-XX:+UseConcMarkSweepGC
-XX:+UseParNewGC
-XX:+CMSParallelRemarkEnabled
-XX:+UseCMSCompactAtFullCollection
-XX:+UseCMSInitiatingOccupancyOnly
-XX:CMSInitiatingOccupancyFraction=85
-XX:+UseFastAccessorMethods
-XX:+CMSPermGenSweepingEnabled
-XX:+HeapDumpOnOutOfMemoryError
Jdk7 jvm params
Xmx6g –Xms6g-Xmn256m
-XX:PermSize=128m-XX:MaxPermSize=256m
-Xss256k
-XX:+DisableExplicitGC
-XX:+UseG1GC
-XX:MaxGCPauseMillis=500
-XX:InitiatingHeapOccupancyPercent
-XX:G1HeapRegionSize=32
2.5 Measurement
综合使用iostat,jstack,jmap,mat,jps,top,jstat等工具,见后续博文
2.6 Topology
3. PerformanceResults
3.1 Connection
连接总数 |
CPU |
内存 |
GC情况 |
TCP状态 |
C100K |
||||
C300K |
||||
C500K |
||||
… |
3.2 Non-Reliable P2P
消息体 |
TPS(Put/Get) |
送达率 |
平均延时 |
50byte |
|||
150byte |
|||
255byte |
|||
1kb |
|||
5kb |
3.3 Reliable P2P
消息体 |
TPS(Put/Get) |
送达率 |
平均延时 |
50byte |
|||
150byte |
|||
255byte |
|||
1kb |
|||
5kb |
3.4 Non-Reliable Pub-Sub
消息体 |
TPS(Put/Get) |
送达率 |
平均延时 |
50byte |
|||
150byte |
|||
255byte |
|||
1kb |
|||
5kb |
3.5 Reliable Pub-sub
暂无
4.References
http://write.blog.csdn.net/postedit/7200738
http://docs.oracle.com/javase/7/docs/technotes/guides/vm/
http://blog.mgm-tp.com/2013/12/benchmarking-g1-and-other-java-7-garbage-collectors/
http://www.oracle.com/technetwork/java/javase/tech/vmoptions-jsp-140102.html
http://blog.mgm-tp.com/2014/04/controlling-gc-pauses-with-g1-collector/
CMSand G1 Collector in Java 7 Hotspot: Overview,Comparisons andPerformance Metrics