Here is my topology below:
Clients Single Web Cache Outer Router Servers
CCC============E=======================O============SSS
100mbps 1ms 10 mbps 100 ms 100mbps 1ms
*All links are duplex.
I have 72 client computers (CCC) connected to a single cache server (E)
for { set i 0} { $i <72 } { incr i }{
$client($i) connect $cache
}
I also connect the cache (C) to N servers (SSS)
for { set i 0} { $i < $N } { incr i }{
$cache connect $server($i)
}
and start each client to connect to all N servers using a nested for loop
$client($i) start-connection $cache $server($j)
Then, I use PagePool/ProxyTrace and a proxy trace file to simulate actual
request from our proxy server. I also defined client num as 72, but there is no
way to define how many servers are needed.
The problem is that in my simulation, the more servers I add, the lesser the
average latency of the client request. It seems there is a bottleneck somewhere
in the simulation that I can't seem to find. The 10 mbps link is not congested
so the bottleneck is not in the network topology. I looked at the source code
for the HTTP/Cient , Http/Server and HTTP/Cache and it does not limit the
number of connections per server per client.
Here is a table regarding the average latency during a 1 hour test period with
varying number of Servers.
============================================================
Servers Ave. Latency Max Latency Size of max latency
============================================================
1 44.273616178936 194.75787848001 63958 bytes
2 19.054615947881 76.160913280016 27259 bytes
3 6.8249151116920 43.515350479991 1958 bytes
5 2.6904748712405 40.750138319997 2026007 bytes
10 1.4624603352714 41.162682479988 2026007 bytes
25 1.1519090774774 40.951985199996 2026007 bytes
50 1.0554587133410 41.318702239997 2026007 bytes
72 1.0247390695315 41.328403759997 2026007 bytes
100 0.9999169117431 41.315464239997 2026007 bytes
500 0.9304826117915 41.332590639997 2026007 bytes
1000 0.9044198998673 39.927294799988 2026007 bytes
Is there a limit to the number of request in a single TCP connection/wrapper
when a client connects to a Server?
How can I simulate N number of servers into a single server node using
Pagepool/ProxyTrace without affecting latency? It is cumbersome and unrealistic
to simulate all Servers in the Internet this way.
Thank you for your time and consideration.
Richard Reyes
Instructor
