saved bandwidth -II

Another thought!!

I setup my linux for measuring traffic metering.

Following traffic groups are my concerns.

 {figure1} traffic class

                 [1] [2]
        client ---------+ +-----------> server
               <----+ | | +----------
                [4] | V | V [3]
                   -----------------
                   | cache server |
                   -----------------

  {figure2} tabular format
        * !<n> means that port is not <n>
        * 80 is squid http accel port

  traffic group interface Source Dest

     [1] IN anywhere/!80 anywhere/80
     [2] OUT cache/!80 anywhere/80
     [3] IN anywhere/80 cache/!80
     [4] OUT anywhere/80 anywhere/!80

I setup above rules with ipfwadm accounting.

From above ipfwadm accounting, I found that it shows real
bandwidth savings.

See following sample data:

     (1) clients -> cache farm: 7179 bits/sec, 10 packets/sec
     (2) cache farm -> servers: 3942 bits/sec, 3 packets/sec
     (3) servers -> cache farm: 27167 bits/sec, 5 packets/sec
     (4) cache farm -> clients: 62333 bits/sec, 10 packets/sec

The difference (4)-(3) shows the data from "locally".
The difference is 35166 bps.

  
     {figure3} typical caching topology

                {our network} ==== {cache} ------------- upstream ISP
                               (1) (2)

In figure3, the link (2) is the target for saving bandwidth.
But the traffic rate of link (2) doesn't be effected by cache(s). (<----
by seeing MRTG graph.) In above example, 35kbps (of cause) is small. but
the difference (saving rate) is above 3~4 Mbps when aggregating all cache
farm. I couldn't notice significant difference between applying of
caches from not applying of those. IMHO, the traffic rate of link (2
) immediately drops to the amount of saving rate from caches.

Is there any points of me to consider ?

--J
Received on Wed Jun 24 1998 - 09:06:23 MDT