Cell Relay Archive

[stjahn@cs.uni-magdeburg.de]

Hi,

I have following problem: When I use FORE's ATM API, I get a
throughput of about 30 mbps on an OC-3 connection. When I use UDP 
(IP over ATM), I get about 130 mbps. A large decrepancy!

Well, i watched the traffic with tcpdump (60 seconds) and checked the
time spaces between two successive AAL5 packets (9184 bytes per packet).
There are two classes, time spaces below and above 1 ms (1/1000 second).

  23741 time spaces with   0.552867 ms average (deviation  0.0283305 ms)
    281 time spaces with 167.237    ms average (deviation 64.6856    ms)

Assumed all packets are sent with 0.552867 ms time spaces, we get a 
throughput of 1000/0.552867*9184*8 bits, about 133 mbps! That would be
o.k.. But why does the transmitter 281 times delay sending?

Accidently i watched another strange behaviour. Is the sending host 
envolved in another network connection (Ethernet or ATM), than increases
the ATM throughput up to 120 mbps! STRANGE! For example, I let the 
sending host simultanously send to Ethernet. - In this example the ATM
throughput is about 80 mbps.

  64848 time spaces with  0.552343 ms average (deviation  0.0389876 ms)
    357 time spaces with 67.820    ms average (deviation 71.9429    ms)

Could someone explain this (strange) behaviour? Perhaps an employee of
FORE?


Hosts:    Ultra Sparc/Solaris 2.5.1
Adapter:  SBA200 
Switch:   ASX200
Software: ForeThough 4.0.0 (1.37) 

I use netperf for the benchmarks. I've done multiple benchmarks with
various adapters, all tests with similar results. In all tests was a
minimum cpu idle time of 10 percent.


Thanx in advance,
Steffen.

--
e-mail: stjahn@cs.uni-magdeburg.de

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet