Enterprise Networking Week 10

Spring 2021, Mondays 5:30-8:00

April 5

pcap example

Issues in comparing TCP performance:

• The sample interval should contain a large number of teeth, if loss-based congestion control is involved
• The graphs I ended up with are for 1 GB.
• one-way delay 200 ms, bw 5 MB/s: bdp = 2 MB = 1300 packets. Plus queue of 400: 1700 packets in transit. Suppose Reno gets half; that's a tooth 400x600 = 240,000 packets, or 360,000 1-K blocks.
• Graph artifacts
• TCP performance for shorter connections is also important

HTTP packets

Statistics => conversations; click on TCP tab. Normally A is 192.168.1.10. Look for connections with large numbers in the bytes A->B column:

• port 44646: 4581 up, 7146 down (TLS)
• port 46288: 4221 up, 80k down; simple GET
• port 53702: 6305 up, 70k down; 4 separate image/zipjs GETs
• port 43120: 2566 up, 5489 down
• port 43122: 1013 up, 793 down (second screenful; simple small GET)
• port 43126: 2512/4759 (3 separate GETs)
• port 43128: 2567/4681
• port 43126: 2214/4802
• port 37202: 2136/6545
• port 39849: 1986/5627
• port 38186: 1380/1258
• port 38188: 2252/1320
• port 47318: 13k/246k
• port 35364: 2056/3232
• port 47414: 2692/2443
• port 46734: 5848/3134   Two large GETs, with cookies
• port 54238: 9808/17k   Lots of traffic, but TLS
• port 39686: 8109/6449: TLS
• port 54510: 5932/6655: TLS
• port 36650: 6060/6405: TLS
• port 37790: 6586/6993: TLS
• port 58046: 7108/7514: TLS
• port 55138: 25k/37k (TLS)
• port 47880: 3524/6081 (HTTP) 4 separate GETs

SDN example

./pox.py forwarding.l2_pairs

./pox.py openflow.nicira --convert-packet-in forwarding.l2_nx

SNMP tables

• get_bulk
• sparse tables
• some examples
•