And we thought the text part of the Starr Report would be bad

Paul Zawada writes:

[snip ...]
You can only fit 1.536 Mbps of real data down a T1 unless you've figured
out a funky way to send it across using the framing bits.

8 b per channel * 24 channels per frame * 8000 frames per s = 1536000 bps
28.8 kbps * 53 = 1.526 Mbps
28.8 kbps * 54 = 1.555 Mbps

So technically they should have used the numbers 52 and 53. :slight_smile:

  There is more to this than meets the eye -- 28.8K is asynchronous
  and has start and stop bits for every byte, so there are a maximum
  of 2880 bytes/sec available over 28.8K. Then there is the issue
  of latency and IP overhead which tends to approximate 89% of the
  available bandwidth, or 2880 x .89 = 2563 bytes/sec. The T1 circuit
  normalized to bytes/sec is (1536000 / 8) = 192000 bytes/sec, so
  accounting for apples and oranges shows 192000 / 2563 = 74.91 sessions.
  Ofcourse, this assumes that all 75 28.8K modems are pulling data
  in the same direction at the same time. In reality, traffic for
  most ISPs runs about 4:1 to 5:1 inflows vs outflows. Because T1
  circuits are full duplex (1536000 bps in two directions at the
  same time), and assuming a 4:1 ratio of inflows to outflows, this
  would allow 75 + (75 / 4) = 93 28.8K connections. Fortunately,
  customers don't actually saturate the lines the entire time they
  are connected -- some actually pause to read the screen once in
  a while -- so the real usage on the line has a tendency to
  approach 8:1, 10:1, or even more sessions. With a conservative
  number of 8:1 inactive to active sessions, 8 x 93 = 744 sessions.
  The Starr Report and other nonsence have a tendency to blow
  these types of calculations out of the water every now and then,
  but then no provider has built their network on the worst case
  scenareo. YMMV.

  Dave Stoddard

Almost every modem supports V42 error correction, which makes the modems
speak a sort of synchronous with each other (actually data is transmitted
in blocks with a start-of-block and end-of-block marker, and a checksum).
That gets you 8 bits in a bith minus some negligeble V42 overhead.
So an 28k8 modem can actually transfer almost 3.6 Kbytes/sec.

Because of the block-oriented approach you do get a bit higher latency
on interactive connections, which is why gamers often turn of V42.

Somehow, nobody seems to know this.