Instead of Frame Relay frames, you have to look at the
payload which is usually IP packets. Here is the formula
that I would use:
BW = IP bandwidth in bits per second (bps)
PS = average IP packet size; typically 256 bytes
AO = AAL5 Overhead = 8 bytes
OH = overhead factor
Ceiling((PS+AO)/48) * 53
Ceiling((256+8)/48) * 53 6 * 53
= ------------------------ = ------ = 1.24 (for our 256B example)
EBW = effective ATM bandwidth in cells per second
BW * OH
Thus for 512 Kbps IP bandwidth, you need 1497 cps ATM bandwidth
assuming 256 byte average packet size. SCR should be at least this.
The PCR could be the line rate or equal to SCR. The MBS could
be 32 cells (1 MTU packet size at least) for low jitter applications
and as high as 256 cells for bursty applications.
Specifying an "average packet size" is rough: I've observed that on
an "average Internet connection" around 35% of packets are 64 bytes,
35% 1500 bytes, and the remainder scattered about. The average is
guaranteed to be a poor match for at least 70% of your traffic.
Not that it matters. Most carriers configure FRATM VCCs as
"1536kbps frame relay = 4500cps", treating fractions accordingly.
The last thing you want to do is exceed this cell rate (around
1710kbps at a 1500-byte packet size), as on a policed link you'll
lose nonconforming cells. (Actually, a 1536kb frame relay seems to
be closer to a 1705kb ATM VCC. Eh, the IWF has to buffer some
anyway.) If you increase your cell rates you potentially
oversubscribe your frame link (and the IWF may disallow it in any
case). So if you're tempted to exceed this ratio be very certain of
your link characteristics.
Add to that: when a sustained rate is defined (VBR service
category) the ATM peak rate probably doesn't mean what you would
expect, so I recommend sustained = peak. (ABR is the best match for
data, but isn't often used.)
As to the original question, I wouldn't recommend translating a
burstable frame to ATM unless you can order an ABR VCC or get UPC
and/or frame discard disabled on your link (don't bet on it). It's
not likely you can match parameters, so without flow control you're
essentially looking at reduced performance or packet loss under load.
Whatever you do get, I recommend you ask your carrier for a spec.
You have plenty of variables and relatively robust protocols, meaning
you could take a shot, miss, and only find out for certain that you
had when you'd really rather not. Get it reasonably right, though,
and it'll do right by you.
Huh. Now that I look at it, I need to rethink my own figures yet
again (I haven't even implemented the last rethink) as I appear to
have tended toward the conservative in some areas. Grrr.
Take it easy.
Peter E. Fry