Link Fragmentation and Interleave (LFI)
A typical network has a range of packet sizes. Small packets can be delayed
waiting for a large packet to be sent out the interface. This can happen even
if LLQ is configured—a small voice packet might be sent immediately to the
hardware queue. However, the hardware queue is FIFO. If a large packet
arrived there just before the voice packet, it is serialized out the interface
first. The voice packet has to wait. This causes delay and jitter.
LFI breaks large packets into smaller segments and intersperses the smaller
packets between the pieces of the big ones. Thus, delay and jitter are reduced
for the small packets.
The target serialization delay for voice is 10–15 ms. At 2 Mbps link speed, a
1500 byte packet can be serialized in 10 ms. Thus, there is typically no need
for LFI on links over E1 speed.