Although these two terms are often used interchangeably, there is a fundamental difference between the two.
Agreed.
If I do an Internet search for "load sharing vs load balancing" my browser's AI results summary has another variation of what the difference is, which its key difference is load sharing is statically done while LB is dynamically done. I.e. for the latter, something is actively monitoring load and shifting traffic to achieve a balanced load. For this definition, the active balancing might be per flow (e.g. Cisco's OER/PfR) or per packet.
Although the above is the key difference in the summary, a quick manual review of return results has another variation that focuses on the results such that load sharing just takes advantage of multiple paths in some manner while LB really attempts to somehow balance load across multiple paths, dynamically or statically.
When I initially attempted a "simple words" answer to @parthrawat979 question, I used a simplistic answer. This answer was further questioned, asking whether it was simply "saying" vs. "doing", to which I responded "not exactly", because it's a little bit more involved.
In @Harold Ritter 's reply I fully agree his example for load sharing is an example of load sharing, but do not agree on his definition. Regarding his LB, as explained, I would say it could be LB or load sharing.
Perhaps the best way to distinguish between the two is the words sharing vs. balancing. Sharing means there's some portion allocation while balancing means the portion allocation is some how proportionly shared based on some rule for the allocation. Where it becomes fuzzy, in networking, whether the balancing is static or dynamic, and how precise the balancing needs to be moment by moment. I.e. depending how you consider precision, you may consider exactly the same network technology load sharing or LB!
In theory, if we wanted ideal balancing we would transmit individual bits across multiple paths. Of course, our data is contained within frames/packets, so splitting a frame/packet, by bits, for parallel transmission would be very unusual. Often we split different "flows" across paths, we might (usually rarely) split a flows' packets across paths, and possibly (very rarely) even split individual packets across paths.
How granular traffic is split often has a major impact in proportional bandwidth usage, especially for short time periods.
BTW I wrote earlier that LB provides equal distribution. Well, that's its goal, but equal's real meaning is better defined as equally proportional over some time period. Load sharing is attempting to take advantage of additional paths somehow over some time period.
Possibly a "simple" case example of load sharing vs. LB. Two links between a pair of routers. ECMP being used. Two flows that qualify for ECMP. Is this load sharing or LB?
Well, if flow to interface is deterministic (default for CEF, I believe), as the two flows may always be directed to same interface, it's load sharing?
If ECMP is using round robin, it's LB?
But if one flow is a backup and the other VoIP, bandwidth usage is very unequal, so is it then load sharing or load balancing? I would say, from a flow perspective it's LB, but from a bandwidth utilization it's load sharing.
This case example may show why there's confusion between load sharing and LB, because much depends on personal perspective and the criteria being used to make the distinction. Also why the two terms are also informally used interchangeably.
Consider MLPPP, whose goal is to effectively treat multiple links like one link of the multiple link aggregate bandwidth. Is it doing load sharing or LB? How does its fragmentation option affect the prior question?
Lastly, although I personally disagree with some of what @Harold Ritter wrote I wouldn't say he is "wrong" because the difference depends much on exactly all the criteria being used to make the distinction. However, basically, I suspect we might all agreed load sharing is a somehow "less" then LB, or laugh, even "just saying" vs. "actually doing".
