The Queue Is the Product: Why Another SIM Won't Fix a Clogged SMS Pipeline

Every modem farm has a moment where the answer seems obvious: the queue is backed up, so buy another SIM. Maybe another modem. Maybe a whole new bank. It feels decisive. It also often makes the queue longer, because the constraint was never the number of SIMs in the rack.

Throughput is a queueing problem. Little's Law is the boring little equation that can save you a few thousand dollars and a weekend of chasing carrier tickets. You do not need a graduate seminar to use it. You need a few counters, a spreadsheet, and the willingness to admit that your SMS pipeline has more than one choke point.

Little's Law Without the Textbook Smell

Little's Law says the average number of items in a system equals the arrival rate multiplied by the average time each item spends in the system:

L = λ × W

In modem-farm terms: L is how many messages are pending or in flight, λ is how many messages per minute are entering the system, and W is how long an average message takes from submission to final result. The units matter more than the math. Messages and minutes are fine. Messages and seconds are fine. Mixing them is how people end up with confident, wrong dashboards.

The useful part is not the formula itself. It is the implication: if you know two of the three numbers, you can estimate the third. If 300 messages are sitting in your queue and your completed throughput is 15 messages per minute, the average message is spending about 20 minutes in the system. That number should make you ask a better question than “How many SIMs do we have?”

Where the Queue Actually Lives

A modem farm is not one queue. It is a stack of them, and the slowest one sets the pace.

Start with the application queue. That is where your gateway or homegrown sender holds messages before handing them to a modem. Then there is the modem pool: how many modems are free, how fast each one can submit, and how often they get stuck. Next comes the carrier path. Your account, long code, short code, or 10DLC campaign may have a throughput ceiling that has nothing to do with how many radios you own. Finally, there is the delivery path. A message can be accepted by the carrier and still fail, filter, or sit in limbo downstream.

If you only watch your application queue, you will blame the modem pool. If you only watch the modem pool, you will blame the carrier. If you only watch the carrier, you will blame your code. Little's Law gives you a common language for all of them.

A Worked Example From a Very Average Rack

Suppose you run 20 modems. Each modem can comfortably submit two messages per minute, so the modem pool can theoretically handle 40 messages per minute. Your carrier account, however, is throttled to 30 messages per minute. Incoming traffic is 60 messages per minute. Twenty percent of failed sends are retried, so the effective arrival rate is closer to 72 messages per minute.

The queue is growing by roughly 42 messages per minute. Buying 10 more modems raises the modem pool’s ceiling to 60 messages per minute. That sounds great until you remember the carrier cap is still 30. Now you have 30 modems and a queue that grows even faster because more workers are feeding the same narrow gate. Congratulations: you have built a very expensive way to make the line longer.

Now change one variable. Say the carrier raises the account ceiling to 90 messages per minute and retry amplification drops to 10 percent after you add backoff and idempotency. Effective arrivals are 66 messages per minute. The modem pool can do 40. The carrier can do 90. The modem pool is now the bottleneck. In that case, adding modems is not stupid. Adding modems is the fix, provided the modem pool is actually where the work stops.

That is the whole game. A new SIM helps when the bottleneck is modem slots and the carrier path has headroom. It hurts when the carrier path is capped, retries are multiplying your arrivals, or your application is serializing work behind a single worker.

Retries Are a Hidden Multiplier

Retries are the most common self-inflicted throughput wound. A 20 percent retry rate does not mean 20 percent more work. It means every failed message gets another shot, and those shots compete with fresh traffic. If you retry aggressively, you can turn a modest failure rate into a sustained arrival spike.

Use exponential backoff with jitter. Cap the number of attempts. Make sends idempotent where the carrier or gateway supports it. Most importantly, separate retry traffic from new traffic in your metrics. If your dashboard shows one big “messages sent” number, you cannot see the storm. You just see a queue and start shopping for SIMs.

Carrier Throttling Is a Service-Rate Ceiling

US carriers are not shy about throughput limits. Account-level caps, campaign limits, per-number limits, and filtering all act like a smaller service rate. That is why “we added 40 modems” can coexist with “throughput went down.” The carrier did not care about your rack expansion. It cared about its own gate.

Track accepted, rejected, filtered, and delivered as separate counters. A message accepted by your gateway is not a message delivered to a human. If your queue drains into a carrier black hole, your throughput numbers look fine while your actual delivery rate collapses.

A 15-Minute Sizing Check Before You Buy

Before the next SIM order, do this:

1. Measure arrivals per minute for a normal hour, not a launch day.
2. Measure completions per minute at each stage: application, modem pool, carrier acceptance, delivery.
3. Count retries separately and calculate your effective arrival rate.
4. Compute average time in system with W = L / λ for the stage that worries you.
5. Find the smallest service rate in the chain. That is your bottleneck.
6. Add capacity only at that stage, and re-measure after the change.

The answer might be a new modem. It might be a carrier conversation. It might be a retry policy that stops treating every soft failure like a personal emergency. The point is to stop guessing from the queue length alone.

Little's Law will not tell you which modem to buy or which carrier to call. It will tell you where the line is actually forming. On a modem farm, that is usually worth more than another SIM.