Menene ke iyakance concurrency a scale? Bayyana contention, false sharing, da backpressure.

Question

IT Interview Practice · Accepted Answer

**A scale, throughput yana iyakance ba ta adadin CPU ba sai dai ta contention, cache effects, da queues marasa iyaka — ƙara threads bayan wani lokaci yana sa abubuwa su yi jinkiri.** Amdahl's da Universal Scalability Law duka suna hasashen wannan: farashin daidaitawa a ƙarshe yana rinjaye.

## Contention

Lokacin da threads da yawa ke faɗa a kan lock ɗaya ko cache line, suna serialize kuma suna ciyar da lokaci a jira maimakon aiki. A ƙarƙashin Universal Scalability Law, contention tare da **coherency cost** na iya sa throughput ya *ragu* bayan adadin thread mai kyau.

**Rage matsala**: shard/partition bayanai, per-thread state (thread-local, striped counters kamar `LongAdder`), read-write locks, ko lock-free structures.

## False sharing

Variables biyu masu zaman kansu waɗanda suka zauna a kan **CPU cache line ɗaya** suna sa cores su invalidate cache na juna a kowane rubutu, ko da yake ba sa taɓa bayanai ɗaya — jinkiri mai zafi, na shiru.

```text
cache line (64B): [ counterA | counterB ]  ← two cores ping-pong the line
Fix: pad so each hot variable owns its own line
```

```java
@Contended long counterA;   // JVM pads to a separate cache line
```

## Backpressure

Lokacin da producers suka wuce consumers, queue mara iyaka yana girma har OOM kuma latency yana fashewa. **Backpressure** yana nuna wa producers su ragu: bounded queues da ke toshewa ko ƙi, credit-based flow control, ko reactive-streams `request(n)`.

```text
producer ──▶ [bounded queue full] ──▶ block/reject ──▶ producer slows
```

## Me ya sa yake da muhimmanci

Waɗannan su ne tasirin da ke raba "yana aiki a cores 4" da "yana scale zuwa 64." Ƙira maras hankali suna bugun bango: locks suna serialize, cache lines suna thrash, queues suna kumbura. Sharding don rage contention, padding don kashe false sharing, da backpressure don ƙuntata queues su ne levers da ke sa latency da throughput su tsaya lafiya a ƙarƙashin ainihin load.

Contention

False sharing

Backpressure

Me ya sa yake da muhimmanci

Tambayoyi masu alaƙa