Explain like I'm five · Database

What is In-Memory Computing (In-Memory DB/IMDG)?

In-memory DB and IMDG, told as a story about an express desk that skips the trip to the deep shelf.

the packer has to walk all the way to the deep shelfthe driver keeps waiting the whole time

The packer has to walk all the way to the deep shelf.The driver keeps waiting the whole time.

the same popular item, yet it's fetched from deep storage every timecouldn't frequently-needed items stay close by?

The same popular item, fetched from deep storage every time.Couldn't frequently-needed items stay close by?

in-memory DB = an express desk that skips the deep shelfdata is kept in memory, so pickup is far faster than from disk

An in-memory DB is an express desk that skips the deep shelf.Data is kept in memory, so pickup is far faster than from disk.

Keep data right in memoryskips the trip to disk
Fast pickupfar lower latency than disk
Volatility riskdata in memory vanishes if power is lost
IMDG — pool many machinespools memory into one big store
how in-memory DB / IMDG gets usedas a cache — cuts deep-shelf lookupse.g. Redisfast lookups like sessions, leaderboardsfar faster than diskIMDG — pools memory across many machinesgrows by shardingvolatile — data vanishes if power is lostbacked up via log/snapshot

In-memory DB/IMDG is used for caching, fast lookups, and pooling memory.Logs or snapshots are kept so data can be restored after a power loss.

frequently-needed data comes out right awaythe rest of the data stays on the deep shelf (disk)memory costs more than disk, so moving all the data into memory is hard

Memory costs more than disk, so moving all the data into memory is hard.So often only frequently-used data gets loaded into memory — that's a cache.

In one breath

In-memory DB/IMDG keeps data right in memory instead of making the trip to the deep shelf (disk), so it can be fetched far faster than from disk.

An in-memory DB keeps data resident in RAM instead of disk, greatly cutting disk I/O latency. An IMDG (In-Memory Data Grid) pools memory across multiple nodes into a large distributed in-memory store. Volatility is its weak point, so write-ahead logging, snapshots, and replication back up its durability.

When grown-ups say it

In-Memory Database
A warehouse that keeps data right in memory. Almost no disk I/O latency on reads.
In-Memory Data Grid
A warehouse pooling memory across many machines. Scales together with sharding. → the warehouse split into zones
Cache
Keeping only the frequently-needed items close by. Holds only a portion of the data in memory.
Volatility
The trait of vanishing when power is lost. Lost data is recovered via log or snapshot.
Durability
The promise that once stamped, it stays. Some in-memory DBs trade off this promise for speed. → the four promises a receipt keeps
NoSQL Modeling
A warehouse reshaping shelves to fit the box. Some products, like Redis, are both key-value and in-memory. → the free-shelf warehouse
Caching (SRE field)
The caching story from the SRE field. The distributed-systems view of caching is covered there. → caching
Concurrency Control
Rules that stop simultaneous edits from colliding. Still needed even when everything lives in memory. → the no-collision rule
Next story
The Control Room That Watches the Whole Warehouse →
Want another term as a picture book? Request →