Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Server Memory - BuildStuff Ukraine 2019
Search
Андрей Листочкин (Andrey Listochkin)
November 18, 2019
Programming
0
26
Server Memory - BuildStuff Ukraine 2019
Андрей Листочкин (Andrey Listochkin)
November 18, 2019
Tweet
Share
More Decks by Андрей Листочкин (Andrey Listochkin)
See All by Андрей Листочкин (Andrey Listochkin)
Everybody Stand Back! I Know Regular Expressions
listochkin
0
120
Command-line scripting with Rust. Wait, what?!
listochkin
0
250
Server Memory - Chernivtsi JS 2019
listochkin
1
99
10 Years Later
listochkin
0
330
Managing Managers - DevTalks iHUB
listochkin
0
34
Time, Numbers, Text
listochkin
1
500
Software Licensing: A Minefield Guide
listochkin
0
120
We Make Bots. For Real
listochkin
0
370
Devops for A Busy Developer - XP Days 2016
listochkin
0
150
Other Decks in Programming
See All in Programming
Folding Cheat Sheet #7
philipschwarz
PRO
0
150
Activities at Cairo Library
cairolibrary720
0
1.2k
初心者がおさえておきたいAWS CDKのベストプラクティス 2024
konokenj
15
7.3k
Async Await: Mastering Python's Time-Bending Tricks - EuroPython2024
yanbo
1
290
Polarsの成長: v0.14からv1.0までの変遷と今後の展望
zerebom
1
350
ピグパーティにおけるMongoDB CommunityバージョンからAtlasへの移行事例
10969hotaka
0
130
実用的かつリーズナブルな 「Azure × Gemini × LINE」~キャラクターBot 実装ライブデモ~
tomodo_ysys
1
170
生成AIをkintoneに連携してみた
hideg
0
230
AWSでゲームサーバーを運用! Amazon GameLiftのお話
iriikeita
0
200
유연한 Composable 설계
l2hyunwoo
0
380
CSC307 Lecture 14
javiergs
PRO
0
220
わかりやすい正解を捨てて、コトに向き合う - スクラムフェス金沢2024 スポンサーセッション
yusukekokubo
0
170
Featured
See All Featured
個人開発の失敗を避けるイケてる考え方 / tips for indie hackers
panda_program
78
15k
[RailsConf 2023] Rails as a piece of cake
palkan
35
4.4k
The Illustrated Children's Guide to Kubernetes
chrisshort
39
47k
How GitHub (no longer) Works
holman
305
140k
Keith and Marios Guide to Fast Websites
keithpitt
408
22k
For a Future-Friendly Web
brad_frost
173
9.2k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
35
6.3k
Optimising Largest Contentful Paint
csswizardry
18
2.6k
KATA
mclloyd
20
13k
Rails Girls Zürich Keynote
gr2m
93
13k
XXLCSS - How to scale CSS and keep your sanity
sugarenia
245
1.2M
Designing for Performance
lara
604
67k
Transcript
None
@listochkin + fb.com/tektonna
Memory Management
A bit deeper than most talks on similar subject
None
CPU Caches RAM Swap Disk Network
Operating System
None
Pages 4k
Virtual
3 processes [0 … ∞] [0 … ∞] [0 …
∞]
Allocate page But not really
None
malloc calloc realloc
None
pointer = malloc(size)
free(pointer)
malloc malloc malloc free malloc free free
Forget to free Free the wrong pointer Read from unallocated
memory
Can we do better?
Reference Counting
None
const players = [ … ] players.forEach(p => makeMove(p))
a = … doX(a) yield a
if ( … ) { let a = … …
}
C++: std::shared_ptr
None
Optimizations
Delayed Counting
Don’t count local references
function (user) { let email = user.email; … … }
Delayed deallocation
Reference Cycles
None
None
Compile time ref-counting
Several types of pointers
Ownership
Forbids cycles by forbidding several owning references at the same
time
None
None
None
Perl 1987 - today Python 1991 - 2001 PHP 1994
- 2009
Multiprocess deploys
fork
Copy on Write
Worker Accepts requests As memory use raises: Stop accepting requests
Complete in-flight requests Terminate
Master Keep track on workers Start new workers Signal them
to terminate when memory pressure is high
“Pre-fork”
Load the framework Load app code Run full GC ----
Start forking process to accept requests
None
None
“Processes” Supervision Trees
Tracing GC
1959 60 years ago
Start at Root references Follow all references Build a live
objects tree Delete all objects not part of the tree
Ruby Java JavaScript Lua Go Default Choice
Roots?
Constants Global variables Local variables Closures Thread-locals ...
Mark & Sweep vs Mark-Compact
None
Team Sweep: Go Ruby* Lua Embedded JS engines Erlang
Pros: Pointers don’t change Native extensions Easier to implement
Team Compact: Java JavaScript Ruby* Haskell
Pros: Less memory fragmentation over time Cons: Harder Takes longer
to do a GC
Incremental Marking
3-colored algorithm by Dijkstra™
None
None
None
None
None
What is a barrier? if statement
CPU branch predictor
Parallel marking
None
Lazy Sweep
Generational
Temporary data data.filter( … ).map( … ) log(`… ${data}`) JSON.stringify(result)
Major vs Minor GC
Pointers from Old objects to new objects
Remembered Set
Major GC Scan roots only Bigger object graph
Minor GC Scan roots + RS Small object graph
None
Modern GCs are hybrid
None
None
C-extensions
Can’t move objects if their references are passed to an
extension
Can’t add WB
RGen GC
2 types of objects WB-protected WB-unprotected
WBu are never OldGen
OldGen -> WBu WBu to Remembered Set
Mark WBus on every minor GC
Adding compaction for WBp
None
V8
Minor GC Parallel
Major Parallel marking Parallel / Concurrent Compact || Sweep
None
When to trigger GC?
Out-of-Bounds GC
Firefox Run GC in background tabs first instead of current
tab
Chrome Animation frame
OS Pages pre-forked processes malloc zones GC Regions Remembered Sets
Barriers
None
None
Practical Advice
Memory Leaks Multiprocess Deploys
Throughput ⇔ Latency
Throughput Batch jobs Queues
Latency Requests Messaging
Keep different types of workloads in different processes
More memory Swap with SSDs
Experiment and measure
@listochkin Hire Me!