6.888 Lecture 6: Network Performance Isola8on Mohammad Alizadeh - - PowerPoint PPT Presentation

▶

Sep 07, 2022 401 likes •733 views

6.888 Lecture 6: Network Performance Isola8on Mohammad Alizadeh Spring 2016 1 Mul8-tenant Cloud Data Centers Shared infrastructure between mul8ple tenants/apps Lack of Performance Predictability GAE memcache read 100 values Unpredictable

SLIDE 1

6.888 Lecture 6: Network Performance Isola8on

Mohammad Alizadeh

Spring 2016

SLIDE 2

Mul8-tenant Cloud Data Centers

Shared infrastructure between mul8ple tenants/apps

SLIDE 3

Lack of Performance Predictability

Unpredictable performance, esp. at the tail

GAE memcache read 100 values 3

SLIDE 4

Conges8on Kills Predictability

4 Apr 2013 4 NSDI 2013

SLIDE 5

?

SLIDE 6

Sharing the Network

Alice’s Switch

VM1 VM2 VMn VM3

… Bob’s Switch

VM1 VM2 VMi VM3

… Customer specifies capacity of the virtual NIC. No traffic matrix.

…

Hose Model (Duffield et al., SIGCOMM’99)

2Ghz VCPU 15GB memory 1Gb/s network

SLIDE 7

Sharing the Network

Tenant selects bandwidth guarantees. Models: Hose, VOC, TAG Place VMs, ensuring all guarantees can be met Enforce bandwidth guarantees & Provide work-conserva8on

VM setup Run8me

Oktopus [SIGCOMM’10] Hadrian [NSDI’13] CloudMirror [SIGCOMM’14] Seawall [NSDI’10] FairCloud [SIGCOMM’12] EyeQ [NSDI’13] Elas8cSwitch [SIGCOMM’13] ….

² Adapted from slide by Lucian Popa

SLIDE 8

Example Run8me System: EyeQ (NSDI’13)

SLIDE 9

Shim Shim Shim Shim

Distributed Rate Alloca8on

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s Shim 10Gb/s pipe (min) Rate Guarantees

EyeQ Shim Layer In the trusted Domain (Hypervisor/NIC)

SLIDE 10

Distributed Rate Alloca8on

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s 5Gb/s 5Gb/s 10Gb/s pipe (min) Rate Guarantees

SLIDE 11

RX Module

Distributed Rate Alloca8on

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s 5Gb/s 5Gb/s

SLIDE 12

Distributed Rate Alloca8on

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s 1Gb/s 1Gb/s 8Gb/s

SLIDE 13

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s 1Gb/s 1Gb/s 8Gb/s 5Gb/s

Distributed Rate Alloca8on

SLIDE 14

RX Module

Distributed Rate Alloca8on

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s 1Gb/s 1Gb/s 5Gb/s 5Gb/s Spare capacity

SLIDE 15

Distributed Rate Alloca8on

VM VM VM VM VM VM 2Gb/s 8Gb/s 2Gb/s 2Gb/s 8Gb/s 8Gb/s 2.5Gb/s 2.5Gb/s 5Gb/s 5Gb/s

SLIDE 16

Transmit/Receive Modules

VM VM VM VM VM VM 2Gb/s 2Gb/s 8Gb/s 8Gb/s 1Gb/s 1Gb/s Conges8on detectors

Rate limit. Rate limit. Rate limit.

RCP: Rate feedback (R) every 10kB (no per-source state needed) Per-des8na8on rate limiters:

nly if dest. is congested… bypass otherwise

F e e d b a c k p k t R a t e : 1 G b / s 2Gb/s 8Gb/s

SLIDE 17

Sharing the Network

Tenant selects bandwidth guarantees. Models: Hose, VOC, TAG Place VMs, ensuring all guarantees can be met Enforce bandwidth guarantees & Provide work-conserva8on

VM setup Run8me

² Adapted from slide by Lucian Popa

Cloud Mirror Uses Elas8cSwitch [SIGCOMM’13]

SLIDE 18

Cloud Mirror

² Slides based on presenta8on by JK Lee (HP)

SLIDE 19

Mo8va8on

Cloud applica8ons are diverse & complex Bandwidth models like pipe and hose not a good fit

19 [Bing.com traffic pattern, Sigcomm’12]

web DB cache web logic

SLIDE 20

Hose model is unfit

Hose aggregates BW towards different components

– Too coarse-grained – Prevents accurate and efficient guarantees on infrastructure

intra-component (self-edge) inter-component

SLIDE 21

Hose is too coarse-grained

web logic DB

Web

… …

Logic DB 400 100 300 200 500 800

TCP-like fair allocation would yield 300:200

3-tier web example Hose model

congestion

SLIDE 22

2B web (N) B

… … …

logic (N) DB (N) 2B

Hose over-provisions physical link bandwidth

Hose model reserva8on at L2 : 2B · N

N: # VMs in each tier B: per-VM per-edge bandwidth Physical deployment example

2X overprovision by Hose Model

2 B N

logic - DB demand = B · N

web (N) logic (N) DB (N)

B B B web + logic DB L1 L

… … …

SLIDE 23

Contribu8ons

1. Tenant Applica8on Graph (TAG)
Accurate for complex apps
Flexible to elas8c scaling
Intui8ve
2. VM Placement Algorithm
Guarantee bandwidth and high availability
Efficient for network and compute resources

SLIDE 24

Tenant Applica8on Graph (TAG)

1. Aggregate pipes (like Hose)
Model simplicity
Mul8plexing gain
2. Preserve inter-component

structure (like Pipe)

Accurately capture applica8on demands
Efficiently u8lize network resources

DB mem web logic DB mem web logic DB mem web logic

Component-level graph

SLIDE 25

Tenant Applica8on Graph (TAG)

web (Nw) DB (ND) Bsnd Brcv Bin

web DB

Bsnd Brcv

TAG model Bsnd = per-VM sending bandwidth (VM-to-component aggrega8on) Brcv = per-VM receiving bandwidth (component-to-VM aggrega8on)

What do self-edges mean?

SLIDE 26

Abstract models in TAG

Self-edge ↔ Hose Direc8onal edge ↔ direc8onal Hose, Virtual Trunk

Total guarantee of virtual trunk = min(Bsnd·Nw, Brcv·ND) Brcv web(Nw) Bsnd … … DB(ND) Bin Virtual Switch Virtual Trunk

web (Nw) DB (ND) Bsnd Brcv Bin TAG model

SLIDE 27

Ques8ons

How are TAGs constructed? How to predict bandwidth demands? What is missing for the TAG model?

SLIDE 28

CloudMirror opera8on

VM placement BW reservation Admission control TAG spec Network topology & BW reservation state Available VM slots

host1 10 host2 50 host3 25

Web (N) DB (N)

B B

SLIDE 29

Discussion

SLIDE 30

Next Time: Centralized Arbitra8on

SLIDE 31