Network Sharing Issues Lecture 15 Aditya Akella
Feb 23, 2016
Network Sharing Issues
Lecture 15Aditya Akella
• Is this the biggest problem in cloud resource allocation? Why? Why not?
• How does the problem differ wrt allocating other resources?
• FairCloud: Sharing the Network in Cloud Computing, Sigcomm 2012
• What are the assumptions? Drawbacks?
Motivation
Network?
Context
Networks are more difficult to share than other resources
X
Context
• Several proposals that share network differently, e.g.:– proportional to # source VMs (Seawall [NSDI11])– statically reserve bandwidth (Oktopus [Sigcomm12])– …
• Provide specific types of sharing policies
• Characterize solution space and relate policies to each other?
FairCloud Paper
1. Framework for understanding network sharing in cloud computing– Goals, tradeoffs, properties
2. Solutions for sharing the network – Existing policies in this framework– New policies representing different points in
the design space
Goals
1. Minimum Bandwidth Guarantees– Provides predictable performance– Example: file transfer finishes within time limit
A1 A2
Timemax = Size / Bmin
Bmin
Goals
1. Minimum Bandwidth Guarantees2. High Utilization– Do not leave useful resources unutilized– Requires both work-conservation and proper
incentives
A
B B B
Both tenants active Non work-conserving Work-conserving
Goals
1. Minimum Bandwidth Guarantees2. High Utilization3. Network Proportionality– As with other services, network should be shared
proportional to payment– Currently, tenants pay a flat rate per VM
network share should be proportional to #VMs (assuming identical VMs)
Goals
1. Minimum Bandwidth Guarantees2. High Utilization3. Network Proportionality– Example: A has 2 VMs, B has 3 VMs
A1
A2
BwA
B1
B3
B2
BwB
BwB
BwA =
23
When exact sharing is not possible use max-min
Goals
1. Minimum Bandwidth Guarantees2. High Utilization3. Network Proportionality
Not all goals are achievable simultaneously!
TradeoffsMin
GuaranteeHigh
UtilizationNetwork
Proportionality
Not all goals are achievable simultaneously!
TradeoffsMin
GuaranteeNetwork
Proportionality
Tradeoffs
BwBBwA
A BAccess Link LCapacity C
BwB = 11/13 CBwA = 2/13 C
10 VMs
Network Proportionality
BwA ≈ C/NT 0
#VMs in the network
BwB = 1/2 CBwA = 1/2 C
Minimum Guarantee
Min Guarantee
Network Proportionality
TradeoffsHigh
UtilizationNetwork
Proportionality
Tradeoffs
L
B1
B3
B2
B4
A1
A3
A2
A4
High Utilization
Network Proportionality
Tradeoffs
L
B1
B3
B2
B4
A1
A3
A2
A4
BwB = 1/2 C
BwA = 1/2 CNetwork Proportionality
High Utilization
Network Proportionality
Tradeoffs
L
B1
B3
B2
B4
A1
A3
A2
A4
P
High Utilization
Network Proportionality
Uncongested path
Tradeoffs
L
B1
B3
B2
B4
A1
A3
A2
A4
BwB BwA <
Network Proportionality
L L
Tenants can be disincentivized to use free resources
If A values A1A2 or A3A4 more than A1A3
BwA+BwA = BwBLLP
P
High Utilization
Network Proportionality
Uncongested path
Network Proportionality
Tradeoffs
L
B1
B3
B2
B4
A1
A3
A2
A4
PNetwork proportionality applied only for flows traversing congested links shared by multiple tenants
High Utilization
Uncongested path Congestion
Proportionality
Network Proportionality
Tradeoffs
L
B1
B3
B2
B4
A1
A3
A2
A4
Uncongested path
P
BwB BwA =
Congestion Proportionality
L L
High Utilization
Congestion Proportionality
Network Proportionality
Tradeoffs
Still conflicts with high utilization
High Utilization
Congestion Proportionality
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4L2
C1 = C2 = C
Network Proportionality
High Utilization
L1
Congestion Proportionality
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4
L1
L2
BwB BwA =
Congestion ProportionalityL1 L1
BwB BwA =L2 L2
Network Proportionality
High Utilization
C1 = C2 = C
Congestion Proportionality
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4
L1
L2
Demand drops to ε
Network Proportionality
High Utilization
C1 = C2 = C
Congestion Proportionality
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4
ε
C - ε
ε
C - ε
Tenants incentivized to not fully utilize
resources
Network Proportionality
High Utilization
C1 = C2 = C
Congestion Proportionality
L1
L2
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4
ε
C - 2εUncongested
ε
C - ε
L1
L2
C1 = C2 = C
Network Proportionality
High Utilization
Tenants incentivized to not fully utilize
resources
Congestion Proportionality
L2
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4
ε
C - 2ε
C/2
C/2
Uncongested
C1 = C2 = C
Network Proportionality
High Utilization
Congestion Proportionality
Tenants incentivized to not fully utilize
resources L1
Tradeoffs
Proportionality applied to each link independently
Congestion Proportionality
Link Proportionality
Network Proportionality
High Utilization
L2
B1
B3
A1
A3
B2
B4
A2
A4
L1
L2
Tradeoffs
B1
B3
A1
A3
B2
B4
A2
A4
Full incentives for high utilization
Congestion Proportionality
Link Proportionality
Network Proportionality
High Utilization
L1
Goals and TradeoffsMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Guiding PropertiesMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Break down goals into lower-level necessary properties
PropertiesMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
Work Conservation
• Bottleneck links are fully utilized• Static reservations do not have this property
PropertiesMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Utilization Incentives
• Tenants are not incentivized to lie about demand to leave links underutilized
• Network and congestion proportionality do not have this property
• Allocating links independently provides this property
PropertiesMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Communication-pattern Independence
• Allocation does not depend on communication pattern
• Per flow allocation does not have this property– (per flow = give equal shares to each flow)
Same Bw
PropertiesMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
Symmetry
• Swapping demand directions preserves allocation• Per source allocation lacks this property– (per source = give equal shares to each source)
Same Bw
Same Bw
Goals, Tradeoffs, PropertiesMin
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
Outline
1. Framework for understanding network sharing in cloud computing– Goals, tradeoffs, properties
2. Solutions for sharing the network – Existing policies in this framework– New policies representing different points in
the design space
Per Flow (e.g. today)Min
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
Per Source (e.g., Seawall [NSDI’11]) Min
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
Static Reservation (e.g., Oktopus [Sigcomm’11])Min
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
New Allocation Policies
3 new allocation policies that take different stands on tradeoffs
Proportional Sharing at Link-level (PS-L)Min
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
Proportional Sharing at Link-level (PS-L)
• Per tenant WFQ where weight = # tenant’s VMs on link
A
B
WQA= #VMs A on L
BwB
BwA =
#VMs A on L#VMs B on L
Can easily be extended to use heterogeneous VMs (by using VM weights)
Proportional Sharing at Network-level (PS-N)Min
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
Proportional Sharing at Network-level (PS-N)
• Congestion proportionality in severely restricted context• Per source-destination WFQ, total tenant weight = # VMs
Proportional Sharing at Network-level (PS-N)
WQA1A2= 1/NA1 + 1/NA2
A1 A2
NA2
NA1
Total WQA = #VMs A
• Congestion proportionality in severely restricted context• Per source-destination WFQ, total tenant weight = # VMs
Proportional Sharing at Network-level (PS-N)
WQB
WQA =
#VMs A#VMs B
• Congestion proportionality in severely restricted context• Per source-destination WFQ, total tenant weight = # VMs
Proportional Sharing on Proximate Links (PS-P)Min
Guarantee
Link Proportionality
High Utilization
Congestion Proportionality
Network Proportionality
Work Conservatio
n
UtilizationIncentives
Comm-PatternIndependence
Symmetry
• Assumes a tree-based topology: traditional, fat-tree, VL2(currently working on removing this assumption)
Proportional Sharing on Proximate Links (PS-P)
• Assumes a tree-based topology: traditional, fat-tree, VL2(currently working on removing this assumption)
• Min guarantees– Hose model– Admission control
Proportional Sharing on Proximate Links (PS-P)
A1
BwA1
A2
BwA2
An
BwAn
…
• Assumes a tree-based topology: traditional, fat-tree, VL2(currently working on removing this assumption)
• Min guarantees– Hose model– Admission control
• High Utilization– Per source fair sharing towards tree root
Proportional Sharing on Proximate Links (PS-P)
• Assumes a tree-based topology: traditional, fat-tree, VL2(currently working on removing this assumption)
• Min guarantees– Hose model– Admission control
• High Utilization– Per source fair sharing towards tree root– Per destination fair sharing from tree root
Proportional Sharing on Proximate Links (PS-P)
Deploying PS-L, PS-N and PS-P• Full Switch Support
– All allocations can use hardware queues (per tenant, per VM or per source-destination)
• Partial Switch Support– PS-N and PS-P can be deployed using CSFQ [Sigcomm’98]
• No Switch Support– PS-N can be deployed using only hypervisors– PS-P could be deployed using only hypervisors, we are currently
working on it
Evaluation
• Small Testbed + Click Modular Router– 15 servers, 1Gbps links
• Simulation + Real Traces– 3200 nodes, flow level simulator, Facebook
MapReduce traces
Many to one
BwBBwA
A B
N
One link, testbedPS-P offers guarantees
BwA
N
MapReduce
One link, testbed
BwBBwA
5
R 5
M
M+R = 10 M
BwB
(Mbp
s)
PS-L offers link proportionality
MapReduce
Network, simulation, Facebook trace
MapReduce
Network, simulation, Facebook trace
PS-N is close to network proportionality
MapReduce
Network, simulation, Facebook trace
PS-N and PS-P reduce shuffle time of small jobs
by 10-15X
Summary• Sharing cloud networks is not trivial• First step towards a framework to analyze network
sharing in cloud computing – Key goals (min guarantees, high utilization and proportionality),
tradeoffs and properties• New allocation policies, superset properties from past work
– PS-L: link proportionality + high utilization– PS-N: restricted network proportional– PS-P: min guarantees + high utilization
• What are the assumptions, drawbacks?