Intelligent QoS Grid for Virtualized Workloads · Web 2.0 Search Database Media Streaming OLTP...
10
1 | Copyright © 2016 Tata Consultancy Services Limited Gaurav Gupta Delivery Head, HiTech Industry Solution Unit Tata Consultancy Services Intelligent QoS Grid for Virtualized Workloads 27 May 2016 SDC India 2016
Transcript of Intelligent QoS Grid for Virtualized Workloads · Web 2.0 Search Database Media Streaming OLTP...
1 | Copyright © 2016 Tata Consultancy Services Limited
Gaurav GuptaDelivery Head, HiTech Industry Solution UnitTata Consultancy Services
Intelligent QoS Grid for Virtualized Workloads
27 May 2016 SDC India 2016
22
Characteristics of Data
Analytics
IOT
BI
Cloud
BigData
OnlineGaming
NoSQL
ImageProcessing
HPC
Web 2.0
Search
Database
MediaStreaming
OLTP
Multi Media
ERP
Static Content
ECM
Communication
SocialNetworking
Data
Latency
Access
IOPsSeek
Virtualized I/O
Cap
acity
Sequ
ential
Backup
I/O blender effect
Mixed Pri
mar
y
CompressionSecondary
PerformanceResponse time
Para
llel
Mixed
Spike
File I/O
Deduplication
SecondaryMultiplexed
Block I/O
Throughput
Ran
do
m
Sequential
Storage
Parallel
R/OI/O Pattern R/W
File I/O
Copyright © 2016 Tata Consultancy Services Limited
33
Load
Time
Workload2
Load
Time
Workload3
Load
Time
Workload4
Load
Time
Workload1 – Railway Reservation
Physical Resource
Load
Time
Workload2 – Social Media
Load
Time
Workload3 – Corporate Portal
Load
Time
Workload4 – Online Retail
Load
Time
Workload1 – Railway Reservation
Application Lifecycle & Challenges in Virtual Environment
Vir
tual
Eco
syst
em
Performance InterferenceEmergency
BookingFestival
Time
World Wide Event
Quarterly Results
Product Promotion / Sale
Virtual Resource
Pool
Hypervisor
CPU Memory Disk NIC
Resource Contention…. ???
Solution
Add more nodes in the
cluster ?
Virtual Machine
Movement?
1
2Un-utilized resource
Challenge
QoS Threshold
Virtual Compute
Virtual Network
Virtual Storage
Load
Time
Workload4
Load
Time
Workload2
Load
Time
Workload3
Load
Time
Workload1
Copyright © 2016 Tata Consultancy Services Limited
44
Elastic Controls - Key QoS
Compute
Storage
Node Boundary
Class of Services (CoS) Tag Control Information Prioritization bits (0 -7) 802.1P
Type of Service (ToS) DSCP (Differential Services
Code Point: 0 to 63) IP Precedence (0 – 7)
# of Virtual CPU % Memory DMA I/O request size
MTU size Max Packet size Max bit rate Network wait time NetIOC Transmission delays
Disk queue length Disk I/O wait time Seek delay
Prioritization Controls
Req1
Layer2 Layer3
Req2
Req3
Network
Copyright © 2016 Tata Consultancy Services Limited
%CPU Pages Exchange Context Switch Concurrence
Disk priority % Read-write Rotational delay
55
QoS Grid – An overview
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Virtual Compute
Virtual Network
Virtual Storage
Physical Resources
Distributed architecture Allows grouping of virtual resources & QoSs Scales up to cluster limit
Federated
Takes into account the historical usage Prioritizes resource allocationAnalytic
Senses the change in workload behavior Dynamically leases virtual resources to others Implements bandwidth-as-a-service model
Intelligent
Copyright © 2016 Tata Consultancy Services Limited
Physical Node(s)
66
QoS Grid - Building Blocks
Discover all virtual stacks in the cluster Allows forming domain for sharing virtual resources Extender
Throttles resource bandwidth across nodes based on change in requirementRegulator
Senses the change in workload behavior Dynamically releases virtual bandwidth to others in
the domainSensor
Copyright © 2016 Tata Consultancy Services Limited
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Qo
SQ
oS
Qo
S
Virtual Compute
Virtual Network
Virtual Storage
Physical Resources
Physical Node(s)
QoS Domain
77
QoS Grid Operations
Sensor Regulator
Tracks resource usage of the lessor VMs against new
QoS baselines
Requests recall leased QoS bandwidth
Adjusts QoS of lease and lessor to
original values
Lease Recall
1
2
3
Lease
Extender Sensor Regulator
DiscoversVMs & QoS
Requests Monitor
Creates QoS Domain
Tracks resource usage against QoS
baseline of each VM
Requests Change in QoS tuning
Borrows resources by re-baseline QoSs in the domain
Adjusts the QoS of the lenders & borrower VMs
1
2 3
4
5
6
Copyright © 2016 Tata Consultancy Services Limited
88
Co-existence of workloads through QoS Grid effect (Indicative)
Virtual Compute
Virtual Network
Virtual Storage
Live streaming of press conference on quarterly results
Pic and video files sharing during festivals
Catalog search and order placement
Emergency Booking
Baseline QoS resources
Shrinked compute & stretched storage at
multi-node in the cluster(Lease Recalled by N3)
Stretched compute & storage from multi-node in the
cluster(Leased from N2 & N3 nodes)
Stretched compute on single node in cluster(Leased with in same
node)
Railways reservation
Social Media
Corporate Portal
Online Retail
Copyright © 2016 Tata Consultancy Services Limited
Domain1N1-2VMs, N2-1VM, N3-1VM
Domain2N1-N5: 1VM each
Domain3N1-N5: 1VM each
9
Click to edit Master title styleQ&A
