Post on 21-Jan-2022
3
AI+5G Cloud Era Requires Diversified Computing
PC era Mobile Internet eraService innovation led
by Cloud, AI, and 5G
x86 PCs
Fixed
network
Midrange
computers/x86 servers
Data center
or cloud
computing
power
ARM intelligent
terminals
2G
3G
4G
x86 servers
Cloud servicesDiverse architecture of IaaS and
PaaS
Industry SaaS
Device
computing
power
NetworkNetwork
virtualization
SDN5G
x86 PCsARM intelligent
terminalsx86 PCs
IoT
devices
AI/big data mining
Network
virtualization
SDN
Embedded AI Low power
consumption
Low latency
Ultra-high
bandwidth
Heterogeneous
computing
Windows 1.x was released.
1985 2007
First-generation iPhone was released. 5G Polar code was
determined as the
channel coding
control scheme.
2019
5G licenses were
issued in China.
2018
The global public cloud
service and infrastructure
expenditure reached
$169.628 USD billion.
2016
4
Cloud Platform Trend
Traditional industry: Build a mature and stable IaaS platform with KVM
Private Public
Hybrid Cloud
Private CloudMulti-DC
ConsolidationHybrid/Public CloudVirtualization
• Computing
virtualization
• Storage virtualization
• Network and security
virtualization
• Automated
management
• Elastic resource
scheduling
• HA and DRS of large
clusters
• Consolidation of
multiple data centers
(DCs)
• Multi-level backup
and DR
• Software-defined
networking (SDN)
• OpenStack/AWS
APIs
• Hardware/Virtual
resource pool
management
• Software-defined
data center
App
Hypervisor
(Host OS)
Hardware
Guest OS
App
Guest OS
• Improves the utilization of CPU, memory, and
I/O resources.• Improves availability and supports load
balancing, dynamic migration, and automatic
fault isolation.• Deploys and manages abstract physical
resources in a simplified manner.
Advantages of KVM Virtualization
5
Cloud Platform Trend
Internet industry: Build a flexible, standardized, and lightweight PaaS platform with
Docker containers
Hardware
infrastructure
Container
technology
Microservice
framework
Agile
infrastructure
Loosely coupled
application
architecture
Monolithic
architectureSOA architecture
Microservice
architecture
Traditional
applications
Web applications with a
few clients
Massive distributed deployment,
clients, and data
Mobile Internet/IoT era
Enterprise Service Bus (ESB)
Mainframe/Midrange
computer and all-in-one
applications
Internet era
• Independent microservices allow quick
rollout of distributed applications in a large
scale• High performance, concurrency, and fast
delivery
Advantages of Docker Container
Technology
6
Kunpeng Supports the Open-source OpenStack +
KVM Cloud Platform
VMVM
OpenStack
Linux
Cinder NeutronNova
Keystone
Glance
Swift
Heat
Ceilometer
Nova-Compute
driver
Cinder-Volume
driver
Neutron
plug-in
VM
Ironic
KVM
QEMU
NIC Driver
QEMU
OVS
Linux
Ceph
Linux
Computing
virtualization
Network
virtualization
Storage
virtualization
Kunpeng server
Management
platform
Virtualization
Hardware
platform
Software Function Support
OpenStackCloud
management
platform
• OpenStack Stein
• Supported components: Nova,
Cinder, Neutron, Swift,
Ceilometer, Heat, Keystone,
Glance, Ironic, Horizon, and
Placement
KVM
Open-source
virtualization
technology
built in Linux
• CentOS 4.14 kernel or later
• Live migration
• Hot swap of vCPUs and memory
is not supported.
QEMUVirtualization
technologies
for I/O devices
• Qemu-2.12.0 or later
• GPU virtualization is not
supported.
7
Kunpeng KVM Virtualization Platform
Performance Tuning
• Check the hardware.
• Check BIOS settings.
• Set memory refresh rate to auto.
• Enable NUMA.
• More
• Update the GCC version.
• Disable transparent huge pages.
• More
KVM virtualization
platform
performance tuning
Hardware
parameter
optimization
OS
Virtualization
settings
optimization
• Bind VMs to CPU cores.
• Prevent cross-die and cross-chip
memory access.
• Enable the memory huge pages.
• More
8
How to Improve the Kunpeng KVM Virtualization
Platform Performance
Tuning Item Description Applicable Scenario Remarks
BIOS configurationSet the memory refresh rate to
auto.Commercial use
It significantly improves memory
bandwidth performance.
NUMA affinity
Ensure that each vCPU and its
memory are in the same
physical node to prevent cross-
die and cross-chip memory
access.
Commercial useIt significantly improves CPU
virtualization performance.
CPU core binding
Bind each vCPU to a CPU
core.Commercial use
It significantly improves CPU
virtualization performance.
Bind vCPUs to CPU cores in a
CPU cluster.PoC tests only
It significantly improves memory
bandwidth performance.
Memory huge
page
Disable the transparent huge
pages and use 512 MB
memory huge pages.
Commercial use
In CentOS 7.6, the CPU
virtualization performance is
improved by about 5%.
9
Kunpeng Supports the Open-source
Kubernetes + Docker Platform
Kunpeng server
Management
platform
Virtualization
Hardware
platform
Software Introduction Support
Kubernetes
Open-source
platform that can
automatically
implement Linux
container operations
Kubernetes
1.15.2 or later
DockerOpen-source
application container
engine
Docker 18.09.8 or
later
10
Kunpeng Docker Platform Performance Tunning
Tuning Item Description Applicable Scenario Remarks
BIOS configuration Set the memory refresh rate to auto. Commercial use
Frequent memory refresh
compromises system performance.
Setting the value to 64 ms can
significantly improve the performance.
NUMA affinity
Ensure that the CPUs and memory
bound to a Docker container are in the
same physical node to prevent cross-
die and cross-chip memory access.
Commercial useIt significantly improves Docker
container computing performance.
CPU core binding
Bind each Docker vCPU to a CPU
core.Commercial use -
Bind Docker vCPUs to CPU cores in a
CPU cluster.PoC tests only
It significantly improves Docker
container computing performance.
NIC interrupt affinity
Bind each interrupt to the CPU of the
NUMA node where the physical NIC is
located.
Commercial use -
11
Supported Mainstream Cloud Platform Software
in ChinaKey Industry Cloud Software Partner Product
Government, finance, and carrier H3C
CloudOS V5
CAS V7
OneStor (Ongoing)
Government, finance
Tencent Cloud
TCE v3.41
TStack
BlueKing Enterprise Edition V1.2
Baidu Cloud ABC Stack (Ongoing)
Jingdong Cloud Jingdong Cloud Agile Private Cloud V1
Kingsoft Cloud Kingsoft Cloud Private Cloud Platform V3.0
UCloud UCloud Stack v1.7
Winhong CNwareKV Service Virtualization 3.2
Government/Public safety
Alibaba CloudDedicated Cloud Agile Edition V3.1
Dedicated Cloud Enterprise Edition V3
SangforSangfor aCloud V5.0
Sangfor sCloud V5.0
Finance
EasyStackECS Stack V5
ECS Private Cloud V5
QingCloudQingCloud V4.3
KubeSphere® V2.1
Rancher Rancher 2.x
GovernmentBingoCloud Infrastructure Cloud Resource Management Software V8.0
Huayun ArcherOS 2020
XC CETC Cloud Nexus (Ongoing)
12
HUAWEI CLOUD Hybrid Cloud Platform Running
on Kunpeng
Cloud
platform
Cloud
services
Kunpeng
Kunpeng servers Storage Network
Hi1822
Management
nodes
Management
VM
KVM
Compute nodes
UVP Euler host OS
NeoKylin
VM
OpenStack
EulerOS
VM
EVS
UVP Euler host
OS
Network nodes
FusionStorageSAN
storage
UVP Euler host
OS
OpenStack components
Management
VM
KVM
...
OpenStack components
KVM
NE
VMNE
VM...
Huawei
UltraPathVBS
Deepin
VMs
CentOS
VM
BMS
nodes
Host OS
User
programs
Computing
service
Storage
service
Network
service
DR &
backup
service
Security
service
PaaS
service
Big data
service
Database
service
Management
service
Cloud services
Infrastructure
Other
Guest OSs
Image
service
Software Function Support
HCS
6.5.1/8.0
HUAWEI
CLOUD
hybrid
cloud
platform
• Hybrid deployment of
x86 and Kunpeng
servers
• Multi-cluster deployment
of containers
• DR HA in the
management zone of a
container; applications
deployed in multiple
AZs in a data zone
• Live migration of VMs
across CPUs in an AZ
• Converged SDN
• Co-deployment of
FusionStorage and SAN
within an AZ
13
Kunpeng Multi-core Computing Architecture
Suitable for Elastic Cloud Platforms
Auto scaling
Multi-core computing
architecture (64 cores+)
VMs
constructed based on
physical cores
VMs
with higher density
Large resource pools with
multiple cores ensure higher
performance of the entire system.
Better isolation ensures higher
reliability.
Fewer servers reduce CAPEX.
14
Hybrid Deployment (1)
Hybrid deployment of Kunpeng and x86 AZs in a region on the OpenStack+KVM cloud
platform
Global
RegionAZ1
Management
nodes
Kunpeng
server
Network
nodes
Kunpeng
servers
AZ2
Compute nodes
Kunpeng servers
Storage nodes
Kunpeng servers
Compute nodes
x86 servers
Storage nodes
x86 servers
New deployment based on Kunpeng
• Hybrid deployment of independent Kunpeng and x86 AZs
in a region
• Management nodes and network nodes use Kunpeng
servers only (hybrid deployment not supported).
Global
RegionAZ1
Management
nodes
x86 servers
Network
nodes
x86 servers
AZ2
Compute nodes
x86 servers
Storage nodes
x86 servers
Compute nodes
Kunpeng servers
Storage nodes
Kunpeng servers
Capacity expansion based on the
x86 cluster
• The original x86 AZ remains unchanged, and AZs
composed of only Kunpeng servers are added.
• Management nodes and network nodes do not support
hybrid deployment. The original x86 architecture remains
unchanged.
Notes: Hybrid deployment requires OpenStack Stein or HCS 6.5.1 & 8.0.
15
Hybrid Deployment (2)
Hybrid deployment of Kunpeng and x86 servers is supported on the Node nodes of the
Docker-K8s cloud platform.
New deployment based on
Kunpeng
• The Master node uses the Kunpeng server.
• The Node nodes support hybrid deployment of
Kunpeng and x86 servers.
Capacity expansion based on the
x86 cluster
Note: Hybrid deployment requires Kubernetes 1.15.2 or later and Docker 18.09.8 or later.
Master
Kunpeng servers
Node
Kunpeng
servers
Node
Kunpeng
servers
Node
x86 servers
Node
Kunpeng
servers
• The Master node originally deployed on the x86 servers
remains unchanged.
• The added Node nodes use Kunpeng servers and hybrid
deployment of nodes in a cluster.
Master
x86 servers
Node
x86 servers
Node
Kunpeng
servers
Node
Kunpeng
servers
Node
x86 servers
16
Summary: Why Kunpeng?
• Ecosystem enablement:
Supports open-source KVM, Docker, HUAWEI CLOUD hybrid cloud, and mainstream commercial
cloud platforms in China.
• Better performance:
Kunpeng multi-core architecture supports more VMs to be created on a device.
• Higher stability:
VMs based on physical cores have small performance deterioration in high-load scenarios.
• Reduced CAPEX:
A smaller number of server clusters are required to build VMs of the same scale.
• Smooth scaling:
Hybrid deployment of Kunpeng and x86 servers implements smooth scaling of clusters on the live
network.
Copyright©2021 Huawei Technologies Co., Ltd. All Rights Reserved.
The information in this document may contain predictive statements including, without
limitation, statements regarding the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that could cause actual
results and developments to differ materially from those expressed or implied in the
predictive statements. Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei may change the
information at any time without notice.
Thank You.