Huawei Converged Infrastructure Vision · 2013. 11. 21. · Huawei Converged Infrastructure Vision...
Transcript of Huawei Converged Infrastructure Vision · 2013. 11. 21. · Huawei Converged Infrastructure Vision...
Huawei
Converged Infrastructure Vision
Santiago Julián, IT Product Manager, Enterprise Business Group
Madrid, Oct 24th, 2013
1
Data Center Development Trends
Converged infrastructure benefits:
1. Adapts to service changes: elastic system expansion reduces service risks
2. Enhances application performance: Computing, storage, and networking resources are optimized.
3. Shortens service online time: The resources pool supply shortens the time required for infrastructure construction from several
months to several hours.
4. Reduces OPEX and IT costs: Unified management for infrastructure and automatic O&M reduces OPEX and IT costs.
Research by Gartner indicates that 30% of data centers will use converged infrastructure by 2015.
Server-centric
• LAN Switch • UNIX, PC Server • Internal storage
• SAN Switch • Independent RAID
array • Separated SAN
network and LAN network
Storage- and Server- centric
• Data Center Ethernet
• FCoE • Unified data
center management
Unified network VM-centric
• A VM is a computing unit • Centralized storage and
dispatching storage on VMs
• Networks are moving towards I/O virtualization
• VM Mobility
Completely automatic architecture
• All computing, storage, and networking resources centrally managed and allocated
• Application transparency
2
Data-Centric Computing Centers and Network Centers
Data centers need to shift the focus of data lifecycles to form a new balance between computing, storage, and networking resources when
application performance bottlenecks occur.
SSD flash storage technology is widely used.
The data center layer 2 network (low latency, lossless Ethernet), virtual network, elastic cloud network, and software defined networking (SDN)
have recently increased in popularity.
Main latency
bottlenecks
Latency: Bandwidth
Latency: Bandwidth
Latency: Bandwidth
Latency: Bandwidth
Latency: Bandwidth
Latency: 1us Bandwidth: 40 Gbit/s
Latency: Bandwidth
Response time: Bandwidth
Port to port Latenc0y: 0.1us Bandwidth: 40 Gbit/s
Port to port Latency: <0.8us Bandwidth: 4-8 Gbit/s
SSD card/disks
Disks
Infiniband card
Ethernet card
FCOE card
Port to port Latency: 2~10us Bandwidth: 1-10 Gbit/s
3
WSC Architecture for Tiered Data Access & Storage
For Data Center with Cloud Computing Scale, the latency for remote RAM/SSD access is far less than local
HDD storage with 2 orders of magnitude difference, while the capacity and bandwidth are at the same level.
-- Abstracted from Google's "Data Center as a Computer".
4
Converged Infrastructure Development
Network
CPU Memory
External
storage
I/O acceleration (proprietary)
Network
CPU Memory
External
storage
I/O acceleration (proprietary)
Virtualization Network
CPU Memory
Storage
I/O acceleration (proprietary)
I/O acceleration (proprietary
or shared)
Unified physical machine
and virtual machine
management High-speed network
CPU Memory Storage
I/O acceleration (shared)
Unified physical and virtual resource pool management
Management platform (monitoring, configuration, automation, and
deployment)
Application deployment template
Traditional infrastructure
Virtualization Converged infrastructure
Converged architecture Converged infrastructure
Fabric-based converged infrastructure (Cloud)
System efficiency and agility increased; OPEX reduced
Serv
er
Serv
er
Hardware
Software
5
Computing and storage separation
1980 - 1990 1990 - 2011
PC/server
Computing and storage
convergence
CPU
MEM
Southbridge
chip
HBA
PCI/PCI-E
SAS/SATA/SCSI
Disk Disk
Disk
Computing node
Computing and storage separation (highly reliable virtualization
integration scenario with large capacity)
CPU
MEM
Southbridge
chip
NIC
PCI/PCI-E
Storage node
RAID Controller
MEM
Southbridge
chip
HBA
PCI/PCI-E
Disk Disk
Disk
+
NIC
Ethernet
SAS/SATA/SCSI
Unbalanced development of computing and storage performances leads to the architectural
separation of computing and storage. The difference between transistor-based computing and
disk-based processing keeps increasing, with data storage attracting more concerns. As a result,
computing and storage are separated to improve resource utilization.
6
2012 + --?
X86 CPU
MEM
HBA
PCI/PCI-E
SAS/SATA/SCSI
Disk Disk
Disk
+
NIC
Computing and storage integration node
X86 CPU
MEM
Southbridge chip
HBA
PCI/PCI-E
SAS/SATA/SCSI
Disk Disk
Disk
NIC
Computing and storage integration node
Cn: Computing blade
Sn: Storage blade Computing blade slots are compatible with storage blade slots.
Mixed insertion of computing blade and storage blade is
supported.
ARM CPU
MEM
Southbridge chip
HBA
PCI/PCI-E
SAS/SATA/SCSI
Disk Disk
Disk
+
NIC
Storage node
X86 CPU
MEM
Southbridge chip
HBA
PCI/PCI-E
SAS/SATA/SCSI
Disk Disk
Disk
NIC
Computing node
Mode 2 OSCA (computing blade+ARM storage blade)
Hardware architecture with integrated computing and storage
Balance server
Balance server
Balance server
...
Balance server
Mode 1 OSCA (computing blade+DAS storage blade) or RH2288
With the use of P2P block storage mechanism in cloud OSs, resource pools and the virtualization technology are applicable to the DAS in the server cluster. The physical separation of computing and storage evolves to the physical convergence and logical separation of architecture to support typical enterprise application consolidation with the minimum cost.
Most IT applications for business transactions of enterprises focus on the instant processing of information instead of the storage or
archiving of data. The convergence of storage and computing can meet this basic requirement of applications.
Southbridge chip
7
IO Performance Enhancement enabled by Scale-Out Computing-Storage Convergence
Page 7
App App
Cache
HDD
FusionCube
App App
Cache
HDD
App App
Cache
HDD
Server2 Server1 Server3
Cache
HDD HDD HDD
SAN
App App App App App App
Server2 Server1 Server3
Server+ SAN
For random IOPS:
Based on Dsware distributed storage engine, total capacity of server cache is more
than 5 times larger than that of centralized SAN controller, thus result in 3-5 times
better hot-data hit-rate and on-line IOPS performance;
SATA can be adopted to substitute SAS to achieve comparable performances with
half the cost
P2P 10GE
PCI-E
GE / FC
SAS
For sequential IOPS (esp. large Files):
Much more number of HDD can serve the same App or VM
instance under the flat P2P architecture, with peak burst MBPS
escalated 3-5 times;
8
Scale-out Distributed Storage
• High Performance Parallel I/O
10X total IO throughput
3-5X IOPS improvement
• High Reliability Replications cross nodes
Quick data rebuild (30min
vs. 12hrs for 1 TB)
• High Scalability Up to 2000 nodes
Linearly scalable in both
capacity and performance
Storage Blades
Server
Hypervisor
Server Server
volume1
Server
volume2
Server
volume3
FusionStorage
VM
Client Client Client
VM VM VM VM VM VM VM VM VM
I/O I/O
FusionSphere FusionSphere FusionSphere Hypervisor
FC/IP network
HDD
SAN/NAS
9
Typical applications of converging computing/storage resources in enterprise appliances
Converged infrastructure Appliance
Converged Fabric (10GE/IB/PCI-E)
Cloud OS DSware
Cloud OS
Converged Fabric (10GE/IB/PCI-E)
DSware
VDI CRM ERM Email ... BOSS
Aggregated multiple small-granularity independent
applications
Virtualization Virtualization
MEM
SSD
DISK
Exclusive single large-granularity distributed
application
HANA/Oracle RAC/
Hadoop Node1
HANA/Oracle RAC/
Hadoop Noden ...
10
iNIC Card
SSD卡
GPU
Computing
Storage
Network
Virtualization
Platform
Unified Central
Management
Distributed
Storage
FusionCube – A True Converged Infrastructure
11
• Start from small
• One chassis - Half
configuration with 4
full-width blades
• Up to 8 chassis concatenation
w/o external switches
• Up to 4096 vCores, 96TB RAM,
and 1.9PB Storage
• Scale out with external switches
• Auto-discover, auto-configure
• Up to 20*3*512 = 30,720 vCores,
240TB RAM and 6PB Storage
Scale On Demand Smoothly
Single
Chassis
One Rack with 3
Chassis
Multiple
Racks
Up to 20
Racks
Copyright©2012 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.
HUAWEI ENTERPRISE ICT SOLUTIONS A BETTER WAY