0% found this document useful (0 votes)
210 views

103-Huawei OceanStor Distributed Storage V1.2

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
210 views

103-Huawei OceanStor Distributed Storage V1.2

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 42

Building a Foundation of Mass and

Diversified Data for an Intelligent World


Huawei OceanStor Distributed Storage Main Slides

Security Level:
Industry Product
OceanStor Insights Capabilities
Distributed
Storage Huawei Best
Strategies Practices

2 Huawei Confidential
Data Integration and Cloud Transformation Drive Distributed
Storage Market

Rapid Growth of Distributed Storage Market Extensive App Scenarios

Cloud transformation
Hyper-converged
Elastic deployment, agile service
development, and quick response
Distributed to traffic surges

Cross-domain integration
PB-level capacity, resource sharing, cost
All-flash
reduction, and efficiency improvement

Converged
Video networking
Linear expansion, One City
One Pool for city security
1 2 3 4 5 6 7
Source: Huawei MI

3 Huawei Confidential
5G+AI Accelerate Data Generation and Flow

Data increments/year 60 TB/day 6 TB


Source: Huawei GIV
Training data Daily output
per vehicle data per device

Autonomous driving Gene sequencing


2025

180 ZB
Data
10x 12 TB/hour
Data growth 8K video data
2018 compared with 4G volume
32 ZB
5G 8K UHD

4 Huawei Confidential
5G and Mass Data Accelerate UHD Video Industry for Ultimate
Experience
8K
7,680 x 4,320
HD 8K
1-hour material 300 MB 12 TB
volume
4K
3,840 x 2,160 1 MB data 80 ms 2 ms
read latency

Full HD 1-hour material 90 seconds 1 hour


1,920 x 1,080 copy duration

2014 2018 2018 2020

2014 FIFA World


Cup in Brazil
First 4K live
CCTV
First 4K UHD
channel
Winter
Olympics in
Pyeongchang
CCTV New
Year's Gala
First 5G + 8K
40x more data 40x shorter latency

broadcast NBC VR live live broadcast


broadcast

5 Huawei Confidential
Big Data Applications: Mass Data Improves Efficiency

Traditional financial

40x Internet transformation


More micro-loan Three Challenges
customers

Challenging deployment:
inefficient physical machine
deployment, and impossible
Practices of public provision of cloud services
80% security big data
Higher case
investigation efficiency
Complex storage: Short-term
storage period due to high costs

Difficult analytics: impossible


Government full analytics due to siloed platform
One-stop big data
business deployment and scattered data
handling

6 Huawei Confidential
AI Applications: Mass Data Enables Autonomous Driving in
Everyday Life
Ultrasonic sensors
Data collection and preprocessing AI training Simulation AI inference

GPS

Data
Lidars Data import preprocessing AI training Simulation Validation

S3/NFS HDFS NFS NFS/CIFS NFS/HDFS

Millimeter wave radar


100 PB-level 100 GB/s-level Mass small files 10 GB/s-level Ultra-low latency:
bandwidth Ultimate OPS and latency bandwidth < 1 ms
Camera

60 TB data/vehicle/day S3, NFS, and HDFS High bandwidth and OPS


From L2 to L3 interfaces requirements
10x increased mileage, 20x more data
100 PB-level data Mass data copy leads Ultra-low latency requires
needs storage. to slow analytics. high performance.

7 Huawei Confidential
Video Surveillance: Mass Data Safeguards Our Health
and Property
A medium-sized city

3 PB  30 PB
HD reconstruction: coverage from urban areas to the
entire city, and retention period from one month to
three months

1 billion  10 billion
Thousands of checkpoints, and tens of billions of
images stored in a year

Quasi-real time  Real time


Tens of billions of vehicle and image data records
searched within seconds

Facial recognition Vehicle recognition Waste identification Gait recognition

8 Huawei Confidential
Industry Product
OceanStor Insights Capabilities
Distributed
Storage Huawei Best
Strategies Practices

9 Huawei Confidential
Storage Challenges for New Data Infrastructure Deployment

Autonomous
8K video AR/VR 5G Telemedicine
driving

 Insufficient storage
Industry apps Government | Finance | Carrier | Smart city | Large enterprise
5G+Cloud+AI drives rapid data growth.
Increasing data needs storage. Storage
scalability and TCO issues need to be resolved. Data enablement Integration | Governance | Development | Service

 Data silos

Full stack
Unstructured data dominates. Service silos Data processing Databases Big data AI
cause horizontal data distribution and vertical
data separation, resulting in low efficiency in
analyzing diversified data. Data storage SAN NAS Object HDFS

 Poor management Diversified


Mass and diversified data leads to complex computing power x86 Arm NPU GPU
data lifecycle management
Data connections

10 Huawei Confidential
Building a Foundation of Mass and Diversified Data for an
Intelligent World

Safe City Digital Intelligent


E-Government Carrier
Banking Factory

Industry

File HDFS Object Block

Services

Full OceanStor

Stack Nodes OceanStor

All-Flash
OceanStor

Performance Capacity

Components Compute Transmission Storage Management

11 Huawei Confidential
Five Scenario-Specific Solutions Enable Digital Transformation of
Thousands of Industries
Decoupled Storage-Compute Big Data
HCS private cloud Solution
Government, finance, and Carriers, public safety, and
carriers government
Virtual storage resource pool Decoupled storage-compute
for elastic expansion architecture, building cost-effective big
data storage

HPC
Large enterprises, education
& research centers, and
supercomputing centers
Storage-compute synergy for
a parallel high-performance
storage solution
Mass resource pool
Finance, carriers, backup and
Video cloud
archiving Safe City, and transportation
100B objects per bucket, EB- Video and images stored in the
level scalability same cluster, and 1,600-channel
video recording per node

OceanStor Distributed Storage


12 Huawei Confidential
A Benchmark of Distributed Storage Industry in Ten Years

3000+ Global Customers by the End of 2019

Industry 1ST 4-in-1

2019

A-A, 3AZ EC
Block, 4.5B [email protected] Ultra Reliability
NO.1 SPC-1 2018
4K UHD
2016
Korea KBS
2nd File 2015
SPEC New Benchmark
2013
1st Gen Object, Block
10x DB Acceleration
2012
1st Gen File
2009

13 Huawei Confidential
Full-Series Dedicated Hardware Provides Optimal User
Experience

• The core binding mechanism


Dedicated improves performance by 20%.
hardware Higher • Multiple algorithms are offloaded to
the Arm chip, achieving 10% higher
OceanStor
performance performance.
Over 15% higher performance than
OceanStor
x86 alternatives when running
• FlashLink enables 0.5 ms ultra-low
same power consumption latency.
OceanStor

• DIF data integrity protection


2U 4U 5U More solid • Comprehensive subhealth check
reliability on network, storage, computing,
and process
Complete hardware fault
• Fine-grained I/O performance
Category F Series P Series C Series detection and isolation
statistics
Ultra-high Performance High-density
Description Capacity model
performance model model
Better • Clear fault demarcation
Height 2U 2U 4U 5U • Component quality assurance
compatibility thanks to strict control and testing
CPU Kunpeng/x86 Kunpeng/x86 Kunpeng/x86 Kunpeng/x86 Integrated deployment
of incoming materials
and delivery

14 Huawei Confidential
New Business Model Ensures Customers Don't Waste a Single Cent

Traditional business model Available capacity business model


Unified hardware and purchase of software based on raw capacity Procurement based on actual business requirements

I need I need
X PB available capacity X PB available capacity

I need X PB available capacity

Vendor Vendor Vendor


A B C

Low Fewer devices and lower Unified


Same language used by
• Customers suffering a loss: Even though different
TCO language business, IT, and procurement
procurement and
vendors provide the same hardware configuration, their
maintenance costs when departments, avoiding errors
software capabilities vary significantly. This means
fulfilling the same data and waste
customers obtain different amounts of capacity from multiple
requirements
vendor products.
15 Huawei Confidential
Industry Insights Product Capabilities
OceanStor
Distributed
Huawei Best
Storage Strategies Practices

16 Huawei Confidential
OceanStor Distributed Storage, the Trusted Choice for Mass Data

for Mass Data Storage


Hardware-, solution-, algorithm-, architecture-level optimization for
30%+ TCO savings

in Using Diversified Data


Multi-protocol interworking with 0 data migration

Data Lifecycle Management


93% of problems are identified in advance and solutions are provided

17 Huawei Confidential
Optimal Cost for Mass Data Storage

18 Huawei Confidential
High-Density Chassis with Ultimate Capacity, Designed for Mass
Data

4,800
512 GB Mate 30 Pro

Ultimate capacity 1,200


4K movies
1.68 PB/5 U 800,000
Human genome data

120 disks/5 U
91%

80 disks/4 U
33%

E vendor I product C series Huawei


Huawei C180 chassis
H model 3 nodes at least 2 chassis at least

C180 storage chassis 20% reduction in 275% improvement


data hall footprint in utilization
19 Huawei Confidential
Elastic EC: The Most Efficient EC Algorithm in the Industry
Greatly Improves Disk Utilization

Vendor C's
3-copy mode Huawei elastic EC

Data
33% 91%

Data
175% more effective disk space utilization
Parity
given the same hardware configuration
Data Data Data

Raw data 3-copy Elastic EC

Maximum redundancy ratio Tolerates the simultaneous


Supports All-flash
supported failure of 4 nodes Huawei Huawei Vendor V Vendor C's
and hybrid storage 3-copy mode
22+2 or 20+4 3-copy EC EC

Huawei EC vs Huawei 3-copy mode: same


reliability, uncompromising performance

20 Huawei Confidential
Dynamic Deduplication and Compression, Enabling High
Space Utilization and Performance
Specifications Huawei vSAN Position
Inline deduplication
Deduplication
Separately controlled Collectively controlled Leading
1. Writes I/Os into the cache. and compression

2. Performs foreground All-flash single node in All-flash single node in


inline deduplication. typical scenarios typical scenarios
Performance Leading
140,000 IOPS, performance 85,000 IOPS, performance
deterioration < 15% deterioration < 20%
SSD/HDD 3. Writes deduplicated data
SSD/HDD
onto disks.

Hybrid (SSD+HDD) and all-


Media type Only all-SSD Leading
SSD

Post-process deduplication Global deduplication and Leading


Scope Group (6 disks) within a node
compression
1. Writes I/Os into the cache.

2. Writes data onto disks.


Workload-based adaptive
3. Re-reads data to the
inline and post-process
cache for deduplication. Dynamic
deduplication and No dynamic adjustment Leading
adjustment
compression, adjustable
4. Writes deduplicated deduplication granularity
data onto disks.
SSD/HDD SSD/HDD SSD/HDD SSD/HDD SSD/HDD

21 Huawei Confidential
100B Objects Stored in a Single Bucket with Efficient Access of
Numerous Small Files

Leading in object count per bucket


Check image Checkpoint IoV
15M 1B 1B 100B image

Vendor X Vendor E Vendor H Huawei

Stable performance • 100B-level objects • In a large city, • Large objects,


TPS
per bucket and 20,000-channel bandwidth > 250
Third-party test data 10B-level objects road traffic cameras MB/s
1M Huawei per set generate 140 • Small objects, TPS
• Performance: TPS billion images of 300 to 600 per
Vendor X 1,000 to 10,000 every year bucket
1M
100B

Meets customers' requirements for single-bucket read and write services, and eliminates the need

22 Huawei Confidential
to reconstruct large-scale applications into buckets.
Decoupled Storage-Compute Big Data Solution: Flexible
Expansion Lowers TCO
Coupled Storage-Compute Decoupled Storage-Compute

Operational Operational
Offline analytics Log retention Offline analytics Log retention
analytics analytics

HDFS components HDFS components

Native HDFS
Protocol

Storage- Storage- Storage- Mgmt. node Compute node Storage node Storage node
Mgmt. node
compute node compute node compute node
OceanStor 100D HDFS
Hadoop compute cluster
Hadoop cluster storage cluster

Decoupled storage- Elastic EC Tens of billions of Broad compatibility


compute deployment Disk utilization improved from files/directory Compatible with Cloudera,
On-demand independent scalability 33% to 91%, and capacity Global namespace does not split and Hortonworks, etc.
saves resources increased by 175% without any directories to simplify operations
cost increase

If required storage capacity is > 512 TB, use the decoupled storage-compute big data solution
powered by OceanStor 100D in the FusionData solution
23 Huawei Confidential
Highest Efficiency in Using Diversified Data

24 Huawei Confidential
Converged Protocols: Industry's First 4-in-1 Storage Service for
Simplified Procurement and O&M
Before After
Check
File sharing image Operational
Operational
Databases File sharing Check images analytics Databases analytics

...

..
.

Block File Object HDFS


... 60% less OPEX
90% shorter TTM OceanStor distributed storage
Block File Object HDFS

 Different storage devices complicate  Unified storage resource pool simplifies


management and O&M management and maintenance
 New devices for new applications lengthen  Dynamic resource allocation on demand for
TTM fast TTM

25 Huawei Confidential
Converged Protocols: Multi-Protocol Data Sharing Eliminates
Data Copies and Reduces Space

Import Analytics Release

Application 0 copy 0 copy


E-invoice image
layer Coexistence of new and old businesses, step-
by-
step transformation from file to object storage
NFS/CIFS HDFS S3
Autonomous driving training

Management system
File HDFS Object Mass data-
based training and analytics without
Data data migration
processing
layer
Enterprise data lake
Distributed storage resource pool One-stop data generation, analysis, and
archiving, one unified copy of all data for th
e optimal use of storage capacity
Hardware
F series P series C series
node layer

26 Huawei Confidential
Cross-Cluster Active-Active Layout Stabilizes Mission-Critical
Workloads

IOPS at 1 ms/node
HyperMetro 168,000
125,000
Active-Active
24/7 service continuity, industry's lowest latency
(databases/VMs)

Traditional high-end OceanStor


storage distributed storage

Kunpeng 920
15%
Computing Higher IOPS with
100 km at 2 ms functions offloaded

AI Fabric 15%
Shenzhen Guangzhou Network Lower latency with
the lossless network

27 Huawei Confidential
System-Level Reliability: Comprehensive Redundancy Tolerates
Simultaneous Failure of Four Cabinets Without Service Interruption

Distributed storage pool


Two data redundancy
Node 1/Cabinet 1 Node 2/Cabinet 2 Node 3/Cabinet 3 Node 4/Cabinet 4 Node 5/Cabinet N
mechanisms: replication and EC
01 02 03 01 02 03 01 02 03 01 02 03 01 02 03
04 05 06 04 05 06 04 05 06 04 05 06 04 05 06
Redundancy levels
01 02 03 01 02 03 01 02 03 01 02 03 … 01 02 03
04 05 06 04 05 06
04 05 06 04 05 06 04 05 06
• Cabinet: Tolerates simultaneous
01 02 03 01 02 03 01 02 03 01 02 03 01 02 03 failure of up to four cabinets
04 05 06 04 05 06 04 05 06 04 05 06 04 05 06

• Node: Tolerates simultaneous


failure of up to four nodes
Traditional centralized
Vendor V OceanStor 100D
storage • Disk: Tolerates simultaneous
failure of up to four disks in a
RAID 0/1/5/6/10 EC and copy modes Distributed RAID technology tolerates cabinet-level
Only tolerates disk-level tolerate the simultaneous failures with the best multi-copy and EC redundancy data chassis
failures. failure of up to two nodes. protection.

28 Huawei Confidential
System-Level Reliability: Failover Within 10s of Node Faults
Minimizes Impact on Applications

1822
Process fault
Huawei Server Smart Network
Interface Card
detection
5S
10s
Fast failover
Fault Fast heartbeat
messages to 3x faster than competitors
groups Performance rivaling that of
Rapid failover 5s traditional high-end storage
Quick service
takeover
OceanStor 100D shortens fault
detection and notification to
Fast I/O milliseconds with Huawei smart NICs,
redirection doubling the overall failover efficiency.
(Available in 2020)

29 Huawei Confidential
Device-Level Reliability: Comprehensive Intelligent Detection and
Preprocessing Addresses Risks Before Faults Occur

Performance
SSD card
Disk • Faults, slow disks, or bad
• Faults, slow disks, or blocks
bad blocks • Over-heating or capacitor
• S.M.A.R.T information failures
p0 • UNC errors • Erase times that exceed
the limit
p2

p1 Network
Server • NIC failures or slow response
pE • Packet loss or rate reduction
• CPU resource exhaustion
• CPU frequency decrease on links
t0 t1 t2 t3 t4 Time • Memory faults • Port failures, intermittent
disconnections, or packet loss
Risk detection and management

System resource performance severely deteriorates, Comprehensive and accurate hardware risk detection
slowing or even disrupting application systems. prevents problems with the Huawei fault library and intelligent
detection algorithms

30 Huawei Confidential
I/O-Level Reliability: DIF for E2E Consistency Check Protects
Data Integrity

2. Check the
1. Insert a parity bit. parity bit. • Online verification
A parity bit is written to the host.
Data Verification occurs when data is written to
Write and read from the host.
Host OS

Memory

Cache

Disk
Data
read
• Back-end verification
The background performs periodic
verification during light service loads.
3. Check the parity bit.

In 2018, a public cloud vendor suffered damaged source data of their file systems • Zero-impact recovery
from silent data corruption in cloud storage. The vendor lost four years of Corrupted data is recoverable with active-
customer and content data. active or local redundant data.

31 Huawei Confidential
QoS Optimization: Intelligent Services for Mission-Critical
Workloads in Mixed Workload Environments

• QoS policies
IOPS
Mixed workload scenarios: Bandwidth- or
Performance burst limit Performance IOPS-based SLA enables precise performance
burst duration
4000 control and on-demand resource provisioning.
The dimensions include bandwidth and

3000 read/write IOPS.

QoS limit
2000
• Burst policies
Policy-based performance burst control
1000
manages intensive interactive services, such
Duration as VM boot storms and online promotion. The
0 1 2 3 4 (hour) dimensions include bandwidth, read/write
IOPS, and duration.

32 Huawei Confidential
Intelligent Algorithm for CPU Partitioning Lowers Latency 20%
Compared to Traditional Storage

CPU grouping algorithm CPU core-based algorithm Intelligent I/O scheduling

I/O read Data exchange Protocol Data


and write channel parsing destaging I/O read 1 I/O read 2 I/O write 1 I/O write 2 Data Top 1
read/write
Advanced Top 2
features
Batch cache Top 3
Core Core Core write
Core Core Core Core Core
Disk
reconstruction
Top 4

Dedicated Dedicated Shared GC Top 5


I/O grouping prevents mutual interference Continuous I/O processing avoids switchover
Dedicated cores for business-critical workloads consumption I/O priority guarantees service
provide sufficient resources and low latency A request is continuously processed on the same core Read/write I/Os are always prioritized.
to prevent same-core thread switchovers for atomic and Other I/Os are restarted after to minimize
Public cores for common workloads balance
lock-free operations. This avoids frequent multi-core read/write latency.
loads
switchovers and improves CPU cache hit ratios.

Stable 0.5 ms latency and fast response to business-critical workloads


33 Huawei Confidential
Simple Data Lifecycle Management

34 Huawei Confidential
Online/Offline Synergy for Intelligent and Streamlined Full-
Lifecycle Management

Training on cloud Training on premises

HUAWEI CLOUD

2+ PB feature data Enhanced training


1000+ scenarios Customized experience

Resource planning Service provisioning System optimization Risk prediction Troubleshooting


60-day forecast of One-click resource provisioning Personalized optimization that Faulty disks identified 14 2000+ fault types
performance and capacity with 1,000+ templates completely meets SLA compliance days in advance Immediate solutions for 93% of
problems discovered

35 Huawei Confidential
Intelligent Risk Prediction Forecasts Storage Performance and
Capacity Trends to Improve Management and Resource Utilization

HUAWEI CLOUD AI-


based deep learning Disk risk detection
• 190,000+ storage systems 14 days Industry-leading techniques eliminate
• 2+ PB feature data potential risks before failures occur
• 1000+ application scenarios

Performance prediction
60 days Analyzes performance patterns and identifies
performance bottlenecks

Romania

Capacity prediction
China
Mexico • Accurately analyzes workloads to improve
365 days resource utilization by 20%
• Reduces TCO with optimal solutions for
capacity expansion

36 Huawei Confidential
Comprehensive Fault Monitoring and Accurate Diagnosis
Provide Immediate Solutions for 93% of Problems Discovered

Comprehensive health scores for high storage Comprehensive fault mode library provides
availability and performance immediate solutions to 93% of problems

Type Example of Monitored Items

Hardware Disks, cables, and fan modules

Integrity, configuration compliance,


Configuration Detection Diagnosis Recovery
and data protection

Pool capacity and reserved space for • 24/7 • Library of 2000+ • Globally shared
Capacity
data protection • Automatic fault fault modes troubleshooting
reporting in experience
Performance CPU usage, latency, and IOPS minutes

Hardware Configuration Capacity Performance

1 min 24 h 24 h 5 min

Data collection intervals

37 Huawei Confidential
Industry Product
OceanStor Insights Capabilities
Distributed
Storage Huawei Best
Strategies Practices

38 Huawei Confidential
China Merchants Bank: OceanStor Decoupled Storage-Compute
Big Data Solution Accelerates Financial Service Innovation
Hadoop coupled storage-compute OceanStor decoupled storage-compute

Detail Precision Financial Detail Precision Financial


records marketing context records marketing context


... ... ... ... Native ...
HDFS
Cluster 1 Cluster 1 Cluster 105 Computing cluster Storage cluster

Compute 11% Compute 65%


Storage 18% Storage 79%

Big data-powered silo


Minutes Decoupled
4x consolidation
Higher resource
Days
Cloud deployment for big data
30% storage-compute
Reduced TCO
utilization increases efficiency

39 Huawei Confidential
China Mobile: Superior Performance and Low Latency Stabilize
Mission-Critical Workloads

10 hr 18 min
5x
faster processing

Hosting VAS OSS MSS BSS 2 hr 9 min

Hosting MMS RBT WAP Signaling Logs NPO OA MSS Cloud BI Report CDR Report CRM Legacy VMAX Distributed
storage block storage

2014 2014 2015 2016 2016 2018/2017 2.5x concurrent users


1 PB 1 PB 2 PB 2 PB 2 PB 6 PB Operational
1000 users
analytics database,
every 100 TB
400 users

OceanStor Distributed Storage 9 ms (traditional VMAX) -> 2


ms (distributed storage)

Legacy VMAX Distributed


storage block storage
Online Video: Three-Site Multi-Active DR for 24/7 Internet Services

Computing
Media Transcoding Content release
resource pool
system system (streaming server)
(VM pool)
Content
upload

Transcoding
Content upload Intermediate file Storage
zone transcoding Buffer zone release resource pool On-demand scalability, exabyte-level
expansion

Content release
databases/Callback

Content placed in
Media imported to

the release zone

Backup and
recovery
45 PB: 90 nodes x TaiShan 4 U x 14 TB SATA disks
S3 S3 S3 S3
Three-site multi-active DR, 24/7
Media and content Backup and archiving service continuity
release zone zone
Bucket A: Libraries that 3AZ EC for service continuity even during a site failure
Bucket C: Resource
permanently store media Bucket B: Streaming
pool space for data
(video sources/original content release zones
protection
images)
69% higher resource utilization,
reduced TCO
Wuxi 3AZ EC 56% storage resource utilization after replacing the 3-copy mode
Yunqiao
Songjiang

OceanStor object storage resource pool


41 Huawei Confidential
Thank you. Bring digital to every person, home and
organization for a fully connected,
intelligent world.

Copyright©2020 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.

You might also like