Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 1
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
Faster Applications &
Lower Infrastructure Costs with
DataCore™ Parallel I/O Technology
Augie Gonzalez, Director, Product Marketing
@AugieGonzalez
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 3
DataCore Shatters I/O World Records
1. Best Price / Performance: $0.08 per SPC-1 IOPS™
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 4
DataCore Shatters I/O World Records
1. Best Price / Performance: $0.08 per SPC-1 IOPS™
2. Fastest Response: 0.32 milliseconds at full load
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 5
DataCore Shatters I/O World Records
1. Best Price / Performance: $0.08 per SPC-1 IOPS™
2. Fastest Response: 0.32 milliseconds at full load
3. Highest IOPS per Rack Unit: 459K SPC-1 IOPS in 2U
3X or better than previously
achieved by any company!
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 6
“Tell me that isn't amazing.”
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 7
The Difference Is Stunning
Hyper-Converged
VNX 8000
Traditional Storage Array
435K < 459K
SPC-1 IOPS™
Faster
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 8
The Difference Is Stunning
Hyper-Converged
VNX 8000
Traditional Storage Array
$0.41 / SPC-1 IOPS
$0.08 / SPC-1 IOPS
435K < 459K
SPC-1 IOPS™
Faster
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 9
SPC is Basis of Comparison
Criteria SPC Benchmark
Industry Standard
Independently Verified & Audited
Peer Reviewed
Covers different types & generations of technology
Maps to “real world” performance (OLTP databases)
Shows cost for achieving performance level
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 10
SPC is Basis of Comparison
Criteria SPC Benchmark
Industry Standard
Independently Verified & Audited
Peer Reviewed
Covers different types & generations of technology
Maps to “real world” performance (OLTP databases)
Shows cost for achieving performance level
Major Storage Vendors with SPC Benchmarks
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 11
100K
200K
300K
400K
$50K $100K $150K $200K
SPC-1
IOPS
PRICE
Dell
Compellent
SC4020
NetApp
EF560
AFA
EMC
VNX 8000
500K
HP
3PAR
7400
Major Storage Brands
Price Performance Leaders
Usually, Price & Performance go hand in hand
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 12
100K
200K
300K
400K
$50K $100K $150K $200K
SPC-1
IOPS
PRICE
Dell
Compellent
SC4020
NetApp
EF560
AFA
EMC
VNX 8000
459K
DataCore
500K
HP
3PAR
7400
Comparison with
Major Storage Brands
Price Performance Leaders
DataCore Price Performance is Off the Curve
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
$0.58
$0.54
$0.41
$0.37
$0.32
$0.24
$0.08
$-
$0.25
$0.50
$0.75
HP 3PAR
7400
258K IOPS
NetApp
EF560
245K IOPS
EMC
VNX8000
435K IOPS
Dell
SC4020
112K IOPS
X-IO ISE
820 G3
253K IOPS
Infortrend
DS3024B
218K IOPS
DataCore
459K IOPS
Cost per SPC-1 IOPS
13
DataCore has significantly
better Price / Performance
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
4.8
3.3
2.1
1.0 0.9 0.9
0.3
0
1
2
3
4
5
Dell
SC4020
$42K
Infortrend
DS3024B
$52K
X-IO ISE
820 G3
$82K
EMC
VNX8000
$177K
NetApp
EF560
$133K
HP 3PAR
7400
$149K
DataCore
$38K
Latency (milliseconds)
14
DataCore has lowest Latency
among Price / Performance Leaders
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
7.4
3.0
1.2 1.1 0.9
0.3
0
2
4
6
8
IBM SVC
v6.2
w/ Storwize
$3.6M
0.52M
IOPS
Kaminario
K2
$1.0M
1.24M
IOPS
NetApp
FAS8080
EX
$1.9M
0.69M
IOPS
Hitachi VS
Platform
G1000
$2.0M
2.01M
IOPS
Huawei
OceanStor
18800 V3
$2.3M
3.01M
IOPS
DataCore
$0.04M
0.46M
IOPS
Latency (milliseconds)
15
DataCore has the Lowest Latency
“The DataCore/Lenovo system had an
average response of 0.32 milliseconds
at 100 percent load.”
“The chart-topping Hitachi VSP G1000
had a 1.15 millisecond average
response at 100 percent load,
72 percent slower and costing
98 percent more.”
“This is just unreal.”
Chris Mellor
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 16
SPC is a Torture Test for Hyper-Converged
Hyper-Converged must be
Powerful enough to do Both!
Stresses Enterprise Database (OLTP)
Applications + their I/O loads
Hyper-Converged
VMVMVMVM VMVMVMVMVMVMVMVM
Separate Servers & Storage
Separate
servers create
I/O load for
storage system
Storage system
only has to focus
on handling I/O
load
Same servers
creating I/O
load need to
handle the I/O
load!
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 17
Price Performance is Even Better
when All Costs are Included
Hyper-Converged
Separate Server
& Storage Components
Servers not
included in
price
VNX 8000
435K SPC-1 IOPS
$177K
(partial price)
$38K
(all inclusive price)
Node includes
compute &
storage
459K SPC-1 IOPS
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 18
435
177
459
38
0
100
200
300
400
500
SPC-1 IOPS
(K)
Price
($K)
6% more SPC-1 IOPS
79% less cost
EMC VNX8000 DataCore
0.41
0.99
0.08
0.32
0
0.3
0.6
0.9
1.2
$ per
SPC-1 IOPS
Latency
(milliseconds)
80% less $ per SPC-1 IOPS
309% faster
EMC VNX 8000 DataCore
DataCore beats EMC VNX
in All Categories
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 19
Additional cost savings from
Space, Power, Cooling…
Hyper-Converged
VNX 8000
Traditional Storage Array
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
 DataCore hyper-converged can be used for mission-
& business-critical workloads
► Suitable for Oracle, MS SQL Server & SAP HANA
 Removing I/O Bottleneck means higher VM density
► More VMs on same hardware for VDI & general workloads
20
Why it Matters for Hyper-converged
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 21
HOW DID DATACORE DO IT ?
WELCOME TO THE WORLD OF PARALLEL I/O
DataCore™ Parallel I/O technology enables
breakthrough performance and revolutionary
productivity gains by harnessing the full power of
today’s readily available multicore servers.
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 22
IO-Starved Virtualized Servers
Increasingly faster
Uni-processors Serial IO
Work
Potential
2010 20202000
CPU clock rates
slow down
More cores
per socket
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
IO Gap
23
IO-Starved Virtualized Servers
Increasingly faster
Uni-processors Serial IO
Work
Potential
2010 20202000
CPU clock rates
slow down
More cores
per socket
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 24
Serial vs. Parallel Processing
Pile of work
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 25
Serial vs. Parallel Processing
Pile of work
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 26
Serial vs. Parallel Processing
Pile of work
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 27
Serial vs. Parallel Processing
Pile of work
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 28
Serial vs. Parallel Processing
Pile of work
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 29
Serial vs. Parallel Processing
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 30
Serial vs. Parallel Processing
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 31
Serial vs. Parallel Processing
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 32
Modern Multi-core CPUs
Worker
1
Worker
2
Worker
3
Worker
4
Worker
5
Worker
6
Worker
7
Worker
8
Worker
9
Worker
10
Multiple “workers” capable of simultaneously handling
compute, networking and I/O loads
10-cores
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 33
Standard use of Multi-core CPUs
in Virtual Servers
VM
1
VM
2
VM
3
VM
4
VM
5
idle I/Oidle idle idle
Parallel
Compute
Serial I/O
VM = Virtual Machine
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 34
Serial I/O Bottleneck in Virtualized Server
 Compute waits on I/O
Compute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 35
Serial I/O Bottleneck in Virtualized Server
 Compute waits on I/O
 CPU cores are wastedCompute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 36
Serial I/O Bottleneck in Virtualized Server
 Compute waits on I/O
 CPU cores are wasted
 Very little work gets done
Compute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 37
Serial I/O Bottleneck in Virtualized Server
 Compute waits on I/O
 CPU cores are wasted
 Very little work gets done
Compute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 38
Serial I/O Bottleneck in Virtualized Server
 Compute waits on I/O
 CPU cores are wasted
 Very little work gets done
Compute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 39
Impact:
Workload
Server
1
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 40
Impact: Many servers needed to spread I/O
Workload
Server
1
Server
2
Server
3
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 41
Impact: Many servers needed to spread I/O
Workload
Server
1
Server
2
Server
3
Server
4
Server
5
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 42
Turbo-Charge through Parallel I/O
 I/O keeps pace with
compute demandsCompute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 43
Turbo-Charge through Parallel I/O
 I/O keeps pace with
compute demands
 CPU cores are fully used
Compute
I/O
Workload
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 44
Turbo-Charge through Parallel I/O
 I/O keeps pace with
compute demands
 CPU cores are fully used
 Lots of work gets done in
very little time
Compute
I/O
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
45
Adaptive Parallel I/O
Workload
45
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
46
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
46
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
47
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
47
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
48
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
48
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
49
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
400,000 IOPS
< 1 millisec
49
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
50
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
400,000 IOPS
< 1 millisec
50
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
51
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
400,000 IOPS
< 1 millisec
51
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
52
Adaptive Parallel I/O
Workload
Response
Time
(millisec)
IOPS
400,000 IOPS
< 1 millisec
52
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
53
Adaptive Parallel I/O
Workload
No more load
400,000 IOPS
< 1 millisec
53
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
54
Adaptive Parallel I/O
Workload
No more load
400,000 IOPS
< 1 millisec
54
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
55
Adaptive Parallel I/O
Workload
No more load
400,000 IOPS
< 1 millisec
55
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
56
Adaptive Parallel I/O
Workload
No more load
400,000 IOPS
< 1 millisec
56
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
57
Adaptive Parallel I/O
Workload
No more load
400,000 IOPS
< 1 millisec
57
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
58
Adaptive Parallel I/O
Workload
No more load
400,000 IOPS
< 1 millisec
58
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
Worker
1
Worker
2
Worker
3
Worker
4
Worker
5
Worker
6
Worker
7
Worker
8
Worker
9
Worker
10
DataCore’s Adaptive use of Multi-core
CPUs in Virtual Servers
VM
1
VM
2
VM
3
VM
4
VM
5
I/O
Parallel
Compute
Parallel
I/O
VM = Virtual Machine
I/OI/O I/O I/O
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 60
Parallel I/O Breakthrough
TIME SAVINGS COST SAVINGS
Work completes in
1/5 the time
2 machines can do
the work of 10
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 61
• 4x 2U Servers
• 3 TB Cache
• 25.6 TB of PCIe Flash
• 168 TB total mirrored capacity
(336TB total raw)
• N-Node Mesh Mirror
• 2 Full Racks
• 256 GB cache
• 17.6TB of array based SSD
• 140TB total mirrored
capacity (280TB total raw)
• 2-Node Mirror
NetApp Proposed Storage
Real World: Parallel I/O’s Impact on Infrastructure
More than Just Hyper-Converged
DataCore Server SAN
Copyright © 2016 DataCore Software Corp. – All Rights Reserved. 62
• 4x 2U Servers
• 3 TB Cache
• 25.6 TB of PCIe Flash
• 168 TB total mirrored capacity
(336TB total raw)
• N-Node Mesh Mirror
• More Capacity
• Faster Performance
• < 1/4 of Space
• 2 Full Racks
• 256 GB cache
• 17.6TB of array based SSD
• 140TB total mirrored
capacity (280TB total raw)
• 2-Node Mirror
NetApp Proposed Storage
Real World: Parallel I/O’s Impact on Infrastructure
More than Just Hyper-Converged
DataCore Server SAN
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
 Parallel I/O changes the storage equation
 Independent benchmark proves DataCore offers
► More IOPS
► Lower Price
► Better Value
► Lower Latency
 Run Mission-Critical applications on DataCore
63
Summary
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
World-Record Price Performance
64
http://www.storageperformance.org/results/benchmark_results_spc1_top-ten
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
 Try Parallel I/O in your environment
► To speed up response for latency-sensitive applications
► To increase the number of concurrent workloads (VMs)
per server
 Download Free Trial of
DataCore™ Hyper-converged Virtual SAN
http://www.datacore.com/resources/free-hyper-converged-virtual-san
 More Info on SPC @
http://www.storageperformance.org/results/benchmark_results_spc1_top-ten
65
Next Steps
Copyright © 2016 DataCore Software Corp. – All Rights Reserved.
Thank You!
www.datacore.com
66

Can $0.08 Change your View of Storage?

  • 1.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 1
  • 2.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. Copyright © 2016 DataCore Software Corp. – All Rights Reserved. Faster Applications & Lower Infrastructure Costs with DataCore™ Parallel I/O Technology Augie Gonzalez, Director, Product Marketing @AugieGonzalez
  • 3.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 3 DataCore Shatters I/O World Records 1. Best Price / Performance: $0.08 per SPC-1 IOPS™
  • 4.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 4 DataCore Shatters I/O World Records 1. Best Price / Performance: $0.08 per SPC-1 IOPS™ 2. Fastest Response: 0.32 milliseconds at full load
  • 5.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 5 DataCore Shatters I/O World Records 1. Best Price / Performance: $0.08 per SPC-1 IOPS™ 2. Fastest Response: 0.32 milliseconds at full load 3. Highest IOPS per Rack Unit: 459K SPC-1 IOPS in 2U 3X or better than previously achieved by any company!
  • 6.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 6 “Tell me that isn't amazing.”
  • 7.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 7 The Difference Is Stunning Hyper-Converged VNX 8000 Traditional Storage Array 435K < 459K SPC-1 IOPS™ Faster
  • 8.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 8 The Difference Is Stunning Hyper-Converged VNX 8000 Traditional Storage Array $0.41 / SPC-1 IOPS $0.08 / SPC-1 IOPS 435K < 459K SPC-1 IOPS™ Faster
  • 9.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 9 SPC is Basis of Comparison Criteria SPC Benchmark Industry Standard Independently Verified & Audited Peer Reviewed Covers different types & generations of technology Maps to “real world” performance (OLTP databases) Shows cost for achieving performance level
  • 10.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 10 SPC is Basis of Comparison Criteria SPC Benchmark Industry Standard Independently Verified & Audited Peer Reviewed Covers different types & generations of technology Maps to “real world” performance (OLTP databases) Shows cost for achieving performance level Major Storage Vendors with SPC Benchmarks
  • 11.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 11 100K 200K 300K 400K $50K $100K $150K $200K SPC-1 IOPS PRICE Dell Compellent SC4020 NetApp EF560 AFA EMC VNX 8000 500K HP 3PAR 7400 Major Storage Brands Price Performance Leaders Usually, Price & Performance go hand in hand
  • 12.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 12 100K 200K 300K 400K $50K $100K $150K $200K SPC-1 IOPS PRICE Dell Compellent SC4020 NetApp EF560 AFA EMC VNX 8000 459K DataCore 500K HP 3PAR 7400 Comparison with Major Storage Brands Price Performance Leaders DataCore Price Performance is Off the Curve
  • 13.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. $0.58 $0.54 $0.41 $0.37 $0.32 $0.24 $0.08 $- $0.25 $0.50 $0.75 HP 3PAR 7400 258K IOPS NetApp EF560 245K IOPS EMC VNX8000 435K IOPS Dell SC4020 112K IOPS X-IO ISE 820 G3 253K IOPS Infortrend DS3024B 218K IOPS DataCore 459K IOPS Cost per SPC-1 IOPS 13 DataCore has significantly better Price / Performance
  • 14.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 4.8 3.3 2.1 1.0 0.9 0.9 0.3 0 1 2 3 4 5 Dell SC4020 $42K Infortrend DS3024B $52K X-IO ISE 820 G3 $82K EMC VNX8000 $177K NetApp EF560 $133K HP 3PAR 7400 $149K DataCore $38K Latency (milliseconds) 14 DataCore has lowest Latency among Price / Performance Leaders
  • 15.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 7.4 3.0 1.2 1.1 0.9 0.3 0 2 4 6 8 IBM SVC v6.2 w/ Storwize $3.6M 0.52M IOPS Kaminario K2 $1.0M 1.24M IOPS NetApp FAS8080 EX $1.9M 0.69M IOPS Hitachi VS Platform G1000 $2.0M 2.01M IOPS Huawei OceanStor 18800 V3 $2.3M 3.01M IOPS DataCore $0.04M 0.46M IOPS Latency (milliseconds) 15 DataCore has the Lowest Latency “The DataCore/Lenovo system had an average response of 0.32 milliseconds at 100 percent load.” “The chart-topping Hitachi VSP G1000 had a 1.15 millisecond average response at 100 percent load, 72 percent slower and costing 98 percent more.” “This is just unreal.” Chris Mellor
  • 16.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 16 SPC is a Torture Test for Hyper-Converged Hyper-Converged must be Powerful enough to do Both! Stresses Enterprise Database (OLTP) Applications + their I/O loads Hyper-Converged VMVMVMVM VMVMVMVMVMVMVMVM Separate Servers & Storage Separate servers create I/O load for storage system Storage system only has to focus on handling I/O load Same servers creating I/O load need to handle the I/O load!
  • 17.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 17 Price Performance is Even Better when All Costs are Included Hyper-Converged Separate Server & Storage Components Servers not included in price VNX 8000 435K SPC-1 IOPS $177K (partial price) $38K (all inclusive price) Node includes compute & storage 459K SPC-1 IOPS
  • 18.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 18 435 177 459 38 0 100 200 300 400 500 SPC-1 IOPS (K) Price ($K) 6% more SPC-1 IOPS 79% less cost EMC VNX8000 DataCore 0.41 0.99 0.08 0.32 0 0.3 0.6 0.9 1.2 $ per SPC-1 IOPS Latency (milliseconds) 80% less $ per SPC-1 IOPS 309% faster EMC VNX 8000 DataCore DataCore beats EMC VNX in All Categories
  • 19.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 19 Additional cost savings from Space, Power, Cooling… Hyper-Converged VNX 8000 Traditional Storage Array
  • 20.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved.  DataCore hyper-converged can be used for mission- & business-critical workloads ► Suitable for Oracle, MS SQL Server & SAP HANA  Removing I/O Bottleneck means higher VM density ► More VMs on same hardware for VDI & general workloads 20 Why it Matters for Hyper-converged
  • 21.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 21 HOW DID DATACORE DO IT ? WELCOME TO THE WORLD OF PARALLEL I/O DataCore™ Parallel I/O technology enables breakthrough performance and revolutionary productivity gains by harnessing the full power of today’s readily available multicore servers.
  • 22.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 22 IO-Starved Virtualized Servers Increasingly faster Uni-processors Serial IO Work Potential 2010 20202000 CPU clock rates slow down More cores per socket
  • 23.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. IO Gap 23 IO-Starved Virtualized Servers Increasingly faster Uni-processors Serial IO Work Potential 2010 20202000 CPU clock rates slow down More cores per socket
  • 24.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 24 Serial vs. Parallel Processing Pile of work
  • 25.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 25 Serial vs. Parallel Processing Pile of work
  • 26.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 26 Serial vs. Parallel Processing Pile of work
  • 27.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 27 Serial vs. Parallel Processing Pile of work
  • 28.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 28 Serial vs. Parallel Processing Pile of work
  • 29.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 29 Serial vs. Parallel Processing
  • 30.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 30 Serial vs. Parallel Processing
  • 31.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 31 Serial vs. Parallel Processing
  • 32.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 32 Modern Multi-core CPUs Worker 1 Worker 2 Worker 3 Worker 4 Worker 5 Worker 6 Worker 7 Worker 8 Worker 9 Worker 10 Multiple “workers” capable of simultaneously handling compute, networking and I/O loads 10-cores
  • 33.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 33 Standard use of Multi-core CPUs in Virtual Servers VM 1 VM 2 VM 3 VM 4 VM 5 idle I/Oidle idle idle Parallel Compute Serial I/O VM = Virtual Machine
  • 34.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 34 Serial I/O Bottleneck in Virtualized Server  Compute waits on I/O Compute I/O Workload
  • 35.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 35 Serial I/O Bottleneck in Virtualized Server  Compute waits on I/O  CPU cores are wastedCompute I/O Workload
  • 36.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 36 Serial I/O Bottleneck in Virtualized Server  Compute waits on I/O  CPU cores are wasted  Very little work gets done Compute I/O Workload
  • 37.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 37 Serial I/O Bottleneck in Virtualized Server  Compute waits on I/O  CPU cores are wasted  Very little work gets done Compute I/O Workload
  • 38.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 38 Serial I/O Bottleneck in Virtualized Server  Compute waits on I/O  CPU cores are wasted  Very little work gets done Compute I/O Workload
  • 39.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 39 Impact: Workload Server 1
  • 40.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 40 Impact: Many servers needed to spread I/O Workload Server 1 Server 2 Server 3
  • 41.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 41 Impact: Many servers needed to spread I/O Workload Server 1 Server 2 Server 3 Server 4 Server 5
  • 42.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 42 Turbo-Charge through Parallel I/O  I/O keeps pace with compute demandsCompute I/O Workload
  • 43.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 43 Turbo-Charge through Parallel I/O  I/O keeps pace with compute demands  CPU cores are fully used Compute I/O Workload
  • 44.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 44 Turbo-Charge through Parallel I/O  I/O keeps pace with compute demands  CPU cores are fully used  Lots of work gets done in very little time Compute I/O
  • 45.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 45 Adaptive Parallel I/O Workload 45
  • 46.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 46 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 46
  • 47.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 47 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 47
  • 48.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 48 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 48
  • 49.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 49 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 400,000 IOPS < 1 millisec 49
  • 50.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 50 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 400,000 IOPS < 1 millisec 50
  • 51.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 51 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 400,000 IOPS < 1 millisec 51
  • 52.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 52 Adaptive Parallel I/O Workload Response Time (millisec) IOPS 400,000 IOPS < 1 millisec 52
  • 53.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 53 Adaptive Parallel I/O Workload No more load 400,000 IOPS < 1 millisec 53
  • 54.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 54 Adaptive Parallel I/O Workload No more load 400,000 IOPS < 1 millisec 54
  • 55.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 55 Adaptive Parallel I/O Workload No more load 400,000 IOPS < 1 millisec 55
  • 56.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 56 Adaptive Parallel I/O Workload No more load 400,000 IOPS < 1 millisec 56
  • 57.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 57 Adaptive Parallel I/O Workload No more load 400,000 IOPS < 1 millisec 57
  • 58.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 58 Adaptive Parallel I/O Workload No more load 400,000 IOPS < 1 millisec 58
  • 59.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. Worker 1 Worker 2 Worker 3 Worker 4 Worker 5 Worker 6 Worker 7 Worker 8 Worker 9 Worker 10 DataCore’s Adaptive use of Multi-core CPUs in Virtual Servers VM 1 VM 2 VM 3 VM 4 VM 5 I/O Parallel Compute Parallel I/O VM = Virtual Machine I/OI/O I/O I/O
  • 60.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 60 Parallel I/O Breakthrough TIME SAVINGS COST SAVINGS Work completes in 1/5 the time 2 machines can do the work of 10
  • 61.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 61 • 4x 2U Servers • 3 TB Cache • 25.6 TB of PCIe Flash • 168 TB total mirrored capacity (336TB total raw) • N-Node Mesh Mirror • 2 Full Racks • 256 GB cache • 17.6TB of array based SSD • 140TB total mirrored capacity (280TB total raw) • 2-Node Mirror NetApp Proposed Storage Real World: Parallel I/O’s Impact on Infrastructure More than Just Hyper-Converged DataCore Server SAN
  • 62.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. 62 • 4x 2U Servers • 3 TB Cache • 25.6 TB of PCIe Flash • 168 TB total mirrored capacity (336TB total raw) • N-Node Mesh Mirror • More Capacity • Faster Performance • < 1/4 of Space • 2 Full Racks • 256 GB cache • 17.6TB of array based SSD • 140TB total mirrored capacity (280TB total raw) • 2-Node Mirror NetApp Proposed Storage Real World: Parallel I/O’s Impact on Infrastructure More than Just Hyper-Converged DataCore Server SAN
  • 63.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved.  Parallel I/O changes the storage equation  Independent benchmark proves DataCore offers ► More IOPS ► Lower Price ► Better Value ► Lower Latency  Run Mission-Critical applications on DataCore 63 Summary
  • 64.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. World-Record Price Performance 64 http://www.storageperformance.org/results/benchmark_results_spc1_top-ten
  • 65.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved.  Try Parallel I/O in your environment ► To speed up response for latency-sensitive applications ► To increase the number of concurrent workloads (VMs) per server  Download Free Trial of DataCore™ Hyper-converged Virtual SAN http://www.datacore.com/resources/free-hyper-converged-virtual-san  More Info on SPC @ http://www.storageperformance.org/results/benchmark_results_spc1_top-ten 65 Next Steps
  • 66.
    Copyright © 2016DataCore Software Corp. – All Rights Reserved. Thank You! www.datacore.com 66

Editor's Notes

  • #3 V16 – Updated October 12, 2015
  • #23 IO has long been a serial function designed for single processor machines. While processor speeds climbed, IO processing benefitted from it and kept pace with computational gains. However, when power and heat concerns flattened the CPU clock speed ramp, chipmakers switched their focus on more cores per socket. Now, multi-core designs have the potential to do far more compute work across separate virtual machines than the serial IO is able to process on a single CPU. In other words, serial IO limits the potential work that multi-core CPUs can accomplish. And that IO gap is widening. The physical machine can’t get enough input and output to keep all those processors busy computing – so CPUs go idle and we are forced to spread the work over more underutilized servers.
  • #24 IO has long been a serial function designed for single processor machines. While processor speeds climbed, IO processing benefitted from it and kept pace with computational gains. However, when power and heat concerns flattened the CPU clock speed ramp, chipmakers switched their focus on more cores per socket. Now, multi-core designs have the potential to do far more compute work across separate virtual machines than the serial IO is able to process on a single CPU. In other words, serial IO limits the potential work that multi-core CPUs can accomplish. And that IO gap is widening. The physical machine can’t get enough input and output to keep all those processors busy computing – so CPUs go idle and we are forced to spread the work over more underutilized servers.
  • #33 Even the most modest server-class machines come equipped with multi-core CPUs. Take Lenovo’s ThinkServer RD530 with 10 CPU Cores for under $2,000. Instead of having just one CPU working for you, the system effectively has 10 workers that can be used simultaneously. Several applications can run concurrently with CPUs individually dedicated to them.