Professional Documents
Culture Documents
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020
Introducing the Dell EMC PowerStore 7000 series storage arrays
In addition to outperforming the HPE Primera A670 in our hands-on testing, Dell EMC PowerStore 7000 series
storage arrays can provide the following features:
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 2
Testing data reduction, performance, and usability on the Dell EMC and HPE solutions
We used default configuration settings for all testing and followed recommendations from each vendor’s
published best practices. Both models in the Dell EMC PowerStore 7000 series are two-node arrays; for
consistency, we also used a two‑node HPE Primera A670 array for our testing. Below, we outline how we tested
data reduction, performance, and usability on the three arrays. For detailed results of all our testing, see the
science behind the report.
We used a storage benchmarking tool called Vdbench to measure data reduction on the Dell EMC PowerStore
arrays and the HPE Primera A670. Both Dell EMC PowerStore arrays achieved the same data reduction ratio.
Performance testing
Dell EMC PowerStore 7000T vs. HPE Primera A670
Using Vdbench, we measured block storage performance on both arrays under different I/O workloads. In this
report, we present the results of performance testing on the Dell EMC PowerStore 7000T and HPE Primera A670.
We configured the hypervisor-enabled Dell EMC PowerStore 7000X to host internal VMs running Vdbench and
compared its storage performance to that of the HPE Primera A670 running Vdbench on externally hosted VMs.
Next, we ran a MongoDB workload on VMs that the PowerStore 7000X hosted internally. Simultaneously, we ran
a Vdbench workload on externally hosted VMs on both solutions. We captured several performance indicators:
• The IOPS that both solutions supported while running Vdbench on internally hosted VMs (on the Dell EMC
PowerStore 7000X) and externally hosted VMs (on the HPE Primera A670).
• The database operations per second and database application latency that the Dell EMC PowerStore
7000X delivered while running a MongoDB workload on internally hosted VMs. Because the HPE Primera
A670 lacks the capability to host VMs internally, we could not make a comparison for this test.
• The bandwidth and IOPS that both solutions supported while running Vdbench on externally hosted VMs.
Note that for this and the above scenario, the Dell EMC PowerStore 7000X was simultaneously running
workloads on internal and external VMs, while the HPE Primera A670 was only running a workload on
externally hosted VMs.
Usability testing
Dell EMC PowerStore 7000X vs. HPE Primera A670
We tested out-of-the-box VM deployment on the PowerStore 7000X versus the HPE Primera A670 (which
required separately configured VM hosts), capturing the time required to deploy a VM in a VMware environment.
We measured how quickly we could access restored data on 10 storage volumes on the PowerStore 7000T
versus the HPE Primera A670.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 3
About the metrics we used to measure storage performance
Our Vdbench testing offers insight into storage performance by showing:
• The number of input/output operations per second (IOPS) a solution can handle, indicating
whether it can process a high volume of requests
• The speed with which a storage solution can respond (storage latency), minimizing the chance
that users and applications will experience long wait times
• The amount of data a storage solution can process per second (bandwidth), indicating how
well it can process a high volume of data
Next, we measured how well each solution deduplicated and compressed the data; that is, how much duplicate
data it recognized and eliminated, and how much data it compressed. We did this by running a write test,
which contained an equal number of 128KB and 256KB blocks, on the data set. The Dell EMC PowerStore 7000
series arrays offered an overall 7.1:1 data reduction ratio compared to the 2.0:1 ratio of the HPE Primera A670,
meaning that the Dell arrays offered about three and a half times the data reduction of the HPE Primera A670
(see Figure 1).
If your organization is heavily utilizing storage, you need a solution that can keep up with capacity demands. As
our testing demonstrated, the Dell EMC PowerStore 7000T and 7000X arrays reduced data more efficiently than
the HPE Primera A670, providing more usable storage capacity.
Figure 1: Amount of physical storage used, after data reduction, while running a 50% 128KB, 50% 256KB write test.
Lower is better. Source: Principled Technologies.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 4
Performance testing
Dell EMC PowerStore 7000T vs. HPE Primera A670 232,602 DELL EMC
POWERSTORE 7000T
Support more IOPS HPE PRIMERA A670
Storage arrays employ data reduction, a feature that cuts down 171,772 (DATA REDUCTION OFF)
on overall storage utilization by reducing incoming data in real HPE PRIMERA A670
time. On the HPE Primera A670, users must make the choice of (DATA REDUCTION ON)
whether to turn data reduction on (to free storage space) or off
(to improve storage performance). On the Dell EMC PowerStore 75,160
arrays, users do not need to make this choice, since the arrays
offer always-on data reduction. We explored how enabling and
disabling data reduction might affect performance on the HPE
IOPS ON AN 8KB RANDOM
Primera A670, and how it compared to the performance of the 100% WRITE WORKLOAD
Dell EMC PowerStore 7000T array. UP TO
First, we ran an 8KB random 100% write workload on the Dell 209%
MORE
EMC PowerStore 7000T. While using always-on data reduction,
the array achieved 232,602 IOPS (see Figure 2). Next, we ran the Figure 2: IOPS supported while running a
Vdbench workload, with data reduction enabled
same workload on the HPE Primera A670 with data reduction
and disabled on the HPE Primera A670. The data
turned off. The HPE array produced 171,772 IOPS. When we reduction feature on the Dell EMC PowerStore
enabled data reduction on the HPE Primera A670 and ran the arrays is always on. Higher is better. Source:
workload again, the array produced 75,160 IOPS—less than Principled Technologies.
half of what it had produced with data reduction disabled.
These results indicate that businesses using the HPE array with data reduction turned on (as is standard, to free
capacity) would see a 53 percent performance reduction compared to running the same workload with data
reduction turned off. The HPE Primera A670 produced 35 percent fewer IOPS than the Dell EMC PowerStore
7000T with data reduction disabled—a mode that also risks the array using and running out of storage space
sooner. Note that, after running this test, we enabled data reduction on the HPE Primera A670 for the remainder
of our testing.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 5
23,239 23,238 DELL EMC
POWERSTORE 7000T
Process more data with
higher bandwidth HPE PRIMERA A670
116% 135%
Primera A670 on both workloads, supporting up
to 135 percent more bandwidth (see Figure 4).
MORE MORE
An array with high bandwidth capabilities helps
process more data for large data requests such Figure 4: Bandwidth (MB/s) provided while running two Vdbench
as streaming video or big data applications. workloads with 256KB data blocks. Higher is better.
Source: Principled Technologies.
72%
application and user experience.
LESS
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 6
Dell EMC PowerStore 7000X vs. HPE Primera A670
Host database VMs internally while providing storage resources to external hosts
In addition to providing storage, the Dell EMC PowerStore 7000X has an embedded hypervisor that enables
users to deploy, host, and manage VMs on the array itself—a capability that could reduce the initial need to
acquire separate servers for storage and compute.
We ran four scenarios to test various aspects of storage performance on both arrays.
• In Scenario 1, we ran a simple Vdbench simulation with internal Vdbench VMs hosted inside the Dell EMC
PowerStore 7000X array, versus external Vdbench VMs hosted on two-socket servers connected to the HPE
Primera A670 array.
• In Scenarios 2-4, we ran a MongoDB database workload with internal VMs hosted inside the PowerStore
7000X array as the array externally ran other Vdbench simulations simultaneously. We ran the same
external Vdbench simulations on servers connected to the HPE Primera A670. Table 1 explains this
scenario in more detail.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 7
Scenarios 2-4: Testing database performance on the Dell EMC PowerStore 7000X
(simultaneously hosting database VMs internally and VMs running Vdbench
simulations externally) vs. the HPE Primera A670 (hosting only VMs running Vdbench
simulations externally)
For these scenarios, we ran two types of workloads simultaneously: (1) Yahoo Cloud Serving Benchmark (YCSB),
which simulates analytics workloads, on internal database VMs, and (2) Vdbench on external VMs. This second
workload type included two variations throughout the test window, which we describe in Table 1 as Scenarios
3 and 4.
Table 1: Testing database performance on the Dell EMC PowerStore 7000X (simultaneously hosting database VMs
internally and VMs running Vdbench simulations externally) vs. the HPE Primera A670 (hosting only VMs running Vdbench
simulations externally).
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 8
Scenario 2: Dell EMC PowerStore 7000X internally hosting
MongoDB VMs running a YCSB workload
The Dell EMC PowerStore 7000X supported
188,320 database operations per second. 0.41 DELL EMC POWERSTORE 7000X
While doing so, the array
0.33
maintained submillisecond read and
update (write) database application
latencies (see Figure 7).
UP TO
188,320
DATABASE OPERATIONS
PER SECOND READ UPDATE
APPLICATION (WRITE) APPLICATION
DATABASE LATENCY (MS) DATABASE LATENCY (MS)
HPE PRIMERA A670 Figure 7: Database application latency (ms) delivered by the
Dell EMC PowerStore 7000X while hosting internal VMs and
CANNOT HOST
simultaneously delivering storage to external hosts. The HPE
INTERNAL VMS, SO WE Primera A670 cannot host internal VMs, so we could not make
COULD NOT MAKE A a comparison. Source: Principled Technologies.
COMPARISON
BANDWIDTH
(MB/S)
31%
MORE
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 9
Scenario 4: Dell EMC PowerStore 7000X and HPE Primera A670 externally hosting VMs running
a Vdbench workload (IOPS)
In addition to providing greater bandwidth, the Dell EMC PowerStore 7000X produced 65 percent more IOPS
than the HPE Primera A670 (see Figure 9), even as it was simultaneously hosting MongoDB VMs internally in
Scenario 2.
The wins in Scenarios 2-4 are particularly striking when you consider that the HPE Primera A670 lacks the
capability to host internal VMs and support internal workloads. Even with applications consuming internal
compute and storage resources on the Dell EMC PowerStore 7000X, the performance this Dell EMC array
delivered to externally based VMs was not just adequate, but greater than that of the HPE array.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 10
Greater versatility with a smaller
data center footprint
The Dell EMC PowerStore 7000X ran compute and
8U storage simultaneously while occupying just 2U of
rack space. If businesses using the HPE Primera
A670 wanted to host VMs in a highly available (HA)
environment, they would need to purchase additional
2U servers, increasing the total footprint of the HPE
Primera A670 array to 8U (the two-node array takes
Dell EMC HPE Primera A670 up 4U, and our testbed used two 2U servers). With its
PowerStore 7000X 2x additional 2U servers compact form factor, the Dell EMC PowerStore 7000X
to host VMs
could help organizations save on data center costs by
delaying the need to expand—all while enjoying the
Figure 10: Amount of space taken up by storage
array plus any servers necessary to provide
flexibility of an AppsON infrastructure.
storage resources while hosting database VMs.
Lower is better. Source: Principled Technologies.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 11
Usability testing
Dell EMC PowerStore 7000X vs. HPE Primera A670 array
Spend less time on out-of-the-box VM deployment DELL EMC POWERSTORE 7000X
Traditionally, storage systems haven’t been able to support embedded virtual HPE PRIMERA A670
machines and serve as storage targets at the same time. The Dell EMC
PowerStore 7000X offers a virtualized VMware-based environment out of the 53s
box that allows admins to provision storage to VMs residing on the array itself
with a few clicks. The Dell EMC PowerStore array also automatically provisions 8m 15s
Tier 1 storage to the embedded VMware environment.
OUT-OF-THE-BOX
Deploying a VM that could access the HPE Primera A670 required us to add VM DEPLOYMENT
physical cable connections, configure the Fibre Channel switch, map to the
host, deploy storage LUNs, and create a VMware datastore. This expanded VM 9X
FASTER
deployment time to 8 minutes and 15 seconds, versus 53 seconds on the Dell
EMC PowerStore 7000X (see Figure 11).
Figure 11: Time to provision a
VM out of the box (minutes and
seconds). Lower is better.
Source: Principled Technologies.
Dell EMC PowerStore 7000T vs. HPE Primera A670
Access snapshot restore data faster
Taking snapshots of storage LUNs is a common task for storage administrators.
DELL EMC POWERSTORE 7000T
Depending on the needs and policies of their data centers, admins might
be taking snapshots on a weekly, daily, or even hourly basis. When data HPE PRIMERA A670
corruption or loss occurs, the admins can then use the snapshots they have
taken to restore data to the system. Businesses typically deploy and manage 1m 7s
large numbers of LUNs—making these snapshot restores a potentially time-
consuming task. 20m 17s
We tested how long the Dell EMC PowerStore 7000T and HPE Primera ACCESS SNAPSHOT
A670 took to conduct a snapshot restore of 10 LUNs mounted as raw RESTORE DATA
device mappings (RDM) onto a single VM. The Dell EMC PowerStore 7000T
accomplished this task in a little over a minute, requiring just 16 steps. The 17.1x
FASTER
HPE Primera A670 array, by contrast, required administrators to un-export
the volumes from the application VM and host.2 It took over 20 minutes (see Figure 12: Time to access
Figure 13) and 31 steps to complete this task. For more details about our snapshot restore data (minutes
and seconds). Lower is better.
testing, see the science behind the report. With faster access to snapshot
Source: Principled Technologies.
restore data, admins could get operations up and running sooner.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 12
Conclusion
We tested two all-flash Dell EMC PowerStore 7000 series arrays against an HPE Primera A670 array. In hands-on
testing, both Dell EMC PowerStore 7000 series arrays reduced data more efficiently and offered greater storage
performance, as measured by IOPS, bandwidth, and latency. The hypervisor-enabled PowerStore array also
saved time and hardware on out-of-the-box VM deployment compared to the HPE Primera A670. With Dell EMC
PowerStore 7000 series arrays, organizations could get more out of their storage.
1 Vdbench uses the LZJB compression algorithm. However, Dell EMC PowerStore arrays use a different compression
algorithm, which may result in different savings. We set the deduplication unit for the HPE Primera A670 to
16 KB to follow the best practices from HPE found here: https://h20195.www2.hpe.com/v2/getdocument.
aspx?docname=a50000209enw. We set the deduplication unit for the Dell EMC PowerStore arrays to 4 KB.
2 HPE, “HPE SSMC 3.6 User Guide1SSMC,” accessed July 10, 2020, https://support.hpe.com/hpesc/public/
docDisplay?docLocale=en_US&docId=emr_na-a00085217en_us.
Principled Principled
Facts matter.® Facts matter.®
Technologies ®
Technologies ®
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 13
Disclaimer:
This document describes what we tested, how we tested, and what we found. To learn how these facts translate
into real-world benefits, read the report Enable greater data reduction and storage performance with Dell EMC
PowerStore 7000 series storage arrays.
We concluded our hands-on testing on June 24, 2020. During testing, we determined the appropriate hardware
and software configurations and applied updates as they became available. The results in this report reflect
configurations that we finalized on June 4, 2020 or earlier. Unavoidably, these configurations may not represent
the latest versions available when this report appears.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020
Our results
Table 1: Results of our testing on the Dell EMC™ PowerStore™ 7000T and HPE Primera A670.
Vdbench testing
50% 128KB and 50% 256KB 7.1:1 (reduced 800 GB of data to 2.0:1 (reduced 800 GB of data to
3.5x the data reduction
write workload 111 GB) 401 GB)
Bandwidth (MB/s)
Latency (ms)
Time (seconds) 11 11
Steps 5 5
Steps N/A 3
Steps N/A 6
Steps 10 4
Steps N/A 5
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 2
Dell EMC PowerStore 7000T HPE Primera A670 Win
Steps N/A 2
Steps N/A 5
Time (seconds) 4 4
Steps 1 1
Total
Table 2: Results of our testing on the Dell EMC PowerStore 7000X and HPE Primera A670.
50% 128KB and 50% 256KB 7.1:1 (reduced 800 GB of data to 2.0:1 (reduced 800 GB of data to
3.5x the data reduction
write workload 111 GB) 401 GB)
Scenario 1: Testing storage performance on the Dell EMC PowerStore 7000X (hosting VMs internally) vs. the HPE Primera A670 (hosting
VMs externally)
Scenarios 2-4: Testing database performance on the Dell EMC PowerStore 7000X (simultaneously hosting database VMs internally and
VMs running Vdbench simulations externally) vs. the HPE Primera A670 (only hosting VMs running Vdbench simulations externally)
Scenario 2: Dell EMC PowerStore 7000X internally hosting MongoDB VMs running a YCSB workload
Database operations
188,320 N/A N/A
per second**
Scenario 3: Dell EMC PowerStore 7000X and HPE Primera A670 externally hosting VMs running a Vdbench workload (bandwidth)
Scenario 4: Dell EMC PowerStore 7000X and HPE Primera A670 externally hosting VMs running a Vdbench workload (IOPS)
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 3
Dell EMC PowerStore 7000X HPE Primera A670 Win
Out-of-the-box VM deployment
Steps N/A 7
Mapping hosts
Steps N/A 8
Steps N/A 6
Steps N/A 7
Steps N/A 6
Time (seconds) 53 47
Steps 12 9
Total
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 4
System configuration information
Table 3: Detailed information on the servers we tested.
Processor
Number of processors 2 2 2
Memory module(s)
Size (GB) 16 16 16
Vendor and model Dell PERC H330 Dell PERC H330 Dell PERC H330
Local storage
Number of drives 2 2 2
Drive vendor and model Samsung® MZ-7LH480A Micron® MTFDDAK480TDC Micron MTFDDAK480TDC
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 5
Server configuration 4x Dell EMC PowerEdge™
4x Dell EMC PowerEdge R740 4x Dell EMC PowerEdge R740
information R740
Storage adapter
Number and type of ports 2x 2-port 32 Gb Fibre Channel 2x 2-port 32 Gb Fibre Channel 2x 2-port 32 Gb Fibre Channel
Power supplies
Storage configuration
Dell EMC PowerStore 7000X Dell EMC PowerStore 7000T HPE Primera A670
information
Software version 1.0.0.0.5.109 1.0.0.0.5.109 4.1.0.27
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 6
Detailed testing procedure
We received three separate testbeds from Dell EMC: one for testing the Dell EMC PowerStore 7000T, one for testing the PowerStore 7000X,
and one for testing the HPE Primera A670. In addition to the storage arrays we tested, each testbed had four Dell EMC PowerEdge R740
servers equipped with two dual-port 32GB Emulex Fibre Channel adapters and VMware ESXi 6.7. With the exception of the internal VMs
running on the PowerStore 7000X array, we performed all testing using Fibre Channel storage protocol and configured the arrays to use 16
ports connected to a 48-port Fibre Channel switch. We used 10GbE switches for testbed management and VM traffic. Where possible, we
verified that the setups on the testbeds were identical, and we configured the hosts, multipathing, and arrays as closely as possible following
best practices published by each storage vendor. For all of our testing, we used CentOS 7 VMs with Vdbench 5.04.07. We configured each
VM with eight vCPUs and 24 GB of RAM.
We performed storage efficiency tests on the HPE Primera A670 array using 16KB deduplication unit sizes, as HPE recommends (https://
h20195.www2.hpe.com/v2/getdocument.aspx?docname=a50000209enw). Both Dell EMC PowerStore arrays can deduplicate 4KB
deduplcation unit sizes, so we tested with this deduplication unit size on the PowerStore 7000X and 7000T arrays.
We used a 50/50 mix of 128KB and 256KB sequential writes with a single thread to fill the LUNs with data:
compratio=3
dedupratio=2
dedupunit=16k (4k for PowerStore arrays)
dedupsets=50%
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
sd=sd3,hd=hd1,lun=/dev/sdd
sd=sd4,hd=hd1,lun=/dev/sde
wd=wd_default,sd=*
wd=wd_prefill,sd=sd*,xfersize=(128k,50,256k,50),seekpct=eof,rdpct=0
rd=default
rd=fill_1,wd=wd_prefill,elapsed=10h,interval=10,iorate=max,forthreads=(1)
rd=fill_2,wd=wd_prefill,elapsed=10h,interval=10,iorate=max,forthreads=(1)
Performance testing
Dell EMC PowerStore 7000T vs. HPE Primera A670
We created 32 800GB volumes and mapped these to all four hosts in each testbed. After rescanning the hosts HBAs, we attached the
volumes as RDM disks to 16 Vdbench test VMs, each of which contained two RDMs. We used Vdbench to prefill the volumes and run the
test. Once the prefilling had completed, we ran through a series of Vdbench I/O simulations and collected the results. After completing the
tests, we unmapped the RDMs and volumes from the VMs and hosts and deleted the volumes on the array. We waited until the number of
uninitialized chunklets on the HPE Primera A670 array returned to zero before we repeated the volume provisioning and testing process
twice to get three runs per array. We present the median results.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 7
Prefilling the data
We used 256KB sequential writes with a single thread to fill all 32 volumes with data:
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
dedupsets=5%
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_prefill,sd=sd*,xfersize=256k,seekpct=eof,rdpct=0
rd=default
rd=rd_prefill,wd=wd_prefill,elapsed=10h,interval=10,iorate=max,forthreads=(1)
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
….
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_256k,sd=sd*,xfersize=256k,seekpct=100
rd=default
rd=read256k_test,wd=wd_256k,iorate=max,warmup=30,interval=10,forrdpct=(100),elapsed=120,forthreads=(4)
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 8
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_256k,sd=sd*,xfersize=256k,seekpct=seqnz
rd=default
rd=read256k_test,wd=wd_256k,iorate=max,warmup=30,interval=10,forrdpct=(100),elapsed=120,forthreads=(4)
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_OLTP2A_RRH,sd=*,rhpct=100,rdpct=100,xfersize=8K,skew=20,range=(0,6m)
wd=wd_OLTP2A_RM,sd=*,rdpct=100,xfersize=8k,skew=45,range=(0,100)
wd=wd_OLTP2A_RW,sd=*,rdpct=0,xfersize=8K,skew=15,range=(0,100)
wd=wd_OLTP2A_SR,sd=*,rdpct=100,seekpct=seqnz,range=(0,100),xfersize=128K,skew=10
wd=wd_OLTP2A_SW,sd=*,rdpct=0,seekpct=seqnz,range=(0,100),xfersize=128K,skew=10
rd=default
rd=rd_OLTP2A,wd=wd_OLTP2A_*,iorate=max,pause=120,elapsed=120,interval=10,warmup=30,forthreads=8
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
….
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
….
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_32k,sd=sd*,xfersize=32k,seekpct=100
rd=default
rd=read32k_
test,wd=wd_32k,iorate=max,pause=120,warmup=30,interval=10,forrdpct=(70),elapsed=120,forthreads=(16)
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 9
Running the 8KB random writes workloads
For these tests, we configured Vdbench to run an 8KB random writes workload with 16 threads:
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_8k,sd=sd*,xfersize=8k,seekpct=100
rd=default
rd=read8k_
test,wd=wd_8k,iorate=max,pause=120,warmup=30,interval=10,forrdpct=(0),elapsed=120,forthreads=(16)
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_OLTP2A_RRH,sd=*,rhpct=100,rdpct=100,xfersize=8K,skew=20,range=(0,6m)
wd=wd_OLTP2A_RM,sd=*,rdpct=100,xfersize=8k,skew=45,range=(0,100)
wd=wd_OLTP2A_RW,sd=*,rdpct=0,xfersize=8K,skew=15,range=(0,100)
wd=wd_OLTP2A_SR,sd=*,rdpct=100,seekpct=seqnz,range=(0,100),xfersize=128K,skew=10
wd=wd_OLTP2A_SW,sd=*,rdpct=0,seekpct=seqnz,range=(0,100),xfersize=128K,skew=10
rd=default
rd=rd_OLTP2A,wd=wd_OLTP2A_*,pause=120,iorate=(107000),elapsed=120,interval=10,warmup=60,forthreads=(8)
After completing these tests, we repeated the 8KB random write test on the HPE Primera A670 with deduplication and compression disabled
on 32x 250GB volumes. We repeated the volume provisioning and testing process twice more to get three test runs. We present the
median results.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 10
Dell EMC PowerStore 7000X with internal VMs vs. HPE Primera A670 with external VMs
The Dell EMC PowerStore 7000X array can allow VMs to consume compute and storage resources within the array without the need for
external server hosts. We refer to VMs using this internal storage deployment model as “internal VMs” in this report. We provisioned four
internal Vdbench test VMs on the Dell EMC PowerStore 7000X and four external VMs on the HPE Primera A670 testbed. On the Dell EMC
PowerStore 7000X, we split these four VMs evenly across the two-node VMware PowerStore cluster. On the Primera array, we distributed the
four VMs evenly between two Dell EMC PowerEdge R740 hosts. We created 16 800GB volumes on both arrays and assigned four volumes
to each VM. We used Vdbench to prefill the volumes and run the test. Once Vdbench completed the prefilling, we ran through a series of
Vdbench I/O simulations and collected the results. After we completed the tests, we unmapped the RDMs and volumes from the VMs and
hosts and deleted the volumes on the array. We waited until the number of uninitialized chunklets on the Primera array returned to zero
before we repeated the volume provisioning and testing process twice more to get three runs per array. We present the median results.
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore arrays)
dedupsets=5%
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
hd=hd3,system=vdb-005
hd=hd4,system=vdb-006
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd15,hd=hd4,lun=/dev/sdd
sd=sd16,hd=hd4,lun=/dev/sde
wd=default,sd=*
wd=wd_prefill,sd=sd*,xfersize=256k,seekpct=eof,rdpct=0
rd=default
rd=rd_prefill,wd=wd_prefill,elapsed=10h,interval=10,iorate=max,forthreads=(1)
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore arrays)
messagescan=no
hd=default,shell=ssh,user=root
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
….
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_4k,sd=sd*,xfersize=4k,seekpct=100
rd=default
rd=read4k_
test,wd=wd_4k,iorate=max,pause=120,warmup=30,interval=10,forrdpct=(0),elapsed=120,forthreads=(32)
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 11
Dell EMC PowerStore 7000X with internal MongoDB database VMs and external Vdbench VMs
vs. HPE Primera A670 with external Vdbench VMs
We set up a MongoDB replication set consisting of three internal MongoDB VMs on the Dell EMC PowerStore 7000X cluster. We used
a separate YCSB VM running on one of the Dell EMC PowerEdge R740 hosts in the PowerStore 7000X testbed to drive the MongoDB
workload. Simultaneously, we created 32 400GB volumes on each array and mapped them to four Dell EMC PowerEdge R740 hosts on each
testbed. We created RDM disks out of the 32 mapped volumes and mounted two RDMs per VM to 16 external VMs. We distributed all 16
VMs evenly between four Dell EMC PowerEdge R740 hosts. We used Vdbench to prefill the volumes and run the I/O test. Once Vdbench
completed the prefilling, we initialized the replica set on the MongoDB VMs and used the external YCSB VM to load the MongoDB database
with 100 million records. The 16 external VMs connected on the PowerStore 7000X and HPE Primera testbeds ran through 256KB and
8KB Vdbench I/O workloads. While the Vdbench tests were running on the PowerStore 7000X, the three internal VMs were also running a
workload of 50 million operations driven by the external YCSB VM, using four separate YCSB processes and 64 total threads. We collected
Vdbench results for the HPE Primera A670 and both Vdbench and MongoDB results for Dell EMC PowerStore 7000X. Below, Figure 1
demonstrates how we simultaneously ran workload on internally and externally hosted VMs on the Dell EMC PowerStore 7000X. Because the
HPE Primera A670 lacks the ability to host VMs internally, we only ran the external Vdbench workloads on it. After the tests completed, we
deleted all the volumes and the MongoDB databases and allowed the number of uninitialized chunklets on the HPE Primera array return to
zero. We repeated the volume provisioning and testing process twice more to get three runs per array. We present the median results.
256KB sequential read Vdbench workload running on external VMs 8KB random write Vdbench workload running on external VMs
0 1 2 3 4 5 6 7 8 9 10
TIME IN MINUTES
Figure 1: Time (minutes) it took to run two Vdbench workloads on externally hosted VMs while simultaneously running a
workload on internally hosted VMs on the Dell EMC PowerStore 7000X. Source: Principled Technologies.
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore arrays)
dedupsets=5%
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_prefill,sd=sd*,xfersize=256k,seekpct=eof,rdpct=0
rd=default
rd=rd_prefill,wd=wd_prefill,elapsed=10h,interval=10,iorate=max,forthreads=(1)
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 12
Running the sequential reads workloads
For these tests, we configured Vdbench to run 256KB sequential reads with three threads:
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore arrays)
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_256k,sd=sd*,xfersize=256k,seekpct=seqnz
rd=default
rd=read256k_test,wd=wd_256k,iorate=max,warmup=30,interval=10,forrdpct=(100),elapsed=300,forthreads=(3)
compratio=3
dedupratio=3
dedupunit=16k (4k for PowerStore array)
messagescan=no
hd=default,shell=ssh,user=root
hd=hd1,system=vdb-001
hd=hd2,system=vdb-002
…
hd=hd15,system=vdb-015
hd=hd16,system=vdb-016
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
…
sd=sd31,hd=hd16,lun=/dev/sdb
sd=sd32,hd=hd16,lun=/dev/sdc
wd=default,sd=*
wd=wd_8k,sd=sd*,xfersize=8k,seekpct=100
rd=default
rd=read8k_
test,wd=wd_8k,iorate=max,pause=300,warmup=30,interval=10,forrdpct=(0),elapsed=300,forthreads=(16)
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 13
Running the MongoDB test (Dell EMC PowerStore 7000X only)
We used an externally hosted YCSB VM to run 50 million operations against our MongoDB replica set using four separate YCSB processes:
Usability testing
Dell EMC PowerStore 7000X vs. HPE Primera A670
Out-of-the-box VM deployment
This testing scenario assumed that the user had initialized both arrays but not performed Fibre Channel zoning or host mapping. Our
scenario also assumed that the HPE Primera A670 could use an available target VMware ESXi host and that the user had deployed a VMware
vCenter® VM on the Dell EMC PowerStore and HPE Primera testbeds.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 14
Creating one virtual volume on the HPE Primera A670
1. Under Primera and 3PAR SSMC dropdown, select Virtual Volumes.
2. Click Create Virtual Volume.
3. On General screen, enter a virtual volume name, switch dedup and compression to Yes. For volume size, enter 1 TiB.
4. Expand Export screen, and click Add.
5. On Add screen, select the appropriate host, and click Add.
6. On Create Virtual Volume screen, click Create.
Creating a single VMFS datastore from a LUN on the HPE Primera A670
1. Right-click the target host, and select StorageNew Datastore.
2. On the Type screen, select VMFS, and click Next.
3. On the Name and device selection screen, enter a datastore name, and select a LUN for provisioning the datastore. Click Next.
4. On the VMFS version screen, select VMFS 6, and click Next.
5. On the Partition configuration screen, use default settings, and click Next.
6. On the Ready to complete screen, click Finish.
Prefilling RDM data on the PowerStore 7000T and HPE Primera A670
For these tests, we used Vdbench to prefill RDM data:
compratio=2
dedupratio=2
dedupunit=16k (4k for PowerStore array)
dedupsets=5%
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=CB_command
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 15
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
sd=sd3,hd=hd1,lun=/dev/sdd
sd=sd4,hd=hd1,lun=/dev/sde
sd=sd5,hd=hd1,lun=/dev/sdf
sd=sd6,hd=hd1,lun=/dev/sdg
sd=sd7,hd=hd1,lun=/dev/sdh
sd=sd8,hd=hd1,lun=/dev/sdi
sd=sd9,hd=hd1,lun=/dev/sdj
sd=sd10,hd=hd1,lun=/dev/sdk
wd=default,sd=*
wd=wd_prefill,sd=sd*,xfersize=256k,range=(0,60),seekpct=eof,rdpct=0
rd=default
rd=rd_prefill,wd=wd_prefill,elapsed=1h,interval=10,iorate=max,forthreads=(1)
Taking a manual snapshot of RDM volumes on the Dell EMC PowerStore 7000T
1. In the PowerStore Manager UI, choose StorageVolume Groups.
2. Select the name of the volume group containing the RDM volumes.
3. On the volume group properties screen, click the Protection tab.
4. Under the Protection tab, click Take Snapshot.
5. On the Take Snapshot screen, keep the defaults, and click Take Snapshot.
Running steady state I/O to RDM volumes on the PowerStore 7000T and HPE Primera A670
For these tests, we used Vdbench to run steady state I/O to RDM volumes:
compratio=2
dedupratio=2
dedupunit=16k (4k for PowerStore array)
dedupsets=5%
messagescan=no
hd=default,shell=ssh,user=root,jvms=1
hd=hd1,system=CB_command
sd=default,openflags=o_direct
sd=sd1,hd=hd1,lun=/dev/sdb
sd=sd2,hd=hd1,lun=/dev/sdc
sd=sd3,hd=hd1,lun=/dev/sdd
sd=sd4,hd=hd1,lun=/dev/sde
sd=sd5,hd=hd1,lun=/dev/sdf
sd=sd6,hd=hd1,lun=/dev/sdg
sd=sd7,hd=hd1,lun=/dev/sdh
sd=sd8,hd=hd1,lun=/dev/sdi
sd=sd9,hd=hd1,lun=/dev/sdj
sd=sd10,hd=hd1,lun=/dev/sdk
wd=default,sd=*
wd=wd_OLTP2A_RRH,sd=*,rhpct=100,rdpct=100,xfersize=8K,skew=20,range=(0,6m)
wd=wd_OLTP2A_RM,sd=*,rdpct=100,xfersize=8k,skew=45,range=(0,30)
wd=wd_OLTP2A_RW,sd=*,rdpct=0,xfersize=8K,skew=15,range=(0,30)
wd=wd_OLTP2A_SR,sd=*,rdpct=100,seekpct=seqnz,range=(0,30),xfersize=128K,skew=10
wd=wd_OLTP2A_SW,sd=*,rdpct=0,seekpct=seqnz,range=(0,30),xfersize=128K,skew=10
rd=default
rd=rd_OLTP2A,wd=wd_OLTP2A_*,iorate=max,elapsed=1h,interval=10,warmup=30,forthreads=(1)
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 16
Quiescing the CentOS application VM on the Dell EMC PowerStore 7000T and HPE Primera A670
1. Open a browser, and navigate to the vCenter vSphere client.
2. Enter credentials, and click Login.
3. Expand the data center.
4. Expand the cluster.
5. Right-click the VM, and choose PowerPower Off.
The HPE Primera A670 cannot promote snapshots if the user exports the target virtual volumes to a host (see https://support.hpe.
com/hpesc/public/docDisplay?docLocale=en_US&docId=emr_na-a00085217en_us). The HPE Primera A670 requires the user complete
the following steps to restore a snapshot successfully:
Removing the LUNs from the ESXi environment on the HPE Primera A670
1. In the vCenter UI, right-click the VM, and choose Edit Settings.
2. Expand Hard Disks.
3. Remove each RDM from the VM, choose to delete files from the datastore, and click OK.
Unexporting the Virtual Volume Set from the ESXi host on the HPE Primera A670
1. In Primera and 3PAR StoreServ Management Console (SSMC), under the Primera and 3PAR SSMC dropdown, click Show All.
2. Click Virtual Volume Sets.
3. Select the virtual volume set containing the base volumes.
4. Click Actions, and choose Unexport.
5. On the Unexport screen, select the volume set, and click Unexport.
6. Check the I have read and understand… checkbox, and click Yes, unexport.
Exporting LUNs back to the ESXi environment on the HPE Primera A670
1. In Primera and 3PAR SSMC, select the virtual volume set containing the base volumes.
2. Click Actions, and choose Export.
3. On the Export screen, in the Export To section, click Add.
4. On the Add screen, choose the ESXi host.
5. On the Export screen, click Export.
Adding LUNs back to the virtualized ESXi environment on the HPE Primera A670
1. In the vCenter UI, if necessary, expand the data center and cluster.
2. To discover newly added storage, right-click the target host, and select StorageRescan Storage.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 17
Adding LUNs back to the VM as RDM volumes on the HPE Primera A670
1. In the vCenter UI, right-click the VM, and choose Edit Settings.
2. Click Add New Device, and click RDM Disk.
3. Select one of the 100GB LUNS, and click OK.
4. Repeat step 2 until you have added all 10 LUNs as RDMs.
5. On Edit Settings screen, click OK.
6. Right-click the CentOS application VM, and choose PowerPower On.
Principled Principled
Facts matter.® Facts matter.®
Technologies ®
Technologies ®
In no event shall Principled Technologies, Inc. be liable for indirect, special, incidental, or consequential damages in connection with its testing, even if advised of the
possibility of such damages. In no event shall Principled Technologies, Inc.’s liability, including for direct damages, exceed the amounts paid in connection with Principled
Technologies, Inc.’s testing. Customer’s sole and exclusive remedies are as set forth herein.
Enable greater data reduction and storage performance with Dell EMC PowerStore 7000 series storage arrays August 2020 | 18