VMware vSphere6 FT arch perf

VMware vSphere6 FT arch perf
VMware vSphere 6
Fault Tolerance
Architecture and Performance
TECHNICAL WHITE PAPER
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Table of Contents
Introduction .................................................................................................................................................................................................................. 3
vSphere FT Architecture ......................................................................................................................................................................................... 3
Storage .....................................................................................................................................................................................................................4
Runtime State .......................................................................................................................................................................................................4
Network ...................................................................................................................................................................................................................4
Transparent Failover ..........................................................................................................................................................................................4
Best Practices...............................................................................................................................................................................................................4
Fault Tolerance Performance ................................................................................................................................................................................ 5
Kernel Compile ..................................................................................................................................................................................................... 5
Netperf ..................................................................................................................................................................................................................... 5
IOMeter.....................................................................................................................................................................................................................8
Swingbench with Oracle 11g ...........................................................................................................................................................................9
DVD Store with Microsoft SQL Server 2012 .......................................................................................................................................... 10
Brokerage Workload Derived from TPC-E ............................................................................................................................................ 12
vCenter Server .................................................................................................................................................................................................... 13
Fault Tolerance Performance Summary ........................................................................................................................................................ 16
Appendix A: Benchmark Setup ......................................................................................................................................................................... 17
Storage Array ...................................................................................................................................................................................................... 17
Primary and Secondary Hosts ..................................................................................................................................................................... 17
Client Machine..................................................................................................................................................................................................... 17
Appendix B: Workload Details ........................................................................................................................................................................... 17
Kernel Compile ................................................................................................................................................................................................... 17
Netperf ................................................................................................................................................................................................................... 17
Iometer ................................................................................................................................................................................................................... 17
Swingbench with Oracle 11g ......................................................................................................................................................................... 18
DVD Store with Microsoft SQL Server 2012 .......................................................................................................................................... 18
Brokerage Workload Derived from TPC-E ............................................................................................................................................ 19
vCenter Server Workload .............................................................................................................................................................................. 19
References ................................................................................................................................................................................................................... 21
TECHNICAL WHITE PAPER / 2
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Introduction
VMware vSphere® Fault Tolerance (FT) provides continuous availability to virtual machines, eliminating
downtime and disruption—even in the event of a complete host failure. This white paper gives a brief description
of the vSphere FT architecture and discusses the performance implication of this feature with data from a wide
variety of workloads.
vSphere FT Architecture
vSphere FT enables a virtual machine to survive a physical server failure by creating an exact replica virtual
machine on another host that can take over at the time of failure. During failover, the transition of a vSphere FT
virtual machine from one physical server to another is similar to a migration using vSphere vMotion®: it is
completely seamless. That means there is zero downtime, zero data loss, zero connection loss, continuous service
availability, and complete transaction integrity.
Figure 1. Fast Checkpointing keeps the secondary virtual machine in sync with the primary virtual machine.
vSphere FT works by continuously replicating an entire running virtual machine from one physical server to
another. The result is that an FT-protected virtual machine has two replicas: the primary virtual machine and the
secondary virtual machine, each running on distinct ESXi hosts. These replicas are logically identical —they
represent a single virtual machine state and a single network identity, but they are physically distinct. Each
replica has its own set of virtual machine files (including VMX and VMDK files), which vSphere FT automatically
keeps in sync. When a physical server fails, one of the replicas will resume execution, and the virtual machine
state, the network identity, and all active network connections for the virtual machine will be identical, ensuring a
seamless failover process.
vSphere FT is made possible by four underlying technologies: storage, runtime state, network, and transparent
failover.
TECHNICAL WHITE PAPER / 3
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Storage
vSphere FT ensures the storage of the primary and secondary virtual machines is always kept in sync. When
vSphere FT protection is enabled, an initial synchronization of the VMDKs happens using VMware vSphere
Storage vMotion® to ensure the primary and secondary have the exact same disk state.
This initial Storage vMotion happens whenever FT is turned on while the virtual machine is running, whenever FT
protection is re-established after a failover occurs, or whenever a powered-off FT virtual machine powers on. The
FT virtual machine is not considered FT-protected until the Storage vMotion completes.
After this initial synchronization, vSphere FT will mirror VMDK write operations between the primary and
secondary over the FT network to ensure the storage of the replicas continues to be identical.
Runtime State
vSphere FT ensures the runtime state of the two replicas is always identical. It does this by continuously
capturing the active memory and precise execution state of the virtual machine, and rapidly transferring them
over a high-speed network, allowing the virtual machine to instantaneously switch from running on the primary
ESXi host to the secondary ESXi host whenever a failure occurs.
Network
The networks being used by the virtual machine are also virtualized by the underlying ESXi host, ensuring that
even after a failover, the virtual machine identity and network connections are preserved. Similar to vSphere
vMotion, vSphere FT manages the virtual MAC address as part of the process. If the secondary virtual machine is
activated, the secondary host sends a gratuitous ARP so the network switch is aware of the new physical location
of the virtual MAC address. Since vSphere FT preserves the storage, the precise execution state, the network
identity, and the active network connections, the result is zero downtime and no disruption to users should an
ESXi host failure occur.
Transparent Failover
vSphere FT ensures that the primary always agrees with the secondary about the state of the virtual machine.
This is achieved by holding externally visible output from the virtual machine, and only releasing it when an
acknowledgement is made from the secondary affirming that the state of the two virtual machines is consistent
(for the purposes of vSphere FT, externally visible output is network transmissions). The cost for maintaining this
zero data loss consistency is that network transmissions from the virtual machine are delayed until the two
virtual machines are in a consistent state.
Best Practices
Based on performance testing described in the next section, a wide range of virtual machine workloads work well
under FT protection. Among these are CPU-bound workloads, I/O-bound workloads, servers, and complex
database workloads. One of the most common performance observations of virtual machines under FT
protection is a variable increase in the network latency of the virtual machine (increasing both latency and jitter).
But as the performance testing in the next section demonstrates, good application performance is typically
possible under FT protection. Workloads that are particularly latency sensitive, such as high frequency trading
(HFT) or real-time systems may experience more overhead.
See the Performance Best Practices for VMware vSphere 6.0 [2] guide for recommendations regarding vSphere
FT setup.
Additionally, this paper applies the recommendations found in the following KB articles to improve performance.
•
KB 2039495
Large packet loss at the guest OS level on the VMXNET3 vNIC in ESXi
TECHNICAL WHITE PAPER / 4
VMware vSphere 6 Fault Tolerance
Architecture and Performance
•
KB 1010071
The output of esxtop shows dropped receive packets at the virtual switch
•
KB 2111976
After you enable Fault Tolerance on a Windows VM (Windows XP and later) configured with Virtual
Hardware version 11 and vSphere 6.0 VM Tools, the virtual machine's workload throughput will drop
significantly
Fault Tolerance Performance
This section discusses the performance characteristics of FT virtual machines using a variety of microbenchmarks and real-life workloads. Micro-benchmarks were used to stress CPU, disk, and network subsystems
individually by driving them to saturation. Real-life workloads, on the other hand, were chosen to represent what
most customers would run, and these workloads have been configured to have a CPU utilization of 60 percent in
steady state. Identical hardware testbeds were used for all the experiments, and the performance comparison
was done by running the same workload on the same virtual machine with and without FT enabled. The
hardware and experimental setup details are provided in Appendix A: Benchmark Setup and Appendix B:
Workload Details. For each experiment (where applicable), the traffic on the FT logging NIC during the steady
state portion of the workload is also provided as a reference.
Kernel Compile
This experiment shows the time taken to do a parallel compile of the Linux kernel. This is a both a CPU- and
MMU-intensive workload due to the forking of many parallel processes. During this benchmark the CPU was 100
percent utilized. This workload did some disk reads and writes, but generated no network traffic. As seen in
Figure 2, FT protection increases the kernel compile time a small amount—about 7 seconds.
Kernel Compile
FT Traffic:
1 vCPU: 2.1 Gbps
2 vCPU: 2.6 Gbps
4 vCPU: 2.4 Gbps
(in seconds)
146.15
1 vCPU
161.03
72.28
2 vCPU
82.90
37.22
4 vCPU
44.09
0
20
40
60
80
FT Off
100
120
140
160
180
FT On
Figure 2. Kernel compilation performance (lower is better)
Netperf
Netperf is a micro-benchmark that measures the performance of sending and receiving network packets. Netperf
was configured with several scenarios and network speeds to demonstrate the throughput and latency
performance of TCP/IP under FT. The results were achieved while applying the recommendations provided in
the following KB articles: KB 2039495, KB 1010071, and KB 2111976.
TECHNICAL WHITE PAPER / 5
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Throughput
This netperf experiment measures unidirectional TCP/IP throughput. One experiment was done in each direction,
when the virtual machine was either receiving or transmitting data. The speed of the virtual machine network is
an important factor for performance, and one experiment was done on a 1 Gbps uplink (Figure 3) and another
was done for a 10 Gbps uplink (Figure 4).
The throughput experiments reveal some important points about network performance when vSphere FT is
enabled:
•
Workload traffic under FT protection can meet or exceed 1 Gbps for both 1 Gbps and 10 Gbps uplinks.
•
The FT network traffic is minimal for workloads where the virtual machine is transmitting a lot of data, but
not receiving much data.
•
The FT network traffic is high for workloads where the virtual machine is receiving a lot of data, but not
transmitting much data.
Receive-heavy workloads (where the virtual machine is receiving large amounts of data) tend to increase FT
traffic due to the requirement to keep the replicas in sync. The influx of data into the primary causes large
differences between the replicas, thus requires more data to be sent over the FT network. Transmit-heavy
workloads (where the virtual machine is transmitting a lot of data), on the other hand, cause very few differences
between the replicas, thus very little FT traffic. Therefore, transmit-heavy applications, such as Web servers or
read-bound databases, tend to have fewer FT traffic requirements.
As shown in Figure 4, the receive workload on a 10 Gbps virtual machine network is challenged when FT
protected, achieving less than 3 Gbps.
Netperf - Throughput (1 Gbps VM Network)
FT traffic:
Receive run:
3.1 Gbps
950
Receive
Transmit run:
331 Mbps
946
971
Transmit
872
0
250
500
750
1000
Mbps
FT Off
FT On
Figure 3. Netperf throughput for receive and transmit on a 1 Gbps VM network (higher is better)
TECHNICAL WHITE PAPER / 6
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Netperf - Throughput (10 Gbps VM Network)
FT traffic:
Receive run:
5.9 Gbps
9.5
Receive
Transmit run:
529 Mbps
2.4
9.5
Transmit
8.0
0.0
2.0
4.0
6.0
8.0
10.0
Gbps
FT Off
FT On
Figure 4. Netperf throughput on a 10 Gbps VM network for receive and transmit (higher is better)
Latency
This netperf experiment measures TCP/IP latency under FT. Fault tolerance introduces some delay to network
output (measureable in milliseconds, as seen in Figure 5). The latency occurs because an FT-protected primary
withholds network transmissions until the secondary acknowledges to the primary that it has reached an identical
runtime state.
In this experiment, netperf was run with the TCP_RR configuration (single stream, no parallel streams) and the
round-trip latency is reported here (it is the inverse of the round-trip transaction rate). TCP_RR is a pure latency
benchmark: the sender transmits a 1-byte message and blocks waiting for a 1-byte response, the benchmark
counts the number of serial transactions completed in a unit time, and there is no parallelization.
In a pure latency benchmark, all latency increases drop throughput (for example, latency increases 57 times and
throughput drops identically). In real-world applications, however, workload performance is generally robust
against network latency increases.
This is because normal server applications are not pure latency benchmarks. They handle multiple connections at
a time, and each connection will transmit several packets worth of data before pausing to hear a response. The
result is that real-world applications tolerate network latencies without dropping throughput. The netperf
throughput experiment is an example of this, and the client/server workloads examined in this paper
demonstrate the same thing. In all these experiments, while low-level network latency may increase, it typically
results in slight visible effects on perceived UI interactivity, latency for database queries, and overall application
throughput.
One aspect not measured by netperf is jitter and latency fluctuation. FT-protected virtual machines can vary
widely in latency depending on the workload, and over time within a given workload. This can cause significant
TECHNICAL WHITE PAPER / 7
VMware vSphere 6 Fault Tolerance
Architecture and Performance
jitter. Highly latency-sensitive applications, such as high frequency trading (HFT), or some voice-over-IP (VoIP)
applications may experience high overhead with FT. However, some voice applications, where the bulk data is
carried by separate processes where only call management traffic is FT-protected, would perform well.
Network Latency
FT traffic:
273.9 Mbps
(Milliseconds)
0.1
6.3
0.0
1.0
2.0
3.0
FT Off
4.0
5.0
6.0
7.0
FT On
Figure 5. Network latency for VMs with FT off and on (lower is better)
IOMeter
IOMeter is an I/O subsystem measurement and characterization tool for Microsoft Windows. It is designed to
produce a mix of operations to stress the disk. This benchmark ran random I/Os of various types. Figure 6 shows
the FT-protected virtual machine achieves nearly as much throughput as the non-protected virtual machine.
TECHNICAL WHITE PAPER / 8
VMware vSphere 6 Fault Tolerance
Architecture and Performance
IOMeter: Throughput
FT Traffic:
2KB reads:
167 Mbps
4.32
4.31
2KB Read
2KB writes:
62 Mbps
2.07
2.12
2KB Write
2KB R/W
(50%/50%)
2KB reads/writes:
79 Mbps
3.47
3.5
37.92
36.42
64KB Read
9.00
9.14
64KB Write
0
10
20
FT Off
30
64KB reads:
208 Mbps
64KB writes:
45 Mbps
40
FT On
Figure 6. IOMeter throughput results for 2KB and 64KB message sizes (higher is better)
Swingbench with Oracle 11g
In this experiment, an Oracle 11g database was driven using the Swingbench 2.2 [3] order entry online transaction
processing (OLTP) workload. This workload has a mixture of CPU, memory, disk, and network resource
requirements. The FT-protected virtual machine is able to achieve nearly as much throughput as the non-FT
virtual machine (Figure 7). The latency of basic operations has increased under FT protection (Figure 8).
TECHNICAL WHITE PAPER / 9
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Oracle Swingbench: Throughput
FT Traffic:
(Transactions/Sec)
1 vCPU: 0.9 Gbps
2 vCPU: 1.3 Gbps
4 vCPU: 1.6 Gbps
96.6
1 vCPU
82.02
145.14
2 vCPU
127.1
189.5
4 vCPU
170.45
0
50
100
FT Off
150
200
FT On
Figure 7. Swingbench OLTP workload run against an Oracle 11g database VM with vs. without FT (higher is better)
Oracle Swingbench: Response Time
(Milliseconds)
Browse
Product
(4 vCPU)
61
142
Process
Order
(4 vCPU)
91
163
Browse
Order
(4 vCPU)
103
192
0
50
100
FT Off
150
200
FT On
Figure 8. Benchmark response time for three common actions run against the Oracle 11g database VM with vs. without FT
(lower is better)
DVD Store with Microsoft SQL Server 2012
In this experiment, the DVD Store 2 [3] benchmark was used to drive the Microsoft SQL Server® 2012 database.
This benchmark simulates online transaction processing of a DVD store. 64 simultaneous user sessions were used
to drive the workload. As with the Swingbench benchmark, this workload has a mixture of CPU, memory, disk,
and networking resource requirements.
TECHNICAL WHITE PAPER / 10
VMware vSphere 6 Fault Tolerance
Architecture and Performance
FT protection incurs throughput overhead for this workload (Figure 9), and latency for basic operations has
increased (Figure 10).
Disclaimer: The throughput here is not meant to indicate the absolute performance of Microsoft SQL Server®
2012, nor to compare its performance to another DBMS. SQL Server was used to place a DBMS workload on ESXi,
and observe the performance of ESXi with and without FT.
Microsoft SQL Server 2012 DVD Store:
Throughput
FT Traffic:
1 vCPU: 0.8 Gbps
2 vCPU: 1.1 Gbps
4 vCPU: 1.6 Gbps
(Operations/min)
2218
1 vCPU
1782
4493
2 vCPU
3599
9041
4 vCPU
6832
0
2000
4000
FT Off
6000
8000
10000
FT On
Figure 9. DVD Store throughput for basic operations of a Microsoft Server 2012 VM with vs. without FT enabled (higher is
better)
TECHNICAL WHITE PAPER / 11
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Microsoft SQL Server 2012 DVD Store:
Response Time
(Milliseconds)
23
1 vCPU
118
18
2 vCPU
113
16
4 vCPU
132
0
20
40
60
80
FT Off
FT On
100
120
140
Figure 10. DVD Store response time for basic transactions against a Microsoft SQL Server 2012 VM with vs. without FT
enabled (lower is better)
Brokerage Workload Derived from TPC-E
This experiment uses a workload modeled on the TPC-E [5] benchmark, which simulates the online transaction
processing workload (OLTP) of a brokerage firm in which customers trade stock, query accounts, and research
markets. This workload has a mixture of CPU, memory, disk, and networking resource requirements. The
workload is particularly stressful and has a large working set.
Disclaimer: The brokerage workload is a non-comparable implementation of the TPC-E business model.
TECHNICAL WHITE PAPER / 12
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Brokerage Workload: Throughput
(Transactions/sec)
FT traffic: 4.4 Gbps
164
91
0
50
100
FT Off
150
FT On
Figure 11. Throughput of the brokerage workload with FT off and on (higher is better)
As shown in Figure 11, there is a reduction in user transaction throughput when FT is enabled. This is due to the
very intensive nature of this brokerage workload.
vCenter Server
This experiment shows the difference in performance of VMware vCenter Server™ when using a vCenter Server
appliance protected versus unprotected by Fault Tolerance. The baseline setup consisted of 32 hosts and 1000
virtual machines managed by a vCenter Server appliance that ran version 6.0 with an embedded vPostgres
database and was configured with 4 vCPUs and 16GB of RAM. Similar results were observed when running
vCenter Server on Windows with an external SQL Server database.
Workloads
Three workloads were tested to highlight the impact of VMware Fault Tolerance on vCenter Server:
1.
Management operations. In these workloads, a number of vCenter Server management operations were
issued to the hosts and virtual machines; for example, powering on virtual machines, cloning them, or
reconfiguring them. This test generated a load of many such operations per minute and measured the
resulting throughput with FT. Two load profiles were measured. Typical customer setups have operational
rates that are less than both profiles, so both loads represented higher load scenarios. “Load A” generated
tens of operations per minute, while “Load B” generated hundreds of operations per minute.
2. Browsing with the vSphere Web Client. The latency of browsing operations (for example, clicking on
summary pages or virtual machine list pages) was measured when FT was enabled.
3. Restart. The latency to restart the vCenter Server process was measured with FT enabled. This determined
the timeframe it takes for vCenter Server to recover from a software crash.
TECHNICAL WHITE PAPER / 13
VMware vSphere 6 Fault Tolerance
Architecture and Performance
These workloads have a mixture of memory churn, disk I/O, and network I/O. The management operations
induce network traffic between vCenter Server and the vSphere hosts, disk traffic to the embedded database,
and memory churn within the vCenter Server Appliance for tracking the creation, deletion, and migration of
various objects like virtual machines. The Web browsing workload generates network bandwidth from the
vCenter Server Appliance to the Web browser and induces memory churn within the appliance to handle session
management, object caching, and database query formulation. Finally, restarting vCenter Server causes network
traffic to retrieve data from vSphere hosts, disk I/O to persist data to the database, and memory churn in order to
fill internal caches in vCenter Server.
Results
The results are shown in Figure 12 (vCenter Server Management Operations), Figure 13 (Web Client Browsing),
and Figure 14 (Restarting vCenter Server).
Management Operations
When running a light load (approximately 27 operations per minute), a vCenter Server using FT can support
about 74% of the throughput of vCenter without FT. Similarly, a vCenter server under heavy load (nearly 200
operations per minute), is able to support about 40% of the throughput of vCenter without FT, as shown in
Figure 12.
vCenter Server Management Operations
FT Traffic:
(Operations/min)
Light: 2.7 Gbps
Heavy: 4.2 Gbps
27
Light Load
20
195
Heavy Load
80
0
50
100
FT Off
150
200
FT On
Figure 12. Operations per minute achieved by vCenter Server Load A (higher is better)
Browsing with the vSphere Web Client
The latency for various browsing operations increases under FT from an average of 1.1s without FT to 1.5s with FT.
The effect this has on UI browsing inside the Web Client, while noticeable, is such that the Web Client remains
comfortably usable.
TECHNICAL WHITE PAPER / 14
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Web Client Browsing
(Time in Seconds)
1.7
VM Summary Page
2.1
0.4
Action Menu for Cluster
0.5
1.67
0.6
Cluster Summary Page
1.0
0.8
Create VM Wizard
1.6
0.0
0.5
FT Off
1.0
1.5
2.0
2.5
FT On
Figure 13. The time it takes in seconds to perform common browsing activities with the vSphere Web Client (lower is
better)
Restarting vCenter Server
Finally, the restart time of vCenter Server with FT is approximately 1.3 times longer than without FT. The restart
time includes the amount of time it takes for the vCenter Server to come back online with the vSphere Web
Client available.
TECHNICAL WHITE PAPER / 15
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Restarting vCenter Server
(Time in Minutes)
1.7
2.3
0.0
0.5
1.0
FT Off
1.5
2.0
2.5
FT On
Figure 14. Restarting vCenter Server to simulate failover due to application crash (lower is better)
For more information about protecting vCenter Server with FT, see the VMware vCenter Server Availability Guide
on page 8 [6].
Fault Tolerance Performance Summary
Based on performance testing described in the previous section, a wide range of virtual machine workloads work
well under FT protection. Among these are CPU-bound workloads, I/O-bound workloads, servers, and complex
database workloads. One of the most common performance observations of virtual machines under FT
protection is a variable increase in the network latency of the virtual machine (increasing both latency and jitter).
But as the performance testing in the previous section demonstrates, good application performance is typically
possible under FT protection; network latency does not dictate overall application throughput for a wide variety
of applications. On the other hand, applications that are highly sensitive to network latency (such as high
frequency trading or real-time workloads) have a higher performance cost under FT protection.
VMware vSphere Fault Tolerance is a revolutionary new technology. It universally applies fault-tolerant
technology to any multi-vCPU workload in a uniquely simple-to-use way. The vSphere FT solution is able to
achieve good throughput for a wide variety of applications.
TECHNICAL WHITE PAPER / 16
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Appendix A: Benchmark Setup
Storage Array
•
System: EMC VNX 5300
•
FLARE OS: 05.31.000.5.720
•
LUNs: RAID 5 LUNs (6 disks), RAID 0 LUNS (6 disks)
Primary and Secondary Hosts
•
System: HP ProLiant DL380 Gen9
•
CPU: Intel® Xeon® Processor E5-2667 v3 @ 3.20GHz
•
Number of cores/socket: 8; Number of sockets: 2; Hyper-Threading: enabled
•
Memory: 128 GB
Client Machine
•
System: Dell PowerEdge R720
•
CPU: Intel® Xeon® Processor E5-2650 @ 2.00GHz
•
Number of cores/socket: 8; Number of sockets: 2; Hyper-Threading: enabled
•
Memory: 64 GB
•
OS: Windows Server 2012 R2 Datacenter Edition, 64-bit
Appendix B: Workload Details
Kernel Compile
•
Virtual machine configuration: 1 vCPU / 2 vCPUs / 4 vCPUs, 2 GB RAM, LSI Logic virtual SCSI adapter
•
OS version: RHEL6
•
Kernel version: 2.16.32-71.el16
•
Benchmark details: Time taken to compile (make bzImage) Linux kernel 2.6.20 was measured. The
experiment was repeated 5 times and the average run time was reported.
Netperf
•
Virtual machine configuration: 4 vCPUs, 8 GB RAM, VMXNET3 virtual NIC, LSI Logic Virtual SCSI adapter
•
OS version for both client/server: Windows Server 2012 R2 Datacenter Edition, 64-bit
•
Netperf configuration: 8 clients started in parallel, remote and local message size: 64K, remote and local
socket size: 4MB
•
The recommended configuration found in KB articles KB 2039495, KB 1010071, and KB 2111976 were applied
Iometer
•
Virtual machine configuration: 4 vCPUs, 8 GB RAM, VMXNET3 virtual NIC, LSI Logic Virtual SCSI adapter
•
OS version: Windows Server 2012 R2 Datacenter Edition, 64-bit
TECHNICAL WHITE PAPER / 17
VMware vSphere 6 Fault Tolerance
Architecture and Performance
•
Benchmark parameters:
– 2k 100% read, 100% write, 50-50% read-write
– 64k 100% read, 100% write
Swingbench with Oracle 11g
•
Virtual machine configuration: 1 vCPU / 2 vCPUs / 4 vCPUs, 8 GB RAM, VMXNET3 virtual NIC, LSI Logic virtual
SCSI adapter
•
OS version: RHEL5
•
Kernel version: 2.16.18-164.el15
•
Oracle Version: 11.1.0.6.0
•
Database Details: max number of processes: 150, SGA buffer size: 1535MB, data file size: 23 GB (index, redo,
and database files on the same location)
•
Swingbench configuration:
– Swingbench version: 2.2, Calling Circle Database
– Number of orders: 23,550,492
– Number of customers: 864,967
– Runtime: 30 mins
– ODBC driver: ojdbc6.jar
– Driver type: Thin
– Number of users: 160
– Pooled: 1
– LogonDelay: 0
– Transaction MinDelay: 50
– Transaction MaxDelay: 250
– QueryTimeout: 60
– Workload Weightage: NewCustomerProcess – 20, BrowseProducts – 50, ProcessOrders – 10,
BrowseAndUpdateOrders – 50
Note: Database was restored from backup before every run.
DVD Store with Microsoft SQL Server 2012
•
Virtual machine configuration: 1 vCPU / 2 vCPUs / 4 vCPUs, 8 GB RAM, VMXNET3 virtual NIC, LSI Logic virtual
SCSI adapter
•
OS version: Windows Server 2012 R2 Datacenter Edition, 64 bit
•
Microsoft SQL Server version: SQL server 2012
•
Database size: 100 GB
•
Database row count: 200 million customers, 10 million orders per month, 1 million products
•
DVD Store benchmark version: 2007/12/03
•
Benchmark parameters
– n_threads: 64
– ramp_rate: 2
– run_time: 30 mins
– warmup_time: 4 mins
– think_time: 0.40 secs
TECHNICAL WHITE PAPER / 18
VMware vSphere 6 Fault Tolerance
Architecture and Performance
– pct_newcsutomers: 40
– n_searches: 5
– search_batch_size: 8
– n_line_items: 10
– db_size_str: L
Note: The database was restored from backup after every run.
Brokerage Workload Derived from TPC-E
•
Virtual machine configuration: 4 vCPUs, 42 GB RAM, VMXNET3 virtual NIC, PVSCSI virtual SCSI adapter
•
OS version: Windows Server 2012, Datacenter Edition, 64-bit
•
Microsoft SQL Server version: SQL Server 2012
•
Database scale: 115,000 customers
•
Number of user connections: 80
•
SQL Server buffer max memory size: 40 GB
vCenter Server Workload
•
Virtual infrastructure configuration:
– VMware vCenter Server 6.0 run in a VM with 4 vCPU and 16 GB memory
– 32 ESXi hosts each containing 125 VMs (4,000 VMs total)
•
Operations performed: See Table 1, below.
Operation
Description
Add Port Group
Create a new port group on a virtual switch.
Remove Port Group
Remove a port group from a virtual switch.
Clone VM
Create a clone of a virtual machine. 1
Create Folder
Create a new folder in the vCenter inventory hierarchy.
Delete Folder
Delete a folder from the vCenter inventory hierarchy.
Create Snapshot
Create a snapshot of a virtual machine.
Delete Snapshot
Delete a snapshot of a virtual machine.
Group Power-On VMs
Power on several virtual machines in a single operation.
vMotion VM
Move a powered-on virtual machine to a different host.
Power On VM
Power on a single virtual machine in a DRS cluster.
Power Off VM
Power off a single virtual machine.
Reconfigure VM
Edit a virtual machine’s configuration settings. 2
Register VM
Add a .vmx file from a datastore to the vCenter inventory.
Unregister VM
Remove a virtual machine from the vCenter inventory without deleting its files from
its datastore.
1
The performance of a clone operation depends on the disk size of the VM being cloned.
2
This benchmark’s Reconfigure operation uses VM Memory Shares as a representative configuration setting.
TECHNICAL WHITE PAPER / 19
VMware vSphere 6 Fault Tolerance
Architecture and Performance
Relocate VM
Move a powered-off virtual machine to a different host and datastore.
Remove VM
Delete a virtual machine from the vCenter inventory, and delete its files from its
datastore.
Reset VM
Reset a single virtual machine.
Suspend VM
Suspend a single virtual machine.
Resume VM
Resume a single virtual machine.
Table 1. Operations performed in vCenter Server workload
For more information about the vCenter Server workload, see “vCenter Server 6.0 Performance and Best
Practices [6].”
TECHNICAL WHITE PAPER / 20
VMware vSphere 6 Fault Tolerance
Architecture and Performance
References
[1] VMware, Inc. (2015, June) Performance Best Practices for VMware vSphere 6.0. [Online].
http://www.vmware.com/files/pdf/techpaper/VMware-PerfBest-Practices-vSphere6-0.pdf
[2] Dominic Giles. Swingbench. [Online]. http://dominicgiles.com/swingbench.html
[3] Todd Muirhead and Dave Jaffe. (2007, December) linux.dell.com. [Online].
http://linux.dell.com/dvdstore/2007-12-03/
[4] Transaction Processing Performance Council. (2015) TPC-E. [Online]. http://www.tpc.org/tpce/
[5] Mohan Potheri, G. Blair Fritz, and Puneet Gupta. (2015, May) VMware vCenter Server 6.0 Availability Guide.
[Online]. http://www.vmware.com/files/pdf/techpaper/VMware-vCenter-Server6-Availability-Guide.pdf
[6] Mike Stunes and Ravi Soundararajan. (2015, August) vCenter Server 6.0 Performance and Best Practices.
[Online]. https://communities.vmware.com/docs/DOC-29203
[7] VMware, Inc. (2015) vSphere Availability. [Online]. http://pubs.vmware.com/vsphere60/topic/com.vmware.ICbase/PDF/vsphere-esxi-vcenter-server-60-availability-guide.pdf
VMware, Inc. 3401 Hillview Avenue Palo Alto CA 94304 USA Tel 877-486-9273 Fax 650-427-5001 www.vmware.com
Copyright © 2016 VMware, Inc. All rights reserved. This product is protected by U.S. and international copyright and intellectual property laws. VMware products are covered by one or more patents listed at
http://www.vmware.com/go/patents. VMware is a registered trademark or trademark of VMware, Inc. in the United States and/or other jurisdictions. All other marks and names mentioned herein may be trademarks
of their respective companies. Date: 4-Jan-16 Comments on this document: https://communities.vmware.com/docs/DOC-30054
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising