testing(vm(density(–why(it(...
Post on 22-May-2020
0 Views
Preview:
TRANSCRIPT
Testing VM density – why it matters, and the 3PAR impact
Jeff Boles – Sr. Analyst / Director of Validation Services jeff@tanejagroup.com
2 © 2013 Taneja Group
Agenda
• What is VM density?
• How does it impact the IT infrastructure and TCO?
• The common culprit that decreases VM density.
• Tackling the common culprit, and a look at some hands-‐on tesJng.
3 © 2013 Taneja Group
What is VM density?
Physical World Virtual World
" Many:1 relationship between applications and hardware
" 1:1 relationship between applications and hardware
1:1
1:1
1:1 Many:1
1:1
1:1
1:1
VM density is the “Many:1” raJo, and idenJfies how many applicaJons you can consolidate on one hypervisor
4 © 2013 Taneja Group
VirtualizaJon has changed the way we pay for IT
Virtual World
" Much licensing is performed at the hypervisor level, and even more licensing can be controlled through the intelligent use of virtual processors.
Physical World
" Everything licensed at the OS and Application
5 © 2013 Taneja Group
There is lots that we pay for…
Factor in Significant AcquisiJon Costs
Management
• Servers • Networking • Storage • Electricity • Datacenter
space
" Licenses for Virtual machines operating system
" Management software
" Management servers and other infrastructure cost
" Databases software and servers
VM Guest OS Management Hardware
" Licenses for virtualization software
Virtualization software
6 © 2013 Taneja Group
Virtual Infrastructure Costs MaXer
6 Confidential
vCloud Suite Edition Lineup
$4,995 $7,495 $11,495
Cloud Infrastructure
Price (per CPU, license only)
Virtualization vSphere Enterprise Plus Virtualized infrastructure with policy-based automation
vCloud Director & vCloud Connector Virtualized datacenters with multi-tenancy and public cloud extensibility
vFabric Application Director Multi-tier application service catalog publishing and provisioning
vCenter Operations Management Suite - Performance monitoring and alerting - Capacity planning, reporting, and optimization
Standard vCloud Networking and Security Software defined networking, security, and ecosystem integration
vCenter Infrastructure Navigator Application discovery, dependency mapping, and management vCenter Chargeback Manager Metered chargeback reporting and accountability
vCenter Configuration Manager Configuration and compliance management of virtual infrastructure Operations
Management
Advanced vCloud Networking and Security Continuous firewall and network availability
vCOps Advanced vCOps Enterprise
vCenter Site Recovery Manager Enterprise Automated disaster recovery planning, testing, and execution
Note: vCloud Suite Editions are licensed per CPU; Unlimited VMs, Unlimited vRAM
vCloud Net & Sec Advanced
vCloud Net & Sec Standard
vCenter SRM
vCloud Suite Standard
vCloud Suite Advanced
vCloud Suite Enterprise
vCloud Net & Sec Advanced
7 © 2013 Taneja Group
These costs shouldn’t be ignored
Scale
Cos
t OpExSavings
CapExSavings
Total Cost of Ownership
Low VM density increases Opex costsand diminishes savings
8 © 2013 Taneja Group
These costs shouldn’t be ignored
Leading Drivers of Server Virtualization
50.1%
51.3%
59.1%
66.3%
30% 35% 40% 45% 50% 55% 60% 65% 70%
Improved availability and/or disaster recovery
Ease of management and increased flexibil ity ofIT operation
Power/cooling/space and related operationalsavings
Capital (CAPEX) savings from consolidatingphysical servers
9 © 2013 Taneja Group
What is VM density?
Physical World Virtual World
" Many:1 relationship between applications and hardware
" Relevant cost metric = COST PER APPLICATION
" 1:1 relationship between applications and hardware
" Relevant cost metric = cost per server
1:1
1:1
1:1 Many:1
1:1
1:1
1:1
VM density matters!
10 © 2013 Taneja Group
The Common Culprit
Storage!
11 © 2013 Taneja Group
A few data points (1 of 4) Average Number of Virtual Servers Per Physical Production Server
24.4%33.1%
15.4%9.3%
7.3%9.0%
1.5%
0 20 40 60 80 100 120 140
1 to 56 to 10
11 to 1516 to 2021 to 25
26 or moreDon't know
Storage Factors Preventing or Slowing the Virtualization of "Mission Critical" Applications
37.6%
14.8%
15.3%
18.1%
18.4%
19.8%
20.9%
21.2%
0 20 40 60 80 100 120 140 160
None of the above
Storage availability/uptime issues or concerns
Disaster recovery issues or concerns
Data protection issues or concerns
Insufficient storage capacity
Lack of required tools for visibil ity and management
Insufficient storage performance
Security issues or concerns
12 © 2013 Taneja Group
A few data points (2 of 4)
Impact of Storage Costs on Overall Server Virtualization ROI
20.1%
21.2%
35.1%
39.8%
0 20 40 60 80 100 120 140 160
Don't know
Storage operational costs (incl:troubleshooting, management, etc.) have
detracted from my return on investment (ROI)
Storage has not impacted my ROI one way oranother
Storage capital costs (buying more storage)have detracted from my ROI
13 © 2013 Taneja Group
A few data points (4 of 4)
Which of the following application areas have the greatest need for greater storage performance in your
business?
14 © 2013 Taneja Group
The 3PAR VM Density Difference
Number of simultaneous VM tiles
3PAR Other
Storage VM Density
2X VM
Density
15 © 2013 Taneja Group
What we tested… !"#$"%&#'()*+,(-.(/#
0.+).+#012/.-# !"###$%&'()*+,-).$/0%1/2#3$45$%&'(),$
0.+).+#"+,3.22,+2#
605$!78)$9)7:$;<=62<#$>?2$378),$-7-'&@$
0.+).+#&%4# A24%$
5./6,+7#012/.-#
BC$!"###$DE8-F'&$!7::)3-$>G&)0EH&+$I87JE,E7:,$JB%K,$':($JLM!,$G78$E:(EJE(F'&$IN+,E3'&$%1/2#3$H&'(),@$
89#9,((.3/*)*/1#
DE'$DE8-F'&$!7::)3-$O&)0=?#$?#4H$;-N)8:)-$P7(F&)$
'/:.+(./#9,((.3/*)*/1#
DE'$DE8-F'&$!7::)3-$5$4H$6/=C78-$OEH8)$!N'::)&$P7(F&)$
0/,+;<.#012/.-#
BC$QCKR$O6##$
=*272# Q6$?/24%$?<S$(E,T,U$,E:V&)$JE8-F'&EW)($('-',-78)$J7&F.)U$RKMX$<$
9;3:.# 54%$!7:-87&$Y$?64%$X'-'$$
!"#$%&%'(#)*+""#,*-(.%"'(/0(&*10".0"*1,2&0/* !"###$%&'()*+,-).$
/0%1/2#3$45$%&'(),$10".0"*3"'4022'"2*
605$!78)$9)7:$;<=62<#$>?2$378),$-7-'&@$
10".0"*5+6* A24%$
70&8'"9*1,2&0/*
BC$!"###$DE8-F'&$!7::)3-$>G&)0EH&+$I87JE,E7:,$JB%K,$':($JLM!,$G78$E:(EJE(F'&$IN+,E3'&$%1/2#3$H&'(),@$
:;*;'((04&%.%&,*
DE'$DE8-F'&$!7::)3-$O&)0=?#$?#4H$;-N)8:)-$P7(F&)$
-&<0"(0&*;'((04&%.%&,*
DE'$DE8-F'&$!7::)3-$5$4H$6/=C78-$OEH8)$!N'::)&$P7(F&)$
1&'"#=0*1,2&0/*
*E.E&'8$3&',,$$6$37:-87&&)8$'88'+$
>%292* Q6$?/24%$?<R$(E,S,T$,E:U&)$JE8-F'&EV)($('-',-78)$J7&F.)T$WKMX$<$
;#4<0* 54%$!'3N)$I)8$37:-87&&)8$$
16 © 2013 Taneja Group
What we tested with…
!
!"#$%&#'()%*+%#%$,-(.%%/01,2,1*#-%345'-4#1)%
%%
"#$%&#'(! )*$+#,-!
./01'23-!4565! 7'$3-8!#9!:#'(;-2!4565!8-,8!8##&!<!=>)?,@!6A;B!=CDE!
./01'23-!>&F-28! ./-0*8-(!:#'(;-2!4565!8-,8@!F2F8F'8-(!$-G#8-&H!=F'!)I.J.>!4!=>)?,@!4;B!=CDE!
IK:!I-$=-$!455<C4!
LML!I8#$-!4!L'8'N',-!O9#$!+1H,F0'&!8-,8!0&F-28!'00-,,P!Q!=>)?,@!<;B!=CDE!
MH'88'!ARQ! S'2(&-(!-/8-$2'&!0#22-08F=F8H!9#$!LML!I8#$-!4!+1H,F0'&!0&F-28!4!=>)?,@!T64EB!=CDE!
• A virtualized applicaJon Jle consisJng of several workloads, much like VMware’s VM Mark.
• Each applicaJon executed different syntheJc benchmarks that were selected for their reasonable ability to simulate real world workloads.
17 © 2013 Taneja Group
How we tested…
• We tested by cloning each of these test Jles repeatedly
• And then running simultaneous
workloads via a proprietary TG Labs test harness unJl failures occurred in the syntheJc tests
18 © 2013 Taneja Group
What this did…
19 © 2013 Taneja Group
What this did…
20 © 2013 Taneja Group
Success or failure
21 © 2013 Taneja Group
Demanding performance, but now with a twist…
22 © 2013 Taneja Group
Measuring results... without snaps
210 2 4 6 8 10 12 14 16 18
35,000
0
5000
10,000
15,000
20,000
25,000
30,000
Number of VMs
Ope
rati
ons
Per
Min
ute
- D
VD
Sto
re 2 HP 3PAR F200
Traditional Array
With HP 3PARperformance scales to16 VM tiles
Traditional arrayperformance peakat 8 VM tiles
23 © 2013 Taneja Group
Measuring results… with snaps
210 2 4 6 8 10 12 14 16 18
35,000
0
5000
10,000
15,000
20,000
25,000
30,000
Number of VMs
Ope
rati
ons
Per
Min
ute
- D
VD
Sto
re 2 HP 3PAR F200
Traditional Array
Traditional Arraycould not supportmore than 4 VM tiles
HP 3PAR F200performance peakedat approximately8 VM tiles
24 © 2013 Taneja Group
Latency makes a difference – 3PAR
Green is write latency, blue is read latency
25 © 2013 Taneja Group
Latency makes a difference – TradiJonal Array
Orange is write latency, red is read latency
26 © 2013 Taneja Group
A snapshot-‐induced latency melt down…
27 © 2013 Taneja Group
So does storage make a difference? • Our 4 blades were still only
seeing slightly more than 50% utilization.
• But the traditional array really fell apart from a lack of controller horsepower and disk efficiency.
• The F200 appeared to have more headroom if we scaled it with more disks.
• The traditional storage system wouldn’t take density anywhere near most customer targets.
top related