AN “APPLES TO APPLES”

COMPARISON
June, 2016
EMC CONFIDENTIAL: FOR INTERNAL
USE ONLY. DO NOT DISTRIBUTE

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 1
 EXECUTIVE SUMMARY:
PERVASIVE GC ISSUES

3PAR simply cannot meet the demands of enterprise iCDM and falls short in 3 key areas:
performance, flexibility and space reclamation:
• Performance
– Snapshot creation degrades production performance by -40%
– Volume performance drops to ZERO for several seconds while creating snapshots
– Performance is degraded indefinitely when snapshots are created with any frequency
– Recovering to pre-snapshot performance takes hours once snapshot creation has ceased
• Flexibility
– Restoring a parent volume from a snapshot results in extended downtime
– Refresh capabilities are extremely limited and pertain only to child objects
• Space Reclamation
– Dirty space constantly accumulates due to slow GC which causes the array to go offline. For successful
testing, utilization was dialed back to meager 25% of usable capacity!
– Tasks of capacity management and reconciliation are nearly impossible with array reporting negative space
savings of > -200% rendering deduplication virtually unusable

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 2
 COPY DATA
MANAGEMENT
✓ ✗
Instant restores Instantly! restore PROD from
Takes minutes/hours
backup or refresh from QA

Full performance ✓ ✗
copies No loss in PROD
performance -40% performance loss

Efficient ✓ ✗
retention/space Delete copies and reclaim Space reclamation requires
management space after retention period deleting all DEV/QA copies !

✓ ✗
Flexible workflow “Catch up” DEV/QA with Not allowed. Only leaf node
PROD – any depth allowed can be refreshed from parent
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 3
 3PAR VOLUMES: TPVV VS TDVV
• TPVV – Thin provisioned virtual volume
– Space efficiency provided via thin provisioning only
– No data reduction capabilities
– Snapshot implementation, COW (copy on write)

• TDVV – Thin deduped virtual volume

– Space efficiency provided via thin provisioning
– Data reduction capabilities via deduplication
– Snapshot implementation, ROW (redirect on write)

**Regardless of which volume type is selected, performance is

severely impacted when creating snapshots!

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 4
 WHILE GC TAKES 8+ HRS (!)
ONLY RUNS ON SUBSET OF VOLUMES

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 5
 CAUSING MASSIVE BUILDUP OF DIRTY SPACE

System reports 58,090GiB allocated

Though only 12,800 GiB has been

provisioned

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 6
 WARNING- ARRAY
FULL, OFFLINE & DU!
• It was impossible to pre-condition the
array at more than ~35% of the
usable capacity
• Attempts to do so were catastrophic:
• Array reached 100% capacity
• CPG went offline
• Volume were offline
• Volumes were corrupted
EMC CONFIDENTIAL: FOR INTERNAL
DANGER
D.U.
USE ONLY. DO NOT DISTRIBUTE
AHEAD

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 7
 PERFORMANCE
PERFORMANCE DEGRADES SUBSTANTIALLY FOR BOTH
COW (TPVV) & ROW (TDVV) SNAPSHOTS
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 8
 HP 3PAR 8440
TESTED ARRAY CONFIGURATION

• 4 controllers (nodes)
• 16 x 16Gb FC Ports (4 per
controller)
• 4 shelves
– 2 x DPE (controllers & SSDs)
– 2 X DAE (SSDs only)
• 48 x SSDs (12 per shelf)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 9
 TDVV SNAPSHOTS - PHASE 1
PERFORMANCE – SNAPSHOT IMPACT & RECOVERY

Purpose:
• To establish a performance baseline for volumes when no
snapshots exist on the array
• Observe and document the performance impact associated with
creating snapshots
• Observe and document the duration required for recovering to
the pre-snapshot performance baseline

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 10
 TDVV SNAPSHOTS - PHASE 1
TEST ENVIRONMENT

• Create 16 x 80GB TDVV volumes (a TDVV is a thinly provisioned deduplicated virtual

volume. The array will attempt to deduplicate the data stored on these volumes)
• Pre-condition using the AFA PoC Toolkit (fill & age)

• Run 12 hour steady state workload (50R/50W, mixed block sizes, 100% random)

• Snapshot schedule:
– No snaps during the first hour of testing to establish baseline
– At 1 hour mark, begin creating snapshots every 15min for a duration of 3:45
(hh:mm)
– Disable snapshot schedules for duration of testing - 7:15 (hh:mm)

**No snapshots were mounted during these tests

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 11
 PRODUCTION PERFORMANCE DROPS -40%
IMMEDIATE IMPACT OF CREATING SNAPSHOTS

1 Create first 2 Performance

snapshot drops -40%

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 12
 SUSTAINED DEGRADED PERFORMANCE

1 Create first 2 Performance 3 Last

snapshot drops -40% snapshot

Performance cannot recover!

(Duration: 3 hours 45 minutes @ 15min intervals)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 13
 PERFORMANCE FINALLY RECOVERS – 5½ HRS!
4 Performance
1 Create first 2 Performance 3 Last finally recovers
snapshot drops -40% snapshot in 5½ HRS!

Performance cannot recover! Takes 5½ hours to regain initial

(Duration: 3 hours 45 minutes @ 15min intervals) performance after snapshots stop

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 14
 SNAPS  HIGH WRITE LATENCY
High write latencies
~10ms caused by
inefficient write handling

No change in Read latency

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 15
 TDVV SNAPSHOTS - PHASE 2
SNAPSHOT PERFORMANCE

Purpose:
• To observe and document the performance characteristics of
the array when both volumes and snapshots of those volumes
are mounted
• To observe the impact of creating snapshots of both the
volumes and mounted snapshots

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 16
 TDVV SNAPSHOTS - PHASE 2
TEST ENVIRONMENT

• Create 16 x 80GB TDVV volumes

• Pre-condition using the AFA PoC Toolkit (fill & age)
• Create snapshots of the volumes
• Mount volumes and snapshots
• Run 12 hour steady state workload (50R/50W, mixed block sizes, 100%
random)
• Create scheduled snapshots of all mounted entities every 15 minutes
**Snapshots were mounted during these tests

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 17
 EMC - XTREMIO
TESTED ARRAY CONFIGURATION

• 4 controllers
• 8 x 8Gb FC Ports
(2 per controller)
• 2 shelves
– 2 X DAE (SSDs only)
• 50 x SSDs (25 per shelf)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 18
 XTREMIO VS 3PAR - IOPS
STEADY STATE IO, 50/50 R/W

XtremIO:
Mean IOPS 144,646

HP 3Par 8440:
Mean IOPS 83,711
(Drop of -33% compared to normal steady state)
INCONSISTENT PERFORMANCE !

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 19
 XTREMIO VS 3PAR - LATENCY
STEADY STATE IO, 50/50 R/W

XtremIO:
Mean Latency 3.54ms
StdDev 0.0695

HP 3Par 8440:
Mean Latency 6.19ms
StdDev 0.822
INCONSISTENT PERFORMANCE !

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 20
 8K PROFILING WITH SNAPSHOTS
MAX IOPS AT 1.03MS

READ % 3PAR 8440 3PAR 8440 % Decrease

IOPS without IOPS with due to
snapshots snapshots snapshots
0% 166,218 113,732 32%
35% 206,800 168,626 18%
50% 247,028 174,880 29%
80% 305,991 245,512 20%
100% 459,004 418,283 9%

3PAR’s SNAPSHOTS SIGNIFICANTLY PENALIZE PERFORMANCE

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 21
 TDVV: ROW SNAPSHOTS
MASSIVE WRITE AMPLIFICATION

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 22
 TDVV SNAPSHOTS - PHASE 3
WRITE AMPLIFICATION

Purpose:
• To compare and contrast the front end bandwidth requested by
the hosts versus the resulting backend bandwidth on the
physical SSDs when writing unique data
• To compare and contrast the front end bandwidth requested by
the hosts versus the resulting backend bandwidth on the
physical SSDs when writing 4:1 dedupe data

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 23
 TDVV SNAPSHOTS - PHASE 2
TEST ENVIRONMENT

• Create 16 x 80GB TDVV volumes

• Pre-condition using the AFA PoC Toolkit (fill & age)
• Mount volumes and snapshots
• Run steady state workload with unique data (100% write, mixed block sizes,
100% random)
• Create scheduled snapshots every 15 minutes
• Run steady state workload with 4:1 dedupe data (100% write, mixed block
sizes, 100% random)
• Create scheduled snapshots every 15 minutes

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 24
 HOST BW VS PHYSICAL DISK BW
PERFORMANCE COMPARISON UNIQUE VS DEDUPE DATA READ

WRITE

TOTAL

Host writing Unique data

Dedupe data

@ ~1.1GB/s

DEDUPE
CHAOS

3PAR writing
to SSDs @
~2.5GB/s

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 25
 HOST BW VS PHYSICAL DISK BW
PERFORMANCE IMPACT OF SNAPSHOTS READ

WRITE

TOTAL

Host writing Unique data

Dedupe data

@ ~0.6GB/s -40% drop

in BW

Create
Create first
first DEDUPE
snapshot
snapshot CHAOS

-20% drop
in BW
3PAR writing
to SSDs @
~2.0GB/s

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 26
 SNAPSHOTS ON THIN VOLUMES
COW SNAPSHOTS

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 27
 TPVV SNAPSHOTS - PHASE 1
PERFORMANCE WITH MOUNTED VOLUMES & SNAPSHOTS

Purpose:
• To establish a performance baseline when IO is running to both
volumes and their mounted snapshots
• Observe and document the performance impact when creating
additional snapshots for:
– Volumes only
– Snapshots only
– Both volumes & snapshots simultaneously

• Observe and document the ongoing performance impact of

snapshots
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 28
 TPVV SNAPSHOTS - PHASE 1
TEST ENVIRONMENT

• 16 x 80GB preconditioned TPVVs (CG1 and CG3)

• 16 R/W Snapshots – 1 per TPVV (CG2 and CG4)
• Steady State – 7 hours
– First 30 minutes: No snapshots
– Next 105 minutes: Create CG snapshots of CG1 & CG3 every 15 minutes
(volumes only)
– Next 105 minutes: Create CG snapshots of CG2 & CG4 every 15 minutes
(snapshot volumes only)
– Next 105 minutes: Create CG snapshots of CG1 – CG4 every 15 minutes
(all volumes & snapshots)
– Last 30 minutes: No snapshots
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 29
 PERFORMANCE DROPS & NEVER RECOVERS
BUT STILL OUTPERFOMS 2 X-BRICKS

-40%

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 30
 3PAR 8440 SNAPS – PERF. GOES TO ZERO

Scenario
1. Create and fill 16 volumes READ

2. Create a RW snapshot of each volume WRITE

TOTAL
3. Mount volumes and snapshots
4. Run steady workload
5. Create CGs
6. Schedule snapshots of CG1 and CG3
CG1: TPVV.0 – TPVV.7, CG2: TPVV.0.S1 – TPVV.7.S1
CG2: TPVV.0.S1 – TPVV.7.S1, CG2: TPVV.0.S1 – TPVV.7.S1
CG Snapshots (8 vols/CG)
1. Every 15 minutes
2. 30 second pause between CG1 and CG3 schedules
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 31
 3PAR 8440 SNAPS: USE THE RIGHT UI
SSMC Screenshot

When viewing
performance charts
generated by SSMC, we
see nice smoothed lines
which give the impression
of semi-consistent
performance.

IMC Screenshot (deprecated)

Performance charts generated by IMC,

paint a very different picture. The
actual performance profile is plagued
with continuous jitter and inconsistent
performance. Notice the intervals
where performance drops to zero!

Performance
drop to zero! READ

WRITE

TOTAL

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 32
 3PAR 8440 SNAPS: ZERO PERF FOR 7-8S READ

WRITE

TOTAL

Pre snapshot performance

~140 IOPs (R+W)
Create Snapshot
Post snapshot performance
Δ -15% to ~120K (R+W)

Write IOPs drop from ~70K to ZERO (~10s) ZERO IOPs (~7s) Recover to ~60K (~7s)

Read IOPs drop from ~70K to ZERO (~10s) ZERO IOPs (~8s) Recover to ~60K (~8s)

Scale 10 seconds
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 33
 HP 3PAR VIRTUAL COPIES
AGILITY: RESTORE & REFRESH

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 34
 SYNOPSIS
Conduct tests that demonstrate the capabilities and limitations of
HP 3PAR virtual copies (snapshots) related to iCDM workflows:
1. Refresh a snapshot
2. Promote a snapshot offline via the GUI
3. Promote a snapshot online via the CLI
4. Promote CG snapshots offline via the GUI
5. Promote CG snapshots online via the CLI
6. Write amplification on TDVVs when processing unique data vs
dedupable data
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 35
 “CATCH UP” WITH PRODUCTION
XtremIO: Its
PRODUCTION

P INSTANT !

BACKUP
COPY

S1.1 S0 (READ ONLY

3PAR:
NOT
TEST
ALLOWED
DEVELOPMENT
COPY S1.2 S1 COPY 1
(READ ONLY)
(READ/WRITE)

S2 TEST
COPY 2
QA COPY (READ ONLY)
(READ ONLY)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 36
 TEST 1 – SNAPSHOT REFRESH
HP 3PAR has limited capabilities on which snapshots
can be refreshed. A snapshot can only be refreshed if
it has no child objects (see array output on next
slide).
Vol Refresh not
supported

S1
Refresh
supported
S1.1

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 37
 TEST 1 – SNAPSHOT REFRESH
HP 3PAR has limitations on how a snapshot is refreshed and who
can refresh a snapshot. This involves executing the updatevv
command via the CLI only, and requires the creation of user
accounts and ACLs which grant access to specified volumes.
In the example below, the setuseracl command was executed to grant the 3paradm user the
necessary rights to run updatevv for volumes matching the specified naming pattern, R6.0*

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 38
 TEST 1 – SNAPSHOT REFRESH
The update command was run for each snapshot

• The only snapshot which was able to be refreshed

was R6.0.S1.144200.14422 since it was the lowest
entity in the snapshot chain

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 39
 SNAPSHOT REFRESH – KEY TAKEAWAYS
• Self service user operation via CLI only
• Requires creating users with edit rights -
Rights are granted to most operations. For example, creating, editing, and removing virtual volumes and
other objects. – excerpt from HP 3PAR Command Line Interface Administrator’s Manual
This creates unnecessary exposure and risk. When would
you ever want someone other than a storage admin to
have access to remove volumes?
• Requires ACLs mapping users to lists of volumes and
snapshots and is only manageable via CLI

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 40
 RESTORE PROD XtremIO: Its
INSTANT !
XtremIO: Its
PRODUCTION
INSTANT !
P 3PAR:
BACKUP TAKES
COPY

S1.1 S0 (READ ONLY) HOURS

TEST
DEVELOPMENT
COPY
S1 COPY 1
(READ ONLY)
(READ/WRITE)

S2 TEST
COPY 2
(READ ONLY)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 41
 REFRESH PROD
XtremIO: Its
PRODUCTION
INSTANT !
P
3PAR:
BACKUP
COPY TAKES
S1.1 S0 (READ ONLY
HOURS
TEST
DEVELOPMENT
COPY S1.2 S1 COPY 1
(READ ONLY)
(READ/WRITE)

S2 TEST
COPY 2
QA COPY
(READ ONLY)
(READ ONLY)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 42
 EXEC SUMMARY
RESTORE PRODUCTION VOLUMES

Activity Time Production Comments

taken Volume
Volume Restore Hours to Offline Only offline restore supported in
via GUI Days GUI. Must remove host mounts.
Volume Restore Minutes to Offline Must remove host mounts (else
via CLI – online Hours causes data corruption)
promote
CG restore via Minutes to Offline GUI does not support restoring
CLI – offline Hours CGs in any mode. Must use CLI
CG restore via Minutes to Offline
CLI – online Hours

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 43
 PROD RESTORE FROM SNAPS TAKES HOURS
A 400GB VOLUME RESTORE TAKE 2HRS:46MINS

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 44
 TEST 2 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA GUI (OFFLINE)

• Create snapshot of volume R6.4 (R6.160627201152)

• Write 80GB of random data, 10% of 800GB dataset,
via vdbench age script
wd=wd_age,sd=sd*,xfersize=(4k,9,8k,47,16k,21,32k,17,64k,6),rdpct=0,seekpct=100
rd=age_!SEQN,wd=wd_age,iorate=max,interval=1,elapsed=!TIME,maxdata=80G,forthreads=!THREAD

• Promote snapshot with high priority

• Wait for promote operation to complete

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 45
 TEST 2 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA GUI (OFFLINE)

HP 3PAR limits promoting snapshots via the GUI (online option to

covered later). A snapshot may only be promoted to a base
volume which has no exports. Attempting to promote a snapshot
to an exported volume results in an error.
In this example, an attempt was made to promote a snapshot of the volume R6.4 while the volume was
exported (mapped)

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 46
 TEST 2 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA GUI (OFFLINE)

For successful promotion using the GUI:

1. Unexport volume R6.4
2. Run promote with high priority to
reduce volume downtime
3. Wait for completion

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 47
 SNAPSHOT PROMOTE - KEY TAKEAWAYS
SINGLE SNAPSHOT VIA GUI (OFFLINE)

• When promoting a snapshot via the GUI, the volume

MUST BE UNEXPORTED AND OFFLINE!
• Snapshot promotion takes a substantial amount of
time to complete and results in extended outages.
~48 minutes is unacceptable to effectively
change 80GB of unique data for a single
volume!

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 48
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

• Create new 800GB volume – WinVol1

• Export to host, map to Windows VM as RDM, mount and format
• Write 400 x 1GB files (fs1_x.txt) using random file generator
• Create snapshot – WinVol1.S1
• Delete 100 files
• Promote WinVol1.S1 via CLI using online flag
• Verify files are restored

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 49
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

1. Create new TDVV – WinVol1

2. Export and mount as RDM to Windows VM
3. Format as new Windows NTFS volume 3PAR_RDM

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 50
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

Use File Generator utility to create

files on the newly created volume:
• 400 files
• 1GB
• Naming pattern FS1_X.txt
• Fill with random data

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 51
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

Verify that all 400 x

1GB files were created
successfully

Verify array volume

utilization matches
size of created files
and is equal to 400GB

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 52
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

Create snapshot of WinVol1

– WinVol1.S1

Delete 100 files –

FS1.301.txt – FS1_400.txt

Use Case – User

initiated accidental
file/folder deletion

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 53
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

• Run promotesv with –online flag to restore WinVol1

from the point in time WinVol1.S1 without removing
volume exports
• Wait for operation to complete
• Verify data is restored

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 54
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

promotesv –online WinVol.S1

Monitor progress

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 55
 TEST 3 – PROMOTE SNAPSHOT
SINGLE SNAPSHOT VIA CLI (ONLINE)

Verify that all files

exist after the
promote operation

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 56
 PROMOTE SNAPSHOT – KEY TAKEAWAYS
ONLINE PROMOTE VIA CLI

• Online refers to the array status of the volume ONLY and ONLY
means that the promote operation can be executed for volumes
that have exported VLUNs
• During online promote the host file system must be un-
mounted or offline. Failure to adhere to this action will result in
data corruption and will require a second restore of the data.
• Results in extended downtime since volume is offline for the
duration of the promote operation

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 57
 TEST 4 – PROMOTE CG SNAPSHOTS
OFFLINE PROMOTE VIA GUI

• Create CG snapshots of volumes R6.0, R6.1, R6.2

and R6.3 (R6.0.CG.1, R6.1.CG.1, R6.2.CG.1,
R6.3.CG.1)
• Write 320GB of random data, 10% of 3200GB
dataset, via vdbench age script
wd=wd_age,sd=sd*,xfersize=(4k,9,8k,47,16k,21,32k,17,64k,6),rdpct=0,seekpct=100
rd=age_!SEQN,wd=wd_age,iorate=max,interval=1,elapsed=!TIME,maxdata=320G,forthreads=!THREAD

• Promote CG snapshots offline with high priority

• Wait for promote operation to complete
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 58
 TEST 4 – PROMOTE CG SNAPSHOTS
OFFLINE PROMOTE VIA GUI

• Run shovv *CG* to list CG snaps

• Run showvvset CG1.Snaps to list member volumes

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 59
 TEST 4 – PROMOTE CG SNAPSHOTS
OFFLINE PROMOTE VIA GUI

Run promotesv with high priority for the volumeset CG1.Snaps

Run showtask to monitor progress

Task completes in ~28:09

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 60
 PROMOTE CG SNAPSHOTS – KEY TAKEAWAYS
OFFLINE PROMOTE VIA GUI

• HP 3PAR provides no option for promoting CG snapshots via the

SSMC GUI
This was previously supported in the Windows based 3PAR Management Console, however this tool is
deprecated and will not be covered

• CG snapshots may be promoted to base volumes which have

no exported VLUNs
• Promoting a set of CG snapshots offline with a 10% change
rate (320GB) required ~28 minutes to complete which equates
to ~195MB/s with no other activity on the array
• Volumes are offline, un-exported and inaccessible for the
duration of the promote operation!
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 61
 TEST 5 – PROMOTE CG SNAPSHOTS
ONLINE PROMOTE VIA CLI

• Create CG snapshots of volumes R6.4, R6.5, R6.6

and R6.7 (R6.4.CG.1, R6.5.CG.1, R6.6.CG.1,
R6.7.CG.1)
• Write 320GB of random data, 10% of 3200GB
dataset, via vdbench age script
wd=wd_age,sd=sd*,xfersize=(4k,9,8k,47,16k,21,32k,17,64k,6),rdpct=0,seekpct=100
rd=age_!SEQN,wd=wd_age,iorate=max,interval=1,elapsed=!TIME,maxdata=320G,forthreads=!THREAD

• Promote CG snapshots via CLI with –online flag

• Wait for promote operation to complete
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 62
 TEST 5 – PROMOTE CG SNAPSHOTS
ONLINE PROMOTE VIA CLI

• Run shovv *CG.2 to list CG snaps

• Run showvvset CG2.Snaps to list member volumes

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 63
 TEST 5 – PROMOTE CG SNAPSHOTS
ONLINE PROMOTE VIA CLI

Run promotesv with –online flag for the volumeset CG2.Snaps

Run showtask to monitor progress

Task completes in ~22:14

EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE

Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 64
 PROMOTE CG SNAPSHOTS – KEY TAKEAWAYS
ONLINE PROMOTE VIA CLI

• Promoting a CG snapshot to an exported volume

requires the use of the –online flag
• Online promotion will ONLY run with the default
medium priority
• Promoting a set of CG snapshots online with a 10%
change rate (320GB) required ~22 minutes to
complete which equates to ~248MB/s with no other
activity on the array
EMC CONFIDENTIAL: FOR INTERNAL USE ONLY. DO NOT DISTRIBUTE
Internal©Use - Confidential
Copyright 2016 EMC Corporation. All rights reserved. 65