AN ALTERNATIVE TO ALL- FLASH ARRAYS: PREDICTIVE STORAGE CACHING - - PowerPoint PPT Presentation

AN ALTERNATIVE TO ALL- FLASH ARRAYS: PREDICTIVE STORAGE CACHING

THE EASIEST WAY TO INCREASE PERFORMANCE AND LOWER STORAGE COSTS

Bruce Kornfeld, Chief Marketing Officer, StorMagic Luke Pruen, Technical Services Director, StorMagic Peter Smith, System Admin, Harris Corporation

STORMAGIC SVSAN: BRIEF OVERVIEW

SvSAN turns the internal disk, SSD and memory of two or more servers into highly available shared storage.

2

ALL-FLASH ARRAYS ARE TEMPTING

• SSDs vs. HDD
• 10x-20x more performance
• 10x more expensive
• Lots of hype from flash array

vendors

• Not all workloads need all-flash
• A balanced approach with

advanced caching algorithms can meet your needs

3

Source: Statistica

SSDs are still 10x more costly than HDD. SSD: $0.50 - $1.00 per GB HDD: $0.05 - $0.10 per GB

PREDICTIVE STORAGE CACHING

Over 400% performance improvement with a patent-pending method for predicting when data will become ‘hot’

4

AUTOMATED

Two caches – system memory and SSD Patent-pending predictive algorithms Automation tracks ‘hot’ data and places in the right cache

BUILT FOR PERFORMANCE

Lower latency – less waiting for spinning disks Data pinning Solves the virtual server ‘I/O blender effect’

COST EFFECTIVE

Use fewer and less expensive disk drives (7.2K) Lower CAPEX – server memory is inexpensive Lower OPEX – power, cooling and maintenance

AN ALTERNATIVE TO ALL- FLASH ARRAYS: PREDICTIVE STORAGE CACHING

THE EASIEST WAY TO INCREASE PERFORMANCE AND LOWER STORAGE COSTS

Luke Pruen, Technical Services Director, StorMagic

TODAY’S STORAGE OPTIONS

• The performance gap

between CPU and storage

• Disk only
• High capacity, low

performance workloads

• All-flash
• Performance at any cost

workloads

• Hybrid
• Most workloads fit here

6

THE IMPORTANCE OF CACHING

• Virtualized environments suffer from the ‘I/O

blender effect’

• Working sets of data change over time
• Advanced caching can solve both problems

7

WRITE CACHING

Stage 1

• All new data written to SSD
• Data marked ‘dirty’ – not committed to HDDs

Stage 2

• Write operation is acknowledged immediately to the server

Stage 3

• ‘Dirty’ data is reordered and grouped based on disk locality
• Data de-staged and written to HDD, sequentially as possible

Stage 4

• SSD cache notified when data successfully written to HDD
• Cached data marked ‘clean’, remains in cache until space

needed

8

WRITE CACHING

Stage 1

• All new data written to SSD
• Data marked ‘dirty’ – not committed to HDDs

Stage 2

• Write operation is acknowledged immediately to the server

Stage 3

• ‘Dirty’ data is reordered and grouped based on disk locality
• Data de-staged and written to HDD, sequentially as possible

Stage 4

• SSD cache notified when data successfully written to HDD
• Cached data marked ‘clean’, remains in cache until space

needed

8

READ AHEAD & DATA PINNING

Read ahead mode

• I/O blender effect aware!
• Identifies sequential interleaved I/Os
• Detects sequential read streams
• Pre-fetches data into memory

9

Data pinning mode

• Pin specific data/workloads in memory
• Delivers most efficient read performance
• DBs, VDI, frequently repeated operations
• Manage multiple pin groups

PREDICTIVE CACHING

Predictive approach

• Tracker module has 7 levels
• All read I/Os monitored and

analyzed

• Most frequently used – higher levels
• Cache placement based on levels

Flexible storage options

• RAM: Most frequently accessed

data

• SSD/flash: Next most frequently

accessed data

• HDD: Infrequently accessed data –

‘cold’ data

10

Advanced, patent-pending method for predicting ‘hot’ data, placing it on the best storage cache available

Sizing

• Assign cache sizes to meet

requirements

• Grow caches as working sets

change

• Use any combination of memory,

SSD/flash and disk Play to the strengths

• Play to the strengths of all mediums
• Memory highest IOPS
• SSD/flash magnetic drives providing

low price per GB

REAL CUSTOMER DATA (BEFORE CACHING)

Workload

• 12 Virtual Machines
• 78 applications

11

Performance characteristics of their existing workloads – measured by StorMagic

500 1000 1500 2000 2500 3000 3500 18:39 21:25 00:11 02:57 05:43 08:29 11:15 14:01 16:47 19:33 22:19 01:05 03:51 06:37 09:23 12:09 14:55 17:41 20:27 23:13 01:59 04:45 07:31 10:17 13:03 15:49 18:35 IOPs Time of Day (UTC)

Throughput IOPs

Read Write 0.001 0.01 0.1 1 10 100 1000 10000 21 MB 110 GB 221 GB 332 GB 443 GB 554 GB 665 GB 776 GB 887 GB 998 GB 1.1 TB 1.2 TB 1.3 TB 1.4 TB 1.5 TB 1.6 TB 1.7 TB 1.8 TB 2.0 TB 2.1 TB 2.2 TB 2.3 TB 2.4 TB 2.5 TB 2.6 TB Number of accesses (logarithmic scale) Thousands

Locality of access

Read Write

5,000,000 10,000,000 15,000,000 20,000,000 25,000,000 1 KB 2 KB 4 KB 8 KB 16 KB 32 KB 64 KB 128 KB 256 KB 512 KB 1 MB 2 MB 4 MB Hit Count Block Size

Block Size Distribution

Writes Read

Read Write Read/Write % 77% 23% Sequential % 49% 39% Average Per Day 991 GB 294 GB Average Block Size 58 KB 54 KB Average IOPS 212 138

REAL CUSTOMER DATA (WITH CACHING)

HDD Only (no caching)

• 1 x RAID5 = 3 x 1.2TB 10K SAS

disks HDD & Memory (with caching)

• 12GB of memory per host for

caching

• 1 x RAID5 = 3 x 1.2TB 10K SAS

disks HDD & SSD (with caching)

• 1 x 200GB Samsung SSD
• 1 x RAID5 = 3 x 1.2TB 10K SAS

disks HDD, SSD & Memory (with caching)

• 12GB of memory per host for

caching

• 1 x 200GB Samsung SSD
• 1 x RAID5 = 3 x 1.2TB 10K SAS

disks

12

Impact of Predictive Storage Caching to increase performance

2400 12341 6165 13061 2000 4000 6000 8000 10000 12000 14000 HDD Only HDD & Memory HDD & SSD HDD, SSD & Memory

Total IOPS

21.86 4.94 9.37 5.21 5 10 15 20 25 HDD Only HDD & Memory HDD & SSD HDD, SSD & Memory

AVG Latency (ms)

HDD Only HDD & Memory HDD & SSD HDD, SSD & Memory Total IOPS 2400 12341 6165 13061 AVG Latency (ms) 21.86 4.94 9.37 5.21

With Caching

SYNTHETIC ‘HERO’ NUMBERS

HDD Config

• 1 x RAID5 = 3 x 1.2TB 10K SAS disks

Tiered Config

• 1 x RAID5 = 3 x 1.2TB 10K SAS disks
• 1 x 200GB Samsung SSD
• 32GB of memory per host for caching

13

High performance achievement in a controlled lab environment

4263 2004 1832 7729 8728 207117 38196 79738 85803 15651 4KB 100% RND RD 4KB 100% RND WR 4KB 70/30 RD/WR 80% RND 32KB 100% SEQ RD 32KB 100% SEQ WR

IOPS

HDD Tiered 60.04 127.17 147.36 33.62 29.33 1.24 6.70 3.21 2.98 16.36 4KB 100% RND RD 4KB 100% RND WR 4KB 70/30 RD/WR 80% RND 32KB 100% SEQ RD 32KB 100% SEQ WR

AVG Latency (ms)

HDD Tiered

Test

• vSphere 6.5
• 2 x Windows VMs
• IOmeter
• VMDK on VMFS

Total IOPS Average Response (ms) Workload HDD Tiered HDD Tiered 4KB 100% Random Read 4263 207117 60.04 1.24 4KB 100% Random Write 2004 38196 127.17 6.70 4KB 70/30 Read Write 80% Random 1832 79738 147.36 3.21 32KB 100% Sequential Read 7729 85803 33.62 2.98 32KB 100% Sequential Write 8728 15651 29.33 16.36

SUMMARY: PREDICTIVE STORAGE CACHING

Increased performance – without the high cost of all-flash arrays Flexible configurations – we’ll help ‘right-size’ the server hardware Lower CAPEX – use slower drives for capacity, low-cost system memory and some SSD for performance Lower OPEX – less power, cooling and maintenance

14

Q&A – HARRIS CORPORATION

Peter Smith

Systems Administrator Network, Platforms and Security Group Harris Corporation

15

NEXT STEPS

Further reading: An overview of SvSAN – http://stormagic.com/svsan/ SvSAN data sheet – http://stormagic.com/svsan-data-sheet/ Predictive Storage Caching – http://stormagic.com/svsan/predictive-storage-caching/

16

Download your free trial of SvSAN

stormagic.com/trial

SvSAN Product Information

Product Options SvSAN license 2, 6, 12 and unlimited TBs License entitlement 2 mirrored servers Maintenance and support Platinum - 24x7 / Gold - 9x5

For further information, please contact: sales@stormagic.com