1 © Hitachi Data Systems Corporation 2011. All Rights Reserved.1

WHY HITACHI VIRTUAL STORAGE PLATFORM DOES SO WELL IN A MAINFRAME ENVIRONMENTHOW FAST CAN A VSP GO?

RON HAWKINS, MANAGER, TECH OPS PERFORMANCENOV. 2, 2011

2

WEBTECH EDUCATIONAL SERIES

Why Hitachi Virtual Storage Platform (VSP) Does So Well in a Mainframe Environment

Hitachi VSP is a new paradigm in enterprise array performance. In this session we will discuss how the architecture of VSP enhances its box-wide performance. The results of performance testing with synthetic host I/O generators and the PAI/O driver will also be presented. 

Attend this WebTech to learn how to:

Improve performance in your environment with VSP

Affect performance in mainframe environments with different RAID architectures

Optimize functionality with wide striping enabled by Hitachi Dynamic Provisioning

3

UPCOMING WEBTECHS

November and December

‒ Increase Your IT Agility and Cost-efficiency with HDS Cloud Solutions, Nov. 9, 9 a.m. PT, 12 p.m. ET

‒ Best Practices for Upgrading to Hitachi Device Manager v7, Nov. 16, 9 a.m. PT, 12 p.m. ET

‒ Hitachi Clinical Repository, Dec. 7, 9 a.m. PT, 12 p.m. ET

Please check www.hds.com/webtech for

‒ Link to the recording, presentation and Q&A (available next week)

‒ Schedule and registration for upcoming WebTech sessions

4

WE ARE TOP GUN

WHAT WE WILL COVER

Unified microprocessor

FICON front-end director

The numbers

Hitachi Dynamic Provisioning for Mainframe

Hitachi Virtual Storage Platform

5

UNIFIED MICROPROCESSOR

THE HEART OF THINGS

6

ANOTHER VIEW OF THE HITACHI VIRTUAL STORAGE PLATFORM

7

MICROPROCESSORS HAVE MOVED OFF THE FRONT-END AND BACK-END DIRECTORS

8

Shared Memory

HITACHI UNIVERSAL STORAGE PLATFORM® V ESTRANGED MICROPROCESSOR

FEDBED

9

Shared Memory

HITACHI UNIVERSAL STORAGE PLATFORM® V ESTRANGED MICROPROCESSOR

FEDBED

10

Shared Memory

HITACHI UNIVERSAL STORAGE PLATFORM® V ESTRANGED MICROPROCESSOR

FEDBED

11

UNIFIED MICROPROCESSOR

XEON Quad Core CPU

4GB Local RAM

Package Memory

Shared Memory

Cache

Write Through

Page In

12

FICON FRONT-END DIRECTOR

AT THE EDGE

13

FICON FRONT-END DIRECTOR

Hitachi transport processor‒ Initial FICON protocol

‒ Open exchanges

MHUB + ASIC‒ Data accelerator circuit

‒ A programmable application-specific integrated circuit

‒ Route command to virtual storage director (VSD)

‒ Direct memory access (DMA) engine

FICON 16-PORT FEATURE

14

HITACHI VIRTUAL STORAGE PLATFORM 8-PORT FICON BOARD

HITACHI TRANSPORT PROCESSOR

Accepts initial channel command request from host

Multi-protocol‒ FICON and high-

performance FICON (zHPF)

Establishes open exchanges (480)‒ Shared on demand by

adjacent ports

This graphic is getting hard to read in the red. Can the type be reversed? -th

Not Really as it is a graphic lifted from another source, we don’ t have the original graphic.

15

OPEN EXCHANGES

HITACHI TRANSPORT PROCESSOR

64 for each host channel‒ 2:1 fan-in is 128‒ Adjacent ports 256

‒ 4:1 fan-in is 256‒ Adjacent ports 512

Open Exchange (OE) exhaustion increases Command Response (CMR) time

‒ Microprocessor busy (>80%)‒ Low cache hits‒ TCz synchronous

CMR Time

16

HITACHI VIRTUAL STORAGE PLATFORM 8-PORT FICON BOARD

MHUB + ASIC

Two chips working together‒ Processor plus

programmable ASIC

‒ Commands to/from virtual storage director

‒ Using LDEV mapping tables

DMA engine‒ Read and write directly to

cache

17

THE NUMBERS

WE ARE TOP GUN

18

WE ARE TOP GUN

All tests are a fully popped VSP (unless stated otherwise)‒ 2 chassis

‒ 8 virtual storage directors

‒ 4 back-end directors

‒ 16 front-end directors (128 channels)

‒ 2048 HDD (10K 300GB)

‒ 512GB cache

HITACHI UNIVERSAL STORAGE PLATFORM® V DUAL-CHASSIS CONFIGURATION

19

WE ARE TOP GUN

Zero

Loca

lity W

rite

Zero

Loca

lity R

ead

Front

End

Writ

e

Front

End

Rea

d0

200000

400000

600000

800000

1000000

1200000

FNP Driver Results

FNP Driver Results

HITACHI UNIVERSAL STORAGE PLATFORM® V AS AN I/O DRIVER

20

PAI/O TESTING

THE LAB TO THE REAL WORLD

Extended format VSAM‒ 4KB and 26KB

High overhead‒ Two CCW per block

Growing format for DB2

Datasets >4GB

What customers will really get

21

WE ARE TOP GUN

CONSTRAINED BY CPU CAPACITY

22

WE ARE TOP GUN

BENEFIT OF ZHPF

23

WE ARE TOP GUN

CONSTRAINED BY VSD 80-85% BUSY

24

WE ARE TOP GUN

BENEFIT OF ZHPF

25

WE ARE TOP GUN

RANDOM WRITE MISS HIT 70% WP

26

WE ARE TOP GUN

CONSTRAINED BY VSD 80% BUSY

27

WE ARE TOP GUN

CONSTRAINED BY VSD 80% BUSY

28

WE ARE TOP GUN

HIGH-PERFORMANCE BED FEATURE

0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,0000

2

4

6

8

10

12

14

16

18

20

PAI/O H44K Read

100% Cache Miss

High Performance (2 BED) Standard Performance (1 BED)

IO Operations per Second

Res

po

nse

Tim

e M

S

29

WE ARE TOP GUN

HIGH-PERFORMANCE BED FEATURE

1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,0000

10

20

30

40

50

60

70

80

90

100

PAI/O G0 - Sequential Read27K Chain Length 30 (1 Cyl)

High Performance (2 Bed) Standard Performance (1 BED)

MB per second

Res

po

nse

Tim

e M

S

30

HITACHI DYNAMIC PROVISIONING FOR MAINFRAME

SPREAD THE LOAD

31

HITACHI DYNAMIC PROVISIONING FOR MAINFRAME (HDPM)

Primary advantages of HDPM‒ Wide striping eliminates skewed I/O to disk‒ Greater throughput for cache miss I/O

‒ Configuration flexibility‒ Custom volume sizes made easy

‒ Dynamic volume expansion

‒ IBM® FlashCopy® space-efficient pools

Performance results focus on skewed I/O

HDPM SPREADS THE LOAD

32

PAI/O DRIVER – F1 (SYMMETRICAL)

0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 180,0000.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0.20

PAI/O Driver - F14K Cache Hit Write Chain(1)

RD662NN RD662HF RD662HN RD571NN RD571HF RD571HNIO Per Second

Res

po

nse

Tim

e in

MS

HDS Confidential

33

SKEW OR ASYMMETRICAL WORKLOADPAI/O DRIVER C* SERIES

HDS Confidential

Figure taken from PAI/O Driver User Guide

34

LET HITACHI DYNAMIC PROVISIONING FOR MAINFRAME (HDPM) SHARE THE LOAD

0 2,000 4,000 6,000 8,000 10,000 12,000 14,0000.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

PAI/O Driver - C64K Skewed Series Chain(1)

70% Read / 30% Write

RD662NN RD662HF RD662HN

I/O Per Second

Res

po

nse

Tim

e in

MS

HDS Confidential

HDPMBenefit Zone

RAID 6 6D+2P PAI/O DRIVER – C6 (SKEWED OR ASYMMETRICAL)

35

FINAL THOUGHT

“I feel the need–the need for speed!”

―Top Gun

3636

QUESTIONS AND DISCUSSION

37

UPCOMING WEBTECHS

November and December

‒ Increase Your IT Agility and Cost-efficiency with HDS Cloud Solutions, Nov. 9, 9 a.m. PT, 12 p.m. ET

‒ Best Practices for Upgrading to Hitachi Device Manager v7, Nov. 16, 9 a.m. PT, 12 p.m. ET

‒ Hitachi Clinical Repository, Dec. 7, 9 a.m. PT, 12 p.m. ET

Please check www.hds.com/webtech for

‒ Link to the recording, presentation and Q&A (available next week)

‒ Schedule and registration for upcoming WebTech sessions

3838

THANK YOU