Post on 09-Mar-2018
IBM Scale-out File Services (SoFS)
and IBM DCS9900
Arik Blum
IBM storage consultant
052-2554726, arieb@il.ibm.com
פתרון אחסון ייחודי עבור ענן המחשוב
A shift to unstructured files
Sources: IDC worldwide enterprise disk in Exabytes from “Changing Enterprise Data Profile”, December 2007
Storage capacity is doubling every 18 months…
� Structured data, such as databases for
transactional workloads, growing at 32%
� Unstructured data, such as user files, medical images, web and rich media content, growing at 63%
� Replicated data, including backup, archive, growing at 49%
3
The Problems with NAS filers today
(We have it : IBM N Series Family)
“I loved my first filer. It
was so easy to
manage. When we installed our 20th,
I started to hate them.”
Classic Filers
• Current NAS solutions do not scale
• Customers have to add box after box and
manage them individually
• Difficult to apply policies across independent
data islands
• Some applications require parallel access and
high data rates
• Backup windows are a big issue and get worse
as the amount of data increases
� All files online, but more than 80% haven't been
accessed during the last 6 months
4
Using a Global Name Space• Many attempts to solve this with a Global Name
Space (IBM Virtual File Manager, Brocade StorageX, ONTAP GX)
– Each individual file is pinned to a single NAS filer
– Maximum single file performance is equal to the performance of the individual hosting filer
– Bottlenecks on individual directory branches
– Islands regarding disks, backup, etc.
/
/sales
/finance
/web
• Each filer is individually accessed
• “Finding the file”becomes…– Finding the server that
has the file
… then …
– finding the file on that server
/sales
/finance
/web
5
Scale-out File Services is Different• SoFS combines a true clustered
file system with a global namespace
• All nodes serve all filesAll nodes serve all filesAll nodes serve all filesAll nodes serve all files
• Maximum single file performance is equal to the aggregated performance of the cluster
• Deployed centrally, managed centrally, backed up centrally and grown centrally. No islands!
• No bottlenecks on single directory branches
/
/sales
/finance
/web
SoFS
6
SoFS
IP LayerIP Layer
Classic Filers
DFS AD
The SoFS Story: Performance & Scalability
/
/sales
/finance
/web
/
/sales
/finance
/web
parallel data flowall nodes serve all files
each file is pinnedto a single filer
From a client perspective SoFS is a single file space
7ט"תשס/סיון/ו"כ
Customer Location
Site A
SoFS
Customer Location
Site B
SoFS
CrossClusterMount
Wide
Area
Network
SoFS with cross-cluster mount
Storage
Storage
8ט"תשס/סיון/ו"כ
Customer Location
Site A
SoFS
Customer Location
Site B
Wide
Area
Network
SoFS with synchronous replication across sites
Storage
Storage
9ט"תשס/סיון/ו"כ
Customer Location
Site A
SoFS
A
Customer Location
Site B
Wide
Area
Network
SoFS with asynchronous replication across sites
SoFS
B
Only changed blocks will be transfered
Storage
Storage
10ט"תשס/סיון/ו"כ
SoFS component view
• CIFS and NFS
• CTDB – Clustered Trivial database Daemon, Controls the cluster
• General Parallel File System (GPFS) -IBM’s high end clustered file system
• Management, administration and monitoring software
• Snapshots
CIFSCIFS
IBM
GPFS
IBM
GPFS
Enterprise LinuxEnterprise Linux
IBM ServerIBM Server
IBM Disk
-DS Family
DCS9900
IBM Disk
-DS Family
DCS9900
IBM TapeIBM Tape
ReportingReporting
MonitoringMonitoring
ProvisioningProvisioning
NFSNFS HTTPSHTTPS FTPFTP
CTDBCTDB
HSM - ArchivingHSM - Archiving
TSM – Backup & RestoreTSM – Backup & Restore
IBM Director - Hw MgtIBM Director - Hw Mgt
iSCSI*iSCSI*
IBM SVCIBM SVC
Other Disk
11
The SoFS Story:
Information Lifecycle Management
• Capacity managed centrally
• Average utilization >80%
• Policy driven
- File movement – between
storage tiers, the least active
files can be migrated to tape
- File expiration – delete files
after they are no longer
needed
SoFS
Just buy the capacity you
really need
DCS9900 or any other
Storage system
IBM DCS9900
The best Price/Performance Storage system
for large Environments (100TB and up) :
NAS And/Or SAN support
System Storage DCS9900 OverviewIBM’s Storage Solution for Unstructured I/O and Backup
Fast Streaming I/O
• Up to 5.7 GB/s - Writes as fast as reads
Dense Packaging
• up to 600 TB/rack with 1TB SATA Drives
• More data per single system – up to 1.2 PB
Extreme Reliability
• Access to all data is maintained independent of disk,
enclosure or controller failures
13
2TB Disk drives will be supported soon.
Build your own PetaByte infrastructure with DCS9900 for many services
FC SAN Clients
NFS/CIFS NAS Clients
Virtual Servers
NetApp ClientsBackup Servers
Backup Servers
SAN Virtualization
(IBM SVC)
DeDup Gateway
Multi-PB Scalability
�1.2PB per system
�6GB/s per system
NetApp Filer
Compare capacity scaling approaches:
� Initial Purchase: 200TB
CompetitorDDN
•More Switches
•More Power
•More Cooling
•More Floor Space
•More Management
•More Complexity
�Capacity Grows to: 400TB� Capacity Grows to: 600TB
Simply Add any Drives
DCS9900 Price is very low !
DCS9900
System Storage DCS9900
6/18/2009DCS9900 Technical Presentation16
Field Programmable
Gate Array (FPGA)
• Hardware enabled RAID 6, which protects data in the event of double disk failure in the same redundancy group
• Support for SATA and/or SAS Drives and/or SSD• Dual controller with 5GB of RAID protected cache
(2.5GB cache per controller)• Eight FC 8Gbit or Infiniband 4X DDR host ports
(4 ports per controller)• 20 * SAS 4-lane (3Gb/s) internal connections (10 per controller)• Full duplex host transfer operation, sustained performance up to 5.7
GB/s in both reads/writes• No performance penalties in degraded mode operation,
very fast rebuild rate• SNMP, GUI, Telnet and API support
IBM System Storage DCS9900 Primary Features
6/18/2009DCS9900 Technical Presentation17
• Large capacities- 1.2PB per system (2.4PB with 2TB Disks)- 60TB in 4U
• Sleep Mode per Tier basis!
• SAS / SATA / SSD Intermix drives
• SAN or NAS environments (with NAS Gateway)
• Availability and reliability
�Hardware based RAID 6 : 8+2…with no performance penalty
�Parity computed on every Read –no SATA silent corruption errors
�Data remains visible even in the case of a disk, enclosure or controller failure
6/18/2009DCS9900 Technical Presentation18
IBM System Storage DCS9900 Primary Features
Argonne Intrepid System• IBM BlueGene/P Architecture
– Sustained 447 Tflop/s
– #3 on the June Top500 List
– 160K Cores
• IBM Provided DCS9900 Systems
– 17 x DCS9900s
– 8PB of Capacity
– 93GB/s of Peak Performance
•5.7GB/s per 9900 with GPFS/SOFS
Enabling Extreme Science
? שאלות -תודה ר בה על ה הקשבה
IBM SoFS + IBM DCS9900 = Perfect Solution for Cloud Computing