4. What are storage spaces ?
Windows 2012 introduces new storage options.
Storage Spaces are a cost-effective storage
virtualization platform that provides the
capabilities required for business-critical
storage.
4
5. Physical layer of storage spaces
Direct-Attached Shared – Availability & Scale
SATA or SAS Shared SAS
OR
5
6. Storage Pools vs Spaces
Storage Pools : Units of Aggregation, Administration & Isolation
Storage Spaces : Resiliency, Provisioning & Performance
Windows Storage Storage Storage
virtualized Spaces Spaces Spaces
storage
Storage Pool Storage Pool Storage Pool
(Shared) SAS or SATA or USB
Physical
storage
6
7. Redundant storage
• Mirrored and Parity Spaces with integrated journaling
• Rapid Recovery per pool hot spare support
• Application driven intelligent error correction
Resilient Spaces Mirrored Parity Mirrored Mirrored Parity Mirrored Mirrored Parity
contained within a Pool Space Space Space Space Space Space Space Space
Industry standard
physical storage
Hot Spare
7
8. Cost effective : ONE MILLION IOPS
Dell® Server
Configuration
5x LSI ® HBAs Components Quantity
Dell® PowerEdge® T710 1
2x Intel® Xeon® X5570, 48GB
memory, 5x LSI® 9205-8e SAS
HBA
StarTech® 4x 2.5” SATA 10
Enclosure Bay
Intel® X25-M 160GB SSD 40
JBOD
Enclosure
40x Intel®
SSDs
8
9. Cost effective : 80 TB highly available storage
Configuration
Node #1
Finance Components Quantity
[SFSDemo3] VM Supermicro® 2U server 2
2x Intel® Xeon® X5570, 12GB
memory, LSI 9200-8e SAS
VHDStore HBA
Node #2
[SFSDemo2] VM
Supermicro® 4U JBOD 1
Seagate® 2TB 3.5” 42
.VMLibrary near-line SAS HDDs
Supermicro®
4U JBOD
9
11. • CSV is a clustered file system in Windows Server 2012
- Enables all servers in a Cluster to access a common volume
- Provides a layer of abstraction above NTFS
• Provides complete abstraction with respect to which
nodes actually own a LUN
• Applications can failover without
requiring drive ownership changes
- No dismounting and remounting of volumes
- FAST failover within a cluster
11
12. Moving forward …
vNext+
Windows • Shared Storage
Server 2012 model for
everyone?
• More workloads
Win2008 R2 starting with file
services
• Support for
first workload
with Hyper-V
12
13. CSV 2.0 motivations and targets
• More workloads and applications
- File Server in addition to Hyper-V
• Improved Backup / Restore
• Improved performance
- Block Level I/O redirection
- Direct I/O for more scenarios
- SMB direct (SMB over RDMA)
• Support for Storage Spaces
• Multi-subnet support
13
14. I/O fault tolerance
• CSV transparently handles node, network, and HBA failures
• CSV proxy file system (CSVFS) virtualizes file objects
When a node failure occurs:
- Cluster service notifies remaining nodes to re-set their SMB
connections for all volumes which were owned by the failing node
- Filter driver holds all I/O requests for “orphaned” CSV volumes until
the volumes are remounted on other nodes
- Filter driver reopens files and remaps “virtual” handles to real ones
Failover is transparent to applications!
14
16. New file system check & repair: Execution Flow
Detect Corruption Online Online Self-
Verification Healing
• NTFS detects a • NTFS will validate • If not transient,
perceived anomaly whether issue is NTFS will attempt to
in file system transient or genuine rapidly self-heal
metadata • Volume remains • Volume remains
online online
Identify & Log Precise & Rapid Correction
• NTFS will identify and log • At the next maintenance window,
corrective actions for issues that volume can be taken offline and
logged corruptions are corrected in
could not be self-healed seconds
• Volume remains online • With CSV I/O is transparently paused
• Administrator will be notified for rapid correction and subsequently
automatically resumed.
16
17. From Hours to Seconds …
• Disk scanning process <3 sec of downtime
separated from repair process 400
350
300
•
Minutes
250
Online volume scanning 200
- Volume offline only to repair 150 Windows
100 Server
- Based on number of errors to fix 2008 R2
50
rather than size of volume
0
100 200 300
• Zero offline time with CSV
MillionMillionMillion
Files Files Files
17
18. How it works on Cluster Shared Volumes
• Chkdsk /Spotfix can seamlessly run on CSV volumes
- Normally Spotfix is an Offline operation
• Spotfix processing on a CSV volume:
- Every minute IsAlive health check runs, cluster checks if Chkdsk /Spotfix is required
- Clustering enumerates NTFS $corrupt to identify corrupt files
- CSV namespace is paused to CSVFS file system
- Underlying NTFS volume is dismounted
- Spotfix is run against corrupted files for a maximum of 15 seconds
- If all $corrupt records have not been processed in the 15 seconds, the health check
will wait 3 minutes before continuing the remaining spotfixes
- If corruption is too large, spotfix is not run and marked to run at next Physical Disk
online
18