Transcript Latest trends and technologies in Storage Networking

Latest trends and technologies
in Storage Networking
By: Gururaja Nittur
Advisor: Dr. Chung E Wang
Second Reader: Dr. Du Zhang
Scope of the Project
 Study the new technologies in the storage
networking arena
–
–
–
–
Fibre channel protocol
NAS, SAN and iSCSI
Storage Virtualization
High Availability
 Demonstrate high availability by writing a
DMP(Dynamic Multi Pathing) driver for Solaris
What is Storage Networking?
“The practice of creating, installing,
administering, or using networks whose
primary purpose is the transfer of data
between computer systems and storage
elements and among storage elements”
Why Storage Networks?
“The total amount of data being stored doubles every year. Also, more
than 90% of companies today would fail to survive a catastrophic
data loss. Businesses face a mission-critical need to protect, access,
and manage their ever-growing volume of storage assets”
 Explosive growth of business data
 Internet and Multimedia
 High Availability
 Management complexity
Why Storage Networks? (Contd..)
Fibre Channel (FC)
 A serial, high-speed data transfer technology
 Open standard, defined by ANSI and OSI
 Data rate upto 100 MB/sec. (200 MB/sec. fullduplex)
 Supports most important higher protocols like IP,
ATM, SCSI etc.
 Does not have its own command set, but
facilitates data transfers between individual FC
devices.
Parallel Transmission
 Set of data signals are sent simultaneously
through 8, 16 or even more wires.
 Problems with parallel transmission
– data sent simultaneously over all the wires have to be
received simultaneously as well
– Total time = t+dt
• ‘t’ - time taken for the signals to reach receiver
• ‘dt’ – additional delay due to hardware inconsistencies
• ‘dt’ increases with cable length causing lesser frequency
 Example - SCSI
• Bus length limitations
• Max bus speed is limited (~40 MB/sec in Ultra SCSI)
• Limited device count
Serial Transmission
 Serial transmission uses single cable
– All signals are delayed the same and arrive at
the receiver in the same order in which they
were sent.
– Higher bus length
 Examples
– SSA (Serial Storage Architecture by IBM)
– Fibre Channel
Current Technology
DAS using SCSI
Emerging Technologies
Network Attached Storage (NAS)
Storage Area Networks (SAN)
Storage over IP (iSCSI)
Network Attached Storage
Network Attached Storage
 Storage device will have a built-in
network interface
 NAS unit can be plugged directly into the
network to allow quick and easy access
 Standard network protocols such as CIFS
and NFS can be used to share data
Network Attached Storage
 NAS engine is usually SCSI for low-end
systems for cost reasons and Fibre
Channel for the enterprise systems
 NAS is easy to install and relatively easy
to maintain
 Network is used exclusively for data
transfer causing additional overhead
 Backup using LAN is really a overhead
Storage Area Networks
Storage Area Networks
 As much as 60% of the traffic on a std
corporate network is made up of
housekeeping actions like Backup
 Storage Area Network has been fuelled
significantly by the desire to get this
housekeeping off the network
 Primary interface for SAN infrastructure is
Fibre Channel
Storage Area Networks
 SAN provides excellent performance and
easier management
 SAN implementations are expensive due
to hardware costs
 Better resource sharing could make up for
the initial investment
 SAN is very flexible in that more storage
and servers can be added easily
iSCSI
 Motivation
– GB Ethernet
 iSCSI is a draft
standard protocol to
encapsulate SCSI
commands into
TCP/IP packets
 Can be used to build
IP based SANs
Storage Virtualization
 The research firm Gartner Group estimated that
80% of the storage costs is used up for managing
the installed storage
 Switch and array management becomes very
difficult with increased storage hardware
 Virtualization provides a logical view and eases
management.
 Examples – Veritas Volume Manager, IBM
Tivoli etc.
Future of network storage
 SAN islands connected by IP networks
 Network Unified Storage (NUS)
– NAS + SAN on GB Ethernet networks
High Availability
Host
Single Point Failure
Disk
Host
Multi Pathing
Disk
Dynamic Multi Pathing
Increased disk availability
Load balancing
Identifies disks uniquely from different
hosts
Dynamic Multi Pathing
/dev/rdsk entries
c1t1d0
c2t1d0
….
cnt1d0
/dev/rdsk entries
c1t1d0
c2t1d0
….
cnt1d0
Host
Host
..
Disk
Implementation Details
 Scan the disks listed in /dev/rdsk
 If no UUID is present, generate a unique UUID
and stamp it in the disk’s private region
 Add this device to a hash table hashed on UUID
 Load this table to the kernel and write the ioctls
to update this info
 Use an algorithm (Currently round robin) to
efficiently load balance the I/O requests.
 If a path is bad for more than five I/O attempts,
mark it bad and do not use it for path selection.
Implementation Details
 User code
– Read /dev/rdsk folder and generate a hashed list of
available disks
– Load this list to kernel. Also provide APIs to push
newly added/removed disks.
 Kernel code
– Filter driver to choose the best path
– ioctls to do the following
•
•
•
•
LOAD_DISKS
NEW_DISK
MODIFY_DISK
GET_DISK_HANDLE …
Questions??