Transcript XSAN

Storage Area Networks
The Basics
Storage Area Networks
SANS are designed to give you:
• More disk space
• Multiple server access to a single disk pool
• Better performance
• Option of disk distributed across multiple
locations
Direct Attached Storage
Classically, for storage we had a single box with
a bunch of disks attached:
Public
Network
LUN0
QuickTi me™ and a
TIFF ( Uncompressed) decompressor
are needed to see thi s pi ctur e.
LUN1
QuickTi me™ and a
TIFF ( Uncompressed) decompressor
are needed to see thi s pi ctur e.
Server
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
SCSI Bus
LUN2
QuickTi me™ and a
TIFF ( Uncompressed) decompressor
are needed to see thi s pi ctur e.
Attached Storage
The server speaks to the SCSI disks using a command
language:
• Read from LUN0, Block 123
• Write to LUN1, Block 456
All this goes over the SCSI bus, which is directly
attached to the server; only that server has access to
the bus
The server would create a filesystem on the disk(s) and
could then make the disk available to other
computers via NFS, Samba, etc.
Network Attached Storage
This idea is easily extended to an appliance
approach. Configure a utility box with some
disk that does only NFS or Samba/SMB, place
on network
Public Network
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
NFS Client
NAS Server
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
NFS Server
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
SCSI Bus
NAS and Servers
Redundant web servers share the same data--but they both
talk to the same NFS server
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Web server, data
NFS mounted
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Web server, data
NFS mounted
Public Network
NAS Server
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
NFS Server
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
SCSI Bus
Attached Storage
We can also do things like place a RAID array on the
NAS server.
This works, but it has some limitations:
• If the server goes down, there is no access to the disk
• File sharing goes through the network storage server
and across the network, which can be slow
• Limitations on location of disks--must be near server,
within range of the disk bus
• Adding or subtracting disk space can be difficult
What we want is a shared disk pool that all servers can
access
Storage Area Network
What we want is something that looks like this:
Public Ethernet Net
NFS Client
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
SAN Participants
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Disk Pool
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Storage Area Network
Notice:
• You can take down a server and still maintain access to the disk
pool via the other SAN participants
• Disk added to the pool is available to all servers, not just one
• Shared, high speed access to the disk pool; can run clustered
copies of SQL database or web server if the SQL databases or
web servers are also SAN participants
• Can still serve up the disk pool via an NFS or SMB server on a
SAN-connected box
• “serverless backups”--just send command to copy blocks from
disk A to disk B. Snapshots easier, shortened backup windows-you can have a SAN particpant handle moving a volume to tape
Storage Area Network
So how does this work? It’s a scaled up version of the
old system.
• The commands being sent are the same disk standard
commands: either SCSI or ATA disk bus commands,
READ, WRITE, etc.
• The network connecting the SAN servers to the disk is
typically (but not always) higher speed, eg
FibreChannel
• Some extra glue to allow for concurrent access by
more than one server--need a shared filesystem
• Special filesystems to allow for concurrent access
Storage Area Network
A popular choice:
• SCSI for the bus commands (commands sent
over the wire)
• Fiber Channel for the SAN network
• EMC or similar for the glue volume software
Fiber Channel is 2+ Gbit/sec, and can be
deployed across up to a 500m distance
(sometimes) and up to 70 KM with special
equipment
Storage Area Network
Another option is to use gigabit ethernet for the
SAN networking.
• Cheap! Commodity equipment, don’t need to
learn new Fiber Channel stuff, reuse existing
gear
• But also lower performance--fibre channel has
higher BW, and can use more of it.
ATA Over Ethernet
AoE uses ethernet plus ATA bus commands
rather than SCSI. Low cost; but since
ethernet frames are not routable all devices
must be on the same network
iSCSI
iSCSI uses SCSI bus commands over ethernet,
encapsulated inside of TCP/IP
• Cheap hardware!
• Well supported in Linux, Solaris, and Windows world
• Because the SCSI is inside of TCP/IP, it is routable-which means you can do a SAN across wide area
networks (with lower performance due to latency)
and do things like mirror for disaster backup, or
across campus on high performance networks
• Processing TCP/IP takes some overhead; some use
TCP offload chips
iSCSI
Each “disk”/LUN is a RAID array that understands iSCSI.
Public Ethernet Net
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
NFS Client
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
iSCSI
The green network is a dedicated (usually) gigabit ethernet
network that carries the SCSI commands encapsulated inside
TCP/IP. The red network connects the SAN participants to other
clients not on the SAN
Important point: TCP/IP is routable. That means that (modulo
latency) the devices can be located anywhere. We could have a
iSCSI SAN participant in Root Hall, and one in Spanagel. The
Root iSCSI server can access the disk pool in Spanagel
We could also have a volume located at Fleet Numeric in the same
SAN
The price we pay for this is having to process the TCP/IP overhead
as iSCSI commands go up the network protocol stack. This can
be alleviated in part by TCP offload chips
Volume Software
Remember, the iSCSI targets are just block
devices. iSCSI says nothing about concurrent
access or multiple hosts accessing the same
devices
For that we need a SAN Filesystem. This
deconflicts concurrent access by hosts to the
block devices
Volume Software
Public Ethernet Net
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Vol2
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
NFS Client
Quic kTime™ and a
TIFF (Unc ompres sed) dec ompres sor
are needed to see this pic ture.
Vol1
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
Quic kT i me™ and a
T IFF (Unc ompres s ed) dec ompres s or
are needed t o s ee thi s pi c ture.
SAN Software
The “volume software” allows you to build a
concurrent access filesystem out of one or
more LUNs
iSCSI
Example: Five compute servers need read
access to one weather data set. If the servers
are all on the SAN, they can directly access
the data
Example: backup. Copy disk blocks directly,
then have a tape drive SAN participant copy
to tape
Example: storage expansion. Just add more
disk, and it is available to all SAN participants
Competitors
iSCSI’s competitor is for the most part fibre channel.
The concept of fiber channel is almost identical, but
the SCSI commands are simply encapsulated in a
fibre channel frame
Fibre channel is typically higher performance--more
data can be pushed across FC, and there is much less
overhead processing FC frames
BUT it is higher cost
ATA Over Ethernet is very similar to FC in concept-directly inserting the ATA commands in ethernet
frames. But it seems to have less market penetration