Introduction to Storage types - basic.ppt

Editor's Notes

• #2: These are the topics that will be covered in this unit.
• #3: Information (data) is constantly flowing in and out of computers. Data coming in is either stored in memory for quick access (but not permanent storage) or written to disk so that it can be retrieved later. How this information is stored depends on a couple of things: - How will it be stored (what will it be stored ON)? - What is the protocol by which this data will be transferred? - What devices will it pass through on the way to being stored?
• #4: This is an example of traditional internal data storage. (Similar to what you might have on your laptop or desktop computer.)
• #5: Data from outside the system is entered via keyboard or some other interface…
• #6: … and goes through the CPU, memory, bus, RAID controller and into internal storage – i.e. a hard disk drive.
• #7: Again, data from outside the system passes through the CPU, memory, bus, and RAID controller. In this case, it then is passed outside the system to some type of external disk enclosure where it is written to disk drives. Even though the data is stored outside the computer system, this is still referred to as Direct Attached Storage since the RAID controller is internal to the computer system, and the disk drives are attached to the controller via a single path.
• #8: Regardless of whether storage is internal to the server or external, there must be a connection between the RAID controller and the disk drives, and a communication protocol must be used to communicate across this connection. In the case of external storage, a cable connects the RAID controller and the external array. Even for internal storage, a short “ribbon” cable is commonly used to connect the RAID controller and disks. Most desktop systems with internal storage utilize the ATA or Serial ATA (SATA) protocols to communicate with disks. For desktop systems, internal disks are either ATA disks or SATA disks, depending on the protocol. ATA is a parallel communication protocol that can be used to communicate with a limited number of devices over a short distance. SATA is a serial form of the ATA protocol, which allows a true cable to be used, and allows a greater distance between the controller and disks. SCSI is a parallel protocol that allows increased distance between the controller and disks. SCSI is often used for internal storage on servers. When SCSI is used for internal storage, SCSI disks are used as well. Although SCSI may be used for internal storage on servers, its main use is with external storage. A single RAID controller may attach to up to 14 disks in an external cabinet. Most commonly, external disks attached to a SCSI RAID controller are SCSI disks. However, protocol converters may be used within the external cabinet to allow ATA or SATA drives to be attached to the SCSI bus. This is commonly done to lower disk costs. In addition, the SATA protocol allows “pure” SATA external arrays to be built. The RAID controller utilizes SATA to communicate with SATA disks, with similar characteristics to external SCSI or Fibre Channel arrays (there will be more about Fibre Channel in the next section). However, SATA arrays are currently slower and less reliable than SCSI arrays. Due to their low cost, they are ideal for backup-to-disk strategies and nearline storage. Evolving specifications for SATA promise competitive characteristics to SCSI or even Fibre Channel in the future.
• #9: A true external storage system is defined as external by the fact that the disk enclosure and “smarts” of the system (RAID controller) are outside the computer that is receiving the data. In this case, the flow of data on the computer system is again through the CPU, memory and a bus, but then it is transferred to the external storage system through a Host Bus Adapter (HBA). From the RAID controller on the external system, the data is then passed into storage on the disk drives.
• #10: To allow an external storage cabinet with an external RAID controller, a more flexible communication protocol is needed than SCSI. The Fibre channel protocol allows the “smart” RAID controller to reside outside the server, connected through a relatively “dumb” Host Bus Adapter (HBA). A single HBA (or two HBAs) may be directly attached to the external array, in a manner similar to SCSI or SATA Direct Attached Storage. However, the external array may contain many more disks than a SCSI array. (Disks may be Fibre, SCSI, SATA, or ATA with appropriate converters). In addition, Fibre Channel connections are not limited to a single direct path. Storage communication paths may split and merge, to form a true network. We will discuss this more under the Storage Area Network section.
• #11: From the perspective of the server, the main difference between storage attached via Fibre Channel and storage attached via SCSI is the nature of the interface hardware. SCSI controllers contain most of the intelligence needed to manage the array “onboard” the server. With Fibre Channel, the RAID controller is actually outside the server, in the external storage cabinet. To communicate with the outside world, the server utilizes a Host Bus Adapter (HBA). An HBA is a simpler device than a RAID controller, simply transferring the data over a cable to an external RAID controller. Although the use of an HBA offloads storage processing to the external RAID controller, it does not guarantee faster performance for any given IO. The same amount of processing for the IO must be done in either SCSI or Fibre Channel scenarios. An external SCSI array may actually be faster for some types of IO. Fibre Channel does enable greater flexibility on how storage is networked with servers. Fibre Channel may also offer greater throughput for multiple simultaneous IOs on one or more servers attached to the same external storage array.
• #12: How data is stored once it gets to the storage disk drive(s) depends on the type of storage selected. Data storage comes in many different formats. We’re all familiar with what it’s like save a file to our hard drive or to a floppy or CD. Those are all forms of storage. Obviously, it can get a lot more complicated than that. Following is a list of the most common types of data storage: Single disk drive (self explanatory) JBOD – just a bunch of disks. This is collection of disk drives pooled together for storage, but without any RAID, striping, etc. Volume – a “logical” disk drive. A concatenation of drives. When one fills up, it goes to the next one. No RAID, no striping. To the OS, a logical volume looks like one disk drive. Storage Array – Also a group of more than one disk joined together – but can do striping and/or redundancy. Implies some type of RAID (whatever the level). SCSI - SCSI stands for Small Computer System Interface. It is a means of attaching additional storage to a computer. For example, a typical RAID Controller is a SCSI device that allows connection to an external storage enclosure with multiple drives. NAS – Network Attached Storage. Sometimes rather than simply attaching storage to one machine, it is attached to the computer network. That way, multiple machines can access the storage. A file protocol must be used to communicate across the network. ISCSI – Internet/SCSI protocol. Another approach to offering storage on a network. Rather then using file protocols to communicate across a TCP/IP network, native SCSI commands are “encapsulated” in TCP/IP packets. An evolving standard that has already been adopted in Windows 2003 Server. SAN - Storage Area Network. Whereas a NAS is storage that is attached to a network, SAN is a storage network in and of itself that can be attached to multiple machines. SAN is an industry-wide term for both the storage and the switching network. A SAN does not have the protocol conversion overhead of NAS or ISCSI, and tends to offer better performance. However, a SAN may require a higher initial investment in infrastructure.
• #13: A NAS (Network Attached Storage) is designed to provide shared access to storage across a standard TCP/IP network. Sharing data across TCP/IP is accomplished by converting block-level SCSI commands to file sharing protocols. Common file sharing protocols include the UNIX Network File System (NFS) and the Windows Common Internet File System (CIFS). Linux or Windows servers may be used to share network files. However, “Appliance” servers are becoming readily available that offer better performance. These servers utilize a stripped-down operating system that is built to optimize file protocol management, and commonly support multiple file sharing protocols.
• #14: Unlike DAS or SAN, there is no RAID Controller or HBA on the server. Instead, a Network Interface Card is used to communicate with the NAS “server” across the TCP/IP network. The NAS server also utilizes a TCP/IP card. The ethernet network can be either a private or public network. Due to the data traffic and security concerns, a VLAN is preferred when using the public network. Native SCSI commands address storage at the block level. However, native TCP/IP can only communicate storage information at a higher logical level – the file protocol level. This means that a server must send file level requests over TCP/IP to the NAS “server”, which must convert file protocol information to block level SCSI information in order to talk to the disks. Returning data must be converted from block level disk information to file protocol once again and send it across the network cable in TCP/IP packets. Although a Gigabit Ethernet network is fast, all of this protocol conversion incurs significant overhead. The situation is even worse for database requests, because the database “talks” only in block level format to the database server, so protocol conversion must occur coming and going. Because of this, NAS may not be appropriate for all databases. Read-only databases may offer acceptable performance on NAS, as well as relatively small transactional databases. However, large transactional databases are rarely placed on NAS, due to perceived performance reasons. Despite the potential drawbacks, a NAS system may offer good performance at a good price, depending on your situation. A high end NAS appliance over a 1 Gigabit ethernet network can offer performance similar to a SAN. The advent of 10 Gigabit ethernet should alleviate any performance concerns.
• #15: iSCSI (Internet SCSI) storage systems are similar to NAS in that communication between servers and storage is accomplished over standard TCP/IP networks. However, iSCSI does not utilize file protocols for data transport. Instead, SCSI commands are encapsulated in TCP/IP packets and sent over the network (encryption may also be performed). iSCSI is supported through the operating system. Both Windows 2003 Server and Linux support iSCSI.
• #16: iSCSI communication can occur through standard Network Interface cards. However, the OS then incurs substantial overhead in managing TCP/IP encapsulation. A new type of NIC for storage is arriving on the market, sort of an iSCSI HBA. These cards use an onboard TCP/IP Offload Engine (TOE). TOEs perform encapsulation at the hardware level, freeing processor cycles on the server. Although the performance of iSCSI is not yet up to the speed of SCSI or SAN storage, the pace of improvement is rapid. The performance is perfectly acceptable for Small to Medium sized Businesses, and works well with non-mission critical databases. With the adoption of 10 Gigabit networks, iSCSI will become increasingly attractive, even for mission critical applications. Recently, high-performance iSCSI systems have been benchmarked at 90% of the performance of Fibre Channel SANs, at an attractive price/oerformance ratio.
• #18: A Fibre Channel SAN is a true network for storage. A SAN consists of one or more servers, one or more switches, and one or more Fibre Channel Storage Arrays. Components are connected by fiber optic cables (copper cabling may also be used). For redundancy and fail-over capabilities, many components may be doubled. This includes HBAs, switches, and RAID Controllers.
• #19: A SAN (Storage Area Network) is designed to provide access to storage over a dedicated private fibre channel network. Fibre Channel networks can connect multiple hosts and multiple storage devices, and can even incorporate fibre channel switches for enhanced connectivity. SAN solutions are implemented more widely than any other networked storage solution.
• #20: The key benefit of a SAN is storage consolidation. Because of the networked nature of a SAN, multiple servers can be attached to one storage array. This leads to significant cost savings. In addition, higher performance can be achieved through increased overall throughput. Multiple paths from each server to the shared SAN can be implemented. This not only enhances performance, but also adds redundancy to the data path, allowing for failover from one data path to another. Redundancy can be added at every level of the data path, offering a storage solution with no single point of failure. This is particularly valuable for cluster implementations. The current generation of mid-tier storage offers 70TB of storage or more in a fully configured RAID array.
• #21: The most widely implemented SAN technology is currently based on the Fibre Channel protocol. Fibre channel features up to 127 devices within a network, with up to 10 KM of cabling. This is substantially more than SCSI offers. FC2 can operate at two gigabit speed, making it faster than the ethernet networks implemented at most sites.
• #22: As quoted from the Garner Group in October of 1999, the benefits of using Fibre Channel storage include: Greater distance between servers and storage (as much as 10km) Links are simpler and more reliable than SCSI Disk storage consolidation is facilitated Tape resources can be shared on loops and switched fabrics in conjunction with resource serialization storage management software FC-connected disk storage and tape libraries, combined with FC switches and appropriate software in a SAN will allow Windows NT servers to support outboard backup. Data moves form disk to tape over the SAN, unloading the servers and eliminating server bus bandwidth as a constraint on backup speed.
• #23: This shows the actual FC frame. Inside the FC frame, a SCSI command is embedded. Also within the FC frame are some check sums and a header that has source and destination information. A Fibre Channel SAN often contains multiple HBAs and multiple processors. Within the FC frame, the source and destination information provide the addressing required to route the correct packet to the right location or device (similar to Ethernet).
• #24: Fiber is the cable; Fibre Channel is the protocol. Fibre Channel does not require fiber optic cable. It can be implemented on copper as well. Some variations of Fibre Channel over IP allow Fibre Channel over TCP/IP networks, utilizing common ethernet cable.
• #25: Fibre Channel is efficient at transmitting data because it is a serial interface. Serial interfaces use a separate send and receive wire. This allows data to be transferred full speed in both directions simultaneously. Because of the separation of send and receive data, no complex signal handling is required.
• #26: Internal to the Fibre Channel protocol, the SCSI command set is used. The device driver talks SCSI. Each SCSI command ride on top of fibre channel. (Similar to how IP rides on top of Ethernet.) The FC command itself goes across the fibre optic or copper cabling. Note the similarity to TCP/IP interface layers.
• #27: Sometimes the term “SCSI” refers to the SCSI protocol and sometimes it is used to refer to a SCSI device. We’ll discuss the protocol first. SCSI is a protocol and command set that provides services similar to TCP/IP. It’s been in use for a long time and when Fibre Channel was introduced, there was no reason to change the protocol that is used to communicate between devices. (Why re-invent the wheel? SCSI was already tested and accepted w/ every major operating system on the market.) So the protocol that Fibre Channel uses is SCSI.
• #28: A FC transmission is sent across only two “wires” (or two fiber optic strands, thus the slim FC cable). A SCSI transmission would send an 8 bit transmission across 8 different wires (thus the thick SCSI cable). There has to be additional signaling to let the receiving device know what is being transmitted – i.e. 8 bits being transmitted “now.”
• #29: The biggest difference between SCSI and Fibre Channel is availability. Fibre Channel was designed w/ all sorts of built-in redundancy. Part of the reason Fibre Channel came about was that people were looking to overcome downtime associated with SCSI. For example, with SCSI, if you break a cable, SCSI is going to die. Unlike SCSI, Fibre Channel is not limited to a single physical loop. A break on one cable in an FC network will not automatically disable devices connected to other cables. In addition, multiple Fibre Channel connections may be utilized between FC devices. This allows for virtually unlimited redundancy, so that the loss of one cable need not disrupt communications at all. SCSI is single bus – FC is redundant network for all components to talk to each other. SCSI hardware: Disk drives and controller – e.g. raid controller is SCSI hardware With fibre channel, you still have a RAID controller, but, by definition, it will be on the external storage. It will do the logical to physical translation of data, but the physical transfer from the internal system to the external storage will be handled by an HBA. 3 predominant SCSI standards: – Ultra2 (80 megabytes/sec) – Ultra 160 – 160 MB per sec – Ultra 320 – 320 MB per sec Fibre Channel standards: - FC1 – 100 MB/sec - FC2 – 200 MB/sec So how important is this difference in bandwidth? It depends on what you’re doing. When you’re talking about database applications, you usually never come close to using the entire bandwidth – it’s hardly ever the limiting factor. Fibre Channel provides a data link layer above the physical interconnect; analogous to Ethernet. This is completely different from SCSI. FC is network of devices that can be media independent – that’s why it can run on copper or fibre optic. SCSI has to run on copper Fibre channel is closer to a network. Devices talk either fiber optic or copper. Fibre channel changes the network – the physical layer - that things run across. There’s no way to run on a different medium on SCSI. SCSI is media specific. The limitations with SCSI and FC are also different. Cables have to be certain length: With SCSI, no more than 3 feet. No more than 15 devices on a SCSI bus (bus sort of analogous to FC loop – which can have 127 devices). SCSI pins have to be a certain way. IF SCSI were to change, then have to change every device to accommodate. With FC, if ran diff media – would have to change cables, but not devices. The number of disk drives that you can connect with SCSI is going to be less than with Fibre Channel. Lets say that you have 4 SCSI buses with 15 devices (a controller or a disk) on each, then you can have 60 devices. Whereas you can get 127 devices on a fibre channel loop. (Assuming no other devices on that loop.) SCSI Storage Characteristics The average SCSI clustered solution allows at most two RAID controllers for clustered disks. For most SCSI systems on the market, this results in a maximum of 24 - 48 disk drives for the cluster’s shared disks. If that amount of storage is not sufficient for the customer’s current and future growth needs, then Fibre Channel solutions should be considered. Fibre Channel Like SCSI, Fibre Channel is another means of attaching storage to a computer. The big differences between Fiber Channel and SCSI is the redundancy that it allows you and the higher availability (don’t know if we want to get into needing terminal on SCSI, etc.) A common misconception about Fiber Channel is that it is faster than SCSI. In reality, it’s not necessarily faster, but may allow more I/Os (and the increased redundancy & higher availability). Fibre Channel Characteristics Fibre Channel provides high availability through redundant Fibre Channel host bus adapters (HBAs) in the servers and dual storage processors in the storage system. Fibre Channel also requires dual standby power supplies (SPS) in the storage system, which provides integrity of the storage processor write-cache in case one power supply fails. Thus write-caching may be enabled, which will improve disk write performance. With SCSI, write-caching may not be used with a failover cluster. Fibre Channel allows for easy storage subsystem growth. It is relatively easy to add disks to Fibre Channel storage system without having to shutdown the system. A directly connected Fibre Channel solution can support greater than 200 disks on one array. For a Storage Attached Network (SAN) solution, it is possible to have greater than 900 disks on one cluster connecting through a Fibre Channel switch. The same switch can also support additional stand-alone systems and clusters. Although Fibre Channel systems are typically more expensive than SCSI, the expandability and flexibility is much greater than SCSI.
• #30: .
• #31: Microsoft’s Simple San Approach is designed to benefit both Fibre Channel SANs and iSCSI SANs.