2.1 NFS File Servers

This section provides a hardware overview of Sun NFS servers. Figure 2-1 and Table 2-1 illustrate under what conditions a particular Sun NFS file server will meet your needs.

Figure 2-1

Figure 2-1 (Continued)

Table 2-1

Server	Maximum Specifications	Positioning	Key Advantages
SPARCcenter 2000 or 2000E	500 NFS clients 36 subnets 731 Gbytes of disk storage Ethernet, FDDI, SunATM, and token ring	Highest capacity Multipurpose enterprise server (the total solution for an entire company)	Centralized administration Maximum headroom for growth Multiprocessor, I/O, and network performance scalability
SPARCcluster 1	500 NFS clients 30 subnets 150 Gbytes of disk storage Ethernet backbone	High capacity Cluster of distributed servers Dedicated NFS server	Simplified administration Centralized control of distributed servers Lowest cost per NFS ops
SPARCserver 1000 or 1000E	300 NFS clients 12 subnets 395 Gbytes of disk storage Ethernet, FDDI, SunATM, and token ring	High capacity Multipurpose workgroup server	Excellent capacity performance Multipurpose server (NFS, compute, database) Affordable Scalability Integrated packaging
SPARCserver 20	125 NFS clients 138 Gbytes of disk storage 4 subnets Ethernet, FDDI, SunATM, and token ring	Low cost Multipurpose workgroup server PC LAN server	Low cost, yet powerful, flexible, and easily redeployed
SPARCserver 5	60 clients 40 Gbytes of disk storage 4 subnets Ethernet	Low cost Multipurpose workgroup server PC LAN server	Low cost Very good NFS performance at a low cost

2.1.1 SPARCserver 5 System

The SPARCserver 5 system is based on the microSPARC^(TM) .. CPU. It provides fast application performance and networking extensibility. It is ideal for a dedicated workgroup NFS server.

2.1.1.1 SPARCserver 5 System Features

The SPARCserver 5 system features are:

• 70 MHz or 85 MHz microSPARC II CPU
• 16 to 256 Mbytes of RAM (8 or 32 Mbyte SIMMs)
• 64-bit memory bus
• Two internal hard drives,
• 644 Mbyte CD-ROM drive (optional)
• 1.44 Mbyte diskette drive (optional)
• Twisted pair Ethernet (can be expanded between 5 and 9 Ethernet networks)

  · Optional AUI support

• Three SBus expansion slots
• Two serial ports, one parallel port
• Greater than 20 Gbytes mass storage capacity with external disk drives
• SBus Prestoserve NFS accelerator (optional)

A SPARCserver 5 system can be upgraded to a SPARCserver 20 system by substituting the system board with a SPARCserver 20 system board. Figure 2-2 shows a front view of the SPARCserver 5 system.

Figure 2-2

2.1.2 SPARCserver 10 System

The SPARCserver 10 system is a high-performance workstation designed to be a workgroup NFS file server or a database server in an office environment.

The system is available in various uniprocessor and multiprocessor configurations. Some of the uniprocessor configurations available are:

• ^(TM) · Model 30LC--single 36MHz SuperSPARC processor, no external cache
• Model 40--single 40MHz SuperSPARC processor, no external cache
• Model 41--single 40MHz SuperSPARC processor, 1 Mbyte of external cache
• Model 51--single 50MHz SuperSPARC processor, 1 Mbyte of external cache

Some of the multiprocessor configurations available are:

• Model 402--two 40 MHz SuperSPARC processors, no external cache
• Model 512--two 50 MHz SuperSPARC processors, each with 1 Mbyte of external cache
• Model 514--four 50 MHz SuperSPARC processors, each with 1 Mbyte of external cache

Figure 2-3 shows a front view of the system.

Figure 2-3

2.1.2.1 SPARCserver 10 System Features

The features include:

• High performance throughput based on balanced coherent bus, graphic, and networking components
• SuperSPARC superscaler processors set in a multiprocessor architecture
• Memory (RAM) ranging from 32 Mbytes to 512 Mbytes
• With four SPARCstorage Array subsystems (Model 101), the disk storage capacity of the disk arrays is 126 Gbytes.
• Twisted Pair Ethernet (standard) and Thicknet (AUI) with an optional adapter cable
• Four SBus expansion slots
• Two serial ports, one parallel port
• 16-bit audio
• NVRAM-NVSIMM and SBus Prestoserve NFS accelerators (optional)

2.1.3 SPARCserver 20 System

The SPARCserver 20 system is designed to be a workgroup NFS file server or a database server in an office environment. It is based on the same MBus and SBus technologies as the SPARCserver 10 system. Performance is increased over the SPARCserver 10 by using faster MBus and SBus technology, and faster SPARC^(R) modules. The SPARCserver 20 system has increased compute and network performance.

The SPARCserver 20 system is available in three uniprocessor configurations and three multiprocessor configurations. The three uniprocessor configurations are:

• Model 50--50 MHz SuperSPARC processor
• Model 51--50 MHz SuperSPARC processor and 1 Mbyte SuperCache
• Model 61--60 MHz SuperSPARC processor and 1 Mbyte SuperCache

The three multiprocessor configurations are:

• Model 502MP--two 50 MHz SuperSPARC processors
• Model 514MP--four 50 MHz SuperSPARC processors and 1 Mbyte SuperCache
• Model 612MP--two 60 MHz SuperSPARC processors and 1 Mbyte SuperCache

Figure 2-4 shows the front view of the SPARCserver 20 system. This system and the SPARCserver 5 system look alike from the front view.

Figure 2-4

2.1.3.1 SPARCserver 20 System Features

The SPARCserver 20 system features include:

• Over 2 Gbytes of internal hard disk storage (two 1.05 Gbyte internal disk drives)
• Provides up to 126 Gbytes of disk storage in the SPARCstorage Array (Model 101) subsystems when directly connected to four SPARCstorage Array subsystems
• 1.44 Mbyte diskette drive (optional)
• 644 Mbyte CD-ROM drive (optional)
• Two serial ports, one parallel port
• Twisted pair Ethernet
• Up to 512 Mbytes of memory (60 ns SIMMs)
• AUI Ethernet (optional) (can have up to 9 Ethernet networks)
• SBus or NVRAM-NVSIMM Prestoserve NFS accelerator (optional)

2.1.4 SPARCcluster 1 System

The SPARCcluster 1 system is a multiple host server designed for NFS management. The SPARCcluster 1, based on clustering technology, uses up to four SPARCserver 20 system cluster nodes to deliver file services to large departments. Thus, the SPARCcluster 1 system spreads NFS workload across multiple systems. The SPARCcluster 1 system is designed for medium to large NFS file server environments that require a dedicated system for file service. It supports up to 500 clients.

The SPARCcluster 1 system provides:

• A high-uptime server with a minimum of downtime
• A large capacity server that delivers superior response
• A server that can expand easily to accommodate many networks, disks, and memory, without compromising one for the other
• A server emphasizing networking, disks, and CPUs, thus eliminating bottlenecks in NFS file server applications

Figure 2-5 shows the front view of the SPARCcluster 1 system.

Figure 2-5

To fully understand how the system works, you must understand the logical architecture and the physical architecture, which are described later in this section.

2.1.4.1 SPARCcluster 1 System Features

The features include:

• Fast network file server
• NFS performance scalability
• High uptime though the cluster architecture
• Modular design allows new technologies to be easily incorporated
• Single integrated package configured for optimal NFS file servers
• Centralized system administration tools to manage users, data, and networks

The SPARCcluster 1 system contains the following hardware components:

• Two or four SPARCserver 20 desktop systems
• Two or four differential SCSI disk trays, each containing from two to six 2.9 Gbyte disk drives
• One Multi-Tape Backup Tray containing two to four 5.0 Gbyte 8 mm tape drives
• Terminal concentrator
• Ethernet switch with 6 to 15 ports
• Power distribution unit
• Cables required for internal SCSI, network, and power connections
• NVRAM-NVSIMM Prestoserve NFS accelerator (optional)

2.1.4.2 SPARCcluster 1 System Logical Architecture

The SPARCcluster 1 system contains two or four SPARCserver 20 systems, each with a differential SCSI disk tray and a single-ended SCSI tape backup device. The SPARCcluster 1 system can be attached to up to eleven NFS client networks, with each host available on all networks. A high-performance Ethernet network switch connects all the hosts to all client networks.

Figure 2-6 shows the logical architecture of the SPARCcluster 1 system using six client networks, as an example.

Figure 2-6

As Figure 2-6 shows, any workstation attached to any client network has free access to any of the SPARCcluster hosts. In addition, a terminal concentrator is connected to the console of each SPARCserver 20 host, allowing the system administrator to access each host for installation and maintenance from a single, networked workstation.

To implement this logical architecture, one host name and IP address is needed for each host-network logical connection. For example, a system with four hosts and six client networks requires 24 (4 hosts x 6 networks) IP addresses and host names.

2.1.4.3 SPARCcluster 1 System Physical Architecture

In contrast to the logical architecture, the SPARCcluster 1 system physical architecture uses only one internal network per server host. Each host is connected directly to the internal Ethernet switch. To complete the connection, all client networks are also connected to the internal Ethernet switch. For more information on the SPARCcluster 1 networking, see Section 2.1.4.4, "SPARCcluster 1 System Network Architecture."

The Ethernet switch and SPARCcluster 1 software perform the multiplexing functions that make each host appear to be connected to every network, even though each host has only one internal network connection. Figure 2-7 shows the physical architecture within the SPARCcluster 1 system.

Figure 2-7

2.1.4.4 SPARCcluster 1 System Network Architecture

Two key concepts underlying the SPARCcluster operation are:

• Port groups
• Ethernet multiplexing

A port group is a collection of network input and output ports that are logically combined to appear as a single IP (Internet Protocol) network. Ethernet multiplexing is a function that combines the network traffic of several port groups over a single, physical Ethernet cable.

Interfaces on the Ethernet switch are divided into client ports and server ports. The server ports are also called mux ports, because they contain functionality that is used to multiplex several IP networks over the same physical cable. A SPARCcluster configuration, by default, has two or four server ports, and up to 11 client ports.

The Ethernet switch, connecting all the hosts to the client networks, is a modified, high-speed bridging device using information specific to the SPARCcluster 1 in packet headers to function more like a router than a bridge. The switch retains the benefits of bridging technology, namely low cost per port, extremely low latency (delay) per packet, and high scalability in terms of the number of ports and throughput of the switch.

Adding the host-resident SPARCcluster network multiplexing software and the switch modifications allows the switch to gain the primary advantage of a router: the ability to connect to multiple IP networks, instead of the single IP network restriction usually associated with a bridge. In short, the Ethernet switch combines the best features of a bridge and router so that many IP networks can be multiplexed together over a single Ethernet connection.

No IP routing is provided by the Ethernet switch. It essentially performs a selective forwarding of packets, operating more like a bridge than as a router. To move packets from one IP network to another, an external router or one of the SPARCcluster hosts must forward the packets and update destination information. The Ethernet switch has no knowledge of IP addresses, no store-and-forward capability and it makes no dynamic routing decisions.

Network multiplexing is done by the Ethernet mux (emux) device driver, which sits in between the Lance Ethernet (le) and IP drivers in the network protocol stack. Figure 2-8 shows the relationship between the emux driver, the le driver, and the TCP/IP code used by NFS.

Figure 2-8

Bridges may be used to join segments connected to the Ethernet switch and other segments that do not attach directly to the SPARCcluster, as shown in Figure 2-9.

Figure 2-9

A common configuration for a SPARCcluster 1 system is a dedicated router, or a set of gateway machines in parallel with the clustered system. Routers and gateways provide high-speed, high-capacity inter-network paths while the

SPARCcluster 1 system is optimized for services between clients and the servers. A typical SPARCcluster 1 system network configuration is shown in Figure 2-10.

Figure 2-10

Routing is done by the hosts in the cluster, not by the Ethernet switch. Therefore, each packet that is routed traverses a server network twice: once from the switch to a routing host, and once from the host to the switch after the packet has been updated with routing information. Routing should be kept to a minimum if the SPARCcluster 1 system is the primary router in the network.

2.1.4.5 SPARCcluster 1 System External Physical Architecture

The primary network is the client network that is used during installation of the SPARCcluster 1 system. An administration workstation with a SunCD^(TM) drive must be connected to the SPARCcluster 1 system, via the primary network for installation and maintenance. In addition, the terminal concentrator is attached to the primary network. This external configuration assures that all system components are on the same network for installation and management functions.

Figure 2-11 shows the internal and external SPARCcluster 1 system connections.

Figure 2-11

2.1.5 SPARCserver 1000 and SPARCserver 1000E Systems

The SPARCserver 1000 and SPARCserver 1000E systems are designed to be departmental NFS file servers with ultimate flexibility for dedicated or multifunctional application environments. They supports less than 200 NFS clients and provide file, database, timeshare, or computing services to a network and attached devices. The SPARCserver 1000 and the SPARCserver 1000E systems feature multiprocessor architecture allowing incremental system expansion.

Modular design makes expansion of system boards, memory and peripherals a routine task you can perform on-site. Processor expansion and upgrades are easily accomplished by adding or replacing SuperSPARC modules. Figure 2-12 shows the front view of the SPARCserver 1000 or the SPARCserver 1000E system.

Figure 2-12

The SPARCserver 1000 and the SPARCserver 1000E systems are built around the XDBus, the system bus that provides very high data transfer bandwidth.

The CPU building blocks consist of:

• Up to 60 MHz SuperSPARC modules (50 MHz SuperSPARC on the SPARCserver 1000 system)
• SuperCache controller
• 1 Mbyte of cache

The system has three SBus slots per system board, which are accessed by the XDBus. The multiple memory banks process in parallel information from multiple CPUs and I/O devices.

2.1.5.1 SPARCserver 1000 and the SPARCserver 1000E System Features

The SPARCserver 1000 and the SPARCserver 1000E systems include the following features:

• Up to four system boards can be installed
• Up to eight SuperSPARC processors (two per system board) can be installed
• Up to 2 Gbytes of main memory (using 32 Mbyte SIMMs) can be installed
• Up to 16.8 Gbytes of internal storage
• 50 MHz system clock speed in the SPARCserver 1000E system (40 MHz in the SPARCserver 1000 system)
• 25 MHz SBus speed in the SPARCserver 1000E system (20 MHz in the SPARCserver 1000 system)
• When connected to 12 SPARCstorage Array subsystems, the SPARCserver 1000 or 1000E systems provide up to 378 Gbytes of external storage
• Up to 12 SBus slots
• Onboard SCSI-2 port and twisted pair Ethernet on each system board
• Internal 5 Gbyte 4mm tape drive (or 10 Gbyte 8mm tape drive)
• Internal CD-ROM drive
• NVRAM-NVSIMM Prestoserve NFS accelerator (optional)

2.1.6 SPARCcenter 2000 or SPARCcenter 2000E Systems

The SPARCcenter 2000 or the SPARCcenter 2000E systems provide the computing solution for a company. As such, the SPARCcenter 2000 system and the SPARCcenter 2000E system are multifunctional network NFS file servers. They supports less than 500 NFS clients and have the flexibility required for dedicated or multifunctional application environments.

The SPARCcenter 2000 system and the SPARCcenter 2000E system provide scalability and extensive expansion in these areas:

• CPU processor power
• Memory capability
• I/O connectivity

They meet the following requirements:

• High capacity I/O requirements of corporate data centers
• Computationally intensive demands of other organizations

The heart of the SPARCcenter 2000 system or the SPARCcenter 2000E system is a high-speed packet-switched bus complex that provides very high data transfer bandwidth. The backplane supports two distinct XDBuses operating in parallel.

The SPARCcenter 2000 system or the SPARCcenter 2000E system use up to twenty SuperSPARC modules in a shared-memory symmetric multiprocessing configuration, meeting most performance needs. You can expand or upgrade the processing capability by adding SuperSPARC modules.

Main memory is configured in multiple logical units that are installed in the bus complex.

The I/O is expandable. For example, you can configure up to 40 SBus slots on 10 independent busses. The large I/O capacity and configurability makes the SPARCcenter 2000 system or the SPARCcenter 2000E system suitable for very large applications.

2.1.6.1 SPARCcenter 2000 and the SPARCcenter 2000E System Features

The SPARCcenter 2000 system and the SPARCcenter 2000E system features are:

• Up to 20 SuperSPARC modules (each with up to 2 Mbytes of built in cache)
• Dual XDbus system bus
• Greater than 5 Gbyte main memory capacity (8 Mbyte or 32 Mbyte SIMMs)
• Up to 10 SBus channels (for a total of 40 SBus slots)
• Two RS232 serial ports on each system board (up to 20 total)
• Twisted pair Ethernet connections via SBus cards
• 10 Mbytes/second SCSI-2 channels via SBus cards
• Up to 52.2 Gbytes of internal SCSI disk storage
• 50 MHz system clock speed in the SPARCserver 2000E system (40 MHz in the SPARCserver 2000 system)
• 25 MHz SBus speed in the SPARCserver 2000E system (20 MHz in the SPARCserver 2000 system)
• When connected to 20 SPARCstorage Array subsystems, Model 101, the total disk capacity for the expansion subsystems is 630 Gbytes
• NVRAM-NVSIMM Prestoserve NFS accelerator (optional)

Figure 2-13 shows the front view of the SPARCcenter 2000 or the SPARCserver 2000E system.

Figure 2-13

2.2 Disk Expansion Units

This section describes an overview of the following disk expansion units:

• SPARCstorage Array subsystem
• Desktop Storage Module
• Multi-Disk Pack
• Desktop Disk Pack

2.2.1 SPARCstorage Array Subsystem

To expand your disk storage, consider the SPARCstorage Array subsystem. This disk array is a high-performance and high-capacity companion unit for the SPARCcenter 2000 or 2000E, SPARCserver 1000 or 1000E, SPARCserver 10, and the SPARCserver 20 systems.

The Model 101 uses 1.05 Gbyte single connector 3.5-inch disk drives. Each disk array contains three drive trays. Each drive tray supports up to ten 3.5-inch single-connector SCSI disk drives. All disk drive SCSI addresses are hardwired. The position of the disk drive in the drive tray automatically sets the SCSI addresses. Each disk array uses six internal fast, wide SCSI buses. Each drive tray contains two SCSI buses that support five disk drives for each SCSI bus.

Figure 2-14 shows a front view of the SPARCstorage Array subsystem.

Figure 2-15 shows various ways you can connect the SPARCstorage Array subsystem to your NFS server.

The SPARCstorage Array subsystem uses RAID (Redundant Array of Inexpensive Disks) technology. The levels of RAID available are:

• RAID 0.......Striping data without parity
• RAID 1.......Mirroring
• RAID 0+1......Mirroring optimized stripes
• RAID 5.......Striping data with parity

Within the disk array, independent disks plus RAID levels 5, 1, 0, 0+1 are available at the same time. This allows you to easily match data layouts to meet the specific requirements for capacity, performance, high availability, and cost.

If any disk in a RAID 5, 1, or 0+1 group fails, an optional hot spare (if configured) is automatically swapped in to replace the failed disk. Continuous, redundant data protection is provided, even if a disk fails.

Warm plug service allows you to replace one or more disks without needing to power down the system, the disk array, or reboot the system. You can obtain warm plug service if multiple disk arrays are configured.

Using the SPARCstorage Array subsystem can improve NFS performance. Its processor manages and schedules disk I/O.

The SPARCstorage Manager software, provided with the disk array, provides similar functionality to Online: Disk Suite software. Since there are often many more disks to manage in the SPARCstorage Array subsystem, the SPARCstorage Manager software has an intuitive GUI interface.

2.2.2 Desktop Storage Module

The Desktop Storage Module (disk unit) is an external hard disk drive unit for desktop servers and the SPARCserver series. It contains one full-height 1.3 Gbyte disk drive. Figure 2-16 shows a front view of the Desktop Storage Module.

Figure 2-16

2.2.3 Multi-Disk Pack

The Multi-Disk Pack is a disk expansion unit for the SPARCserver 5, SPARCserver 10, and SPARCserver 20 systems. It comes in two configurations. The first configuration is four 1.05 Gbyte fast SCSI disk drives totaling 4.2 Gbytes of disk storage. The second configuration is either two or four 2.1 Gbyte fast SCSI disk drives totaling 4.2 or 8.4 Gbytes of disk storage. The SCSI addresses are set by SCSI address jumpers on the disk drives. Figure 2-17 shows a front view of the Multi-Disk Pack.

Figure 2-17

2.2.4 Desktop Disk Pack

The Desktop Disk Pack is a disk expansion unit containing one SCSI disk. Some of the disk capacities are:

• 207 Mbyte low profile disk drive
• 424 Mbyte disk drive
• 535 Mbyte low profile disk drive
• 1.05 Gbyte low profile disk drive

The SCSI address is set by a SCSI address switch at the rear of the unit. This expansion unit is useful for the following desktop server systems:

• SPARCstation 5 system
• SPARCstation 10 system
• SPARCstation 20 system

Figure 2-18 illustrates a front view of the Desktop Disk Pack.

Figure 2-18