- What is a server?
- Serving up some history
- Client-side computing
- Client-server to the fore
- Server hardware on the market
- Server technical specifications
- Warranty and support
Vendors supply technical specifications with servers, much as they do with other IT hardware. These technical specifications provide the detail you need to determine if the server is the right model for your requirements.
If you know what to look for in a PC, you are a long way to understanding what to look for in a server. The following overview of server components will help you align the technical specifications sheet with the overall function of the server.
CPUThe Central Processing Unit (CPU) is a microprocessor that performs most of the data processing. It is the main processor in the computer and often one of the more expensive components in a server. Servers can have one or more processors.
Servers can have CPUs from a range of vendors. Intel, AMD, Sun Microsystems, Hewlett-Packard, IBM and Motorola all make server-specific CPUs. The primary provider of CPUs for entry-level and workgroup servers is Intel and AMD.
- Intel server CPUs: Pentium III Xeon, Xeon, Xeon MP, Itanium, Itanium2.
- AMD server CPUs: AMD Opteron.
One of the primary differences between CPUs used in PCs and those in servers is that servers have multiple CPUs, which provides improved performance and availability.
For a detailed explanation of processors please read the CPU Buying Guide.
MemoryRAM (Random Access Memory) is a high-speed form of storage that gives the CPU quick access to data.
The basic rule of thumb is, the faster the RAM - and the more of it - the faster the computer performs. Servers can generally accommodate significant amounts of specific RAM designed for speed and accuracy.
Most current Intel- or AMD-based servers will use DDR (Double Data Rate) SDRAM (Synchronous Dynamic Random Access Memory). Some current servers will use SDRAM.
RDRAM (Rambus Dynamic Random Access Memory) is an alternative memory technology designed to work with motherboards and CPUs that use an 800MHz Front Side Bus.
Memory may also be available in ECC (Error Correcting Code) format. The memory itself adds extra information to the data that it processes in order to ensure its correctness.
Based on the current state of the memory market, it is advisable to configure servers with at least 1GB of ECC DDR SDRAM. Your server software provider may recommend more.
StorageData is stored on hard disk drives. Hard disk drives in file servers are based on technology, which provides the fastest transfer of information while at the same time offering the best-of-breed data integrity.
Servers are generally configured with multiple hard drives, deployed in such a way as to vastly decrease the chances of data loss due to hard disk failure. This also provides greater storage volumes.
If physical space is insufficient inside the server to support the number of hard disk drives required, external storage system may be used.
Hard Disk Drives: Commonly, hard disks in servers differentiate themselves by their interface connections, disk spin speeds, latency times and buffers. These four items all affect performance and price. The SCSI interface has been the most common in servers as it allows for far more data to be throughput than an IDE connection. (The relatively recent advent of the Serial ATA (SATA) interface provides for throughput comparable to SCSI drives at a lower cost.) Disk spin speeds refers to the rotation rate - or revolutions per second - of the disks within the drives.
Higher spin speeds improve performance. SCSI throughput is comparable to IDE throughput. The benefit of SCSI is parallelization. That is, as you add more disks, you scale better than linearly. This isn't the case with large IDE systems. SCSI disks are also designed with reliability in mind. For example, many IDE disks ship configured to *not* flush buffers even when instructed. Only a very small number of SCSI disks are configured by default in this way. Latency is how long you wait before you start receiving data (throughput is the rate at which you receive it, once you start receiving it). Lower latency results in better drive performance. Finally, the size of disk buffers is measured in megabytes and refers to a temporary storage area that assists the drive interface to process the data with integrity.
Hard drive interface speeds
- ATA100 (also ATA 6): 100MBps
- ATA133 (also ATA7): 133MBps
- Serial ATA (also Ultra ATA): 150MBps to 300MBps to be ratified shortly
- Ultra3 SCSI: 160MBps
- Ultra160 SCSI: 160MBps
- Ultra320 SCSI: 320MBps
RAID: RAID (Redundant Array of Independent Disks) allows multiple disks to be connected together as one virtual drive for performance, redundancy, or both. In most cases RAID is used to provide redundancy. This allows a hard disk to fail without loss of data or uptime. If a hot-swap disk drive is available, the RAID system can be configured to switch over to using the spare disk. If the disk drives are hot pluggable, the faulty disk can be replaced and configured to become the new hot-swappable drive. It is essential to check that the operating system you intend to run will support the RAID controller in the server you are going to buy.
In an entry-level/workgroup server, you may wish to consider mirrored (RAID), ATA or SCSI drives onto which the operating system and applications are installed.
For more information on RAID, see the table below.
ConnectivityWithout connectivity a server is useless. Ethernet running at 10, 100 or 1000Mbps is the de facto local area networking standard. Most entry-level/workgroup servers will include 10/100Mbps Ethernet NICs (network interface cards). If you require Gigabit Ethernet or other networking technology you will probably have to specify that at purchase time, or buy the network cards separately.
Be sure to check that the NICs are supported by the operating system you intend to run.
You may wish to include redundant NICs in your server. This will allow the server to transparently reroute network traffic from a NIC in the server that has failed to another working NIC in the server, without having to stop users from working by shutting down the network to effect repairs. You will have to configure your server operating system and network to support this.
This is a RAID
RAID Level 0: This is the most basic model. Features: data striping, without fault tolerance, works on a normal hard drive, data is stored on consecutive sectors of the same disk. Uses a minimum of two disk drives and divides data into blocks that range from 512 bytes to several MBs, which are written alternately to the disks. When the system reaches the final drive in the array, it writes to the next available segment of Drive 1, and so forth. Striping the data distributes the I/O load evenly across all the drives. And since drives can be written to or read from simultaneously, performance increases noticeably. Well suited to applications such as video production and editing or image editing.
RAID Level 1: Is disk mirroring and duplexing- everything written to Disk 1 is also written to Disk 2 and can be read from either disk. This provides instant backup but requires the highest number of disk drives and doesn't improve performance. Offering the best performance and fault tolerance in a multiuser system, RAID 1 is the easiest configuration to implement, and it works best for accounting, payroll, financial and high-availability data.
RAID Level 2: Was developed for mainframes and supercomputers. It corrects data on the fly, but RAID 2 is prone to high error-checking and correcting ratios.
RAID Level 3: Includes data striping, but it also assigns one drive to store parity information. This provides some fault tolerance and is especially useful in data-intensive or single-user environments for accessing long sequential records. RAID 3 doesn't overlap I/O, and it requires synchronised-spindle drives to prevent performance degradation with short records.
RAID Level 4: Includes large stripes so that records can be read from any single drive. It's rarely used because it lacks support for multiple simultaneous write operations.
RAID Level 5: Is similar to Level 0, but instead of dividing data into blocks, it stripes the bits of each byte across multiple disks. This byte-striping adds overhead, but if a drive fails, it can be replaced and the data reconstructed from parity and error-correcting codes. RAID 5 overlaps all read/write operations. It requires three to five disks for the array and is best suited to multiuser systems that don't need critical performance or that do few write operations.
RAID Level 6: Is rarely implemented commercially. It extends RAID 5 using a second parity scheme distributed over different drives. It can sustain multiple simultaneous drive failures, but performance, especially for write operations, is poor, and the system requires an extremely complex controller.
RAID Level 7: Includes a real-time embedded operating system as a controller and high-speed bus for caching. It gives fast I/O, but it's expensive.
RAID Level 10 or 0 + 1: A combination of RAID Levels that utilises multiple RAID 1 (mirrored) sets into a single array. Data is striped across all mirrored sets. As a comparison to RAID 5 where lower cost and fault tolerance is important, RAID 0+1 utilises several drives to stripe data (increased performance) and then makes a copy of the striped drives to provide redundancy. Any disk can fail and no data is lost as long as the mirror of that disk is still operational. The mirrored disks eliminate the overhead and delay of parity. Offers high data transfer advantages of striped arrays and increased data accessibility. System performance during a drive rebuild is also better than that of parity based arrays, since data does not need to be regenerated from parity information, but is copied from the other mirrored drive.
RAID level 0 + 5 or 50: A combination of RAID levels that utilises multiple RAID 5 sets striped in a single array. A single hard drive failure can occur in each of the RAID 5 sides without any loss of data on the entire array. If, however more than one disk is lost in any of the RAID 5 arrays all the data in the array is lost. As the number of hard drives increase in an array, so does the possibility of a single hard drive failure. Although there is an increased write performance, once a hard drive fails and reconstruction takes place, there is a noticeable decrease in performance, data/program access will be slower, and transfer speeds on the array will be effected.
RAID Level 53: Is implemented as a Level 0 striped array, in which each segment is a RAID 3 array. It has the same redundancy and fault tolerance as RAID 3. This could be useful for IT systems needing a RAID 3 configuration with high data-transfer rates, but it's expensive and inefficient.