The student will become familiar with:
The various physical categories of server,
The advantages and disadvantages of each class,
Factors in deciding on the physical class.
The various well defined roles of the server,
The basic objectives of each role,
The logical grouping of roles.
The student will become familiar with the various physical categories or form factors of the server family of IBM PC compatible computers including the advantages and disadvantages of each category and factors involved in choosing the server form factor as it relates to the servers role(s) and the physical server framework. The student will become familiar with the major well defined roles as recognized within the IT industry and the basic functionality expected from each type of server. Based on this information the student can make intelligent choices concerning how to group logically related roles into a single server while designing the logical server framework.
A few of the basic server form factors have already been mentioned in the server planning phase of the server project. Now this subject will be explored in further depth. It should be noted that the "IBM PC Compatible" or "industry standard microcomputer" is only one class of general purpose computer that can fulfill the server roles of the organization.
|Supercomputer: Most powerful (the day they are built), most expensive, special order designed and built, "installation level" (whole buildings or wings are devoted to housing them), usually proprietary and require specially trained staff to install, configure, maintain, and use them.|
|Mainframe: Most powerful regular production models, very expensive, "installation level" (whole buildings or wings are devoted to housing them), usually proprietary and require specially trained staff to install, configure, maintain, and use them, usually feature standard interfaces for easier integration with microcomputer (and other) client systems.|
|Minicomputer: Expensive, usually fully compatible with the same manufacturer's mainframe systems and expansion modules, usually proprietary and require specially trained staff to install, configure, maintain, and use them, usually feature standard interfaces for easier integration with microcomputer (and other) client systems.|
|Microcomputer: Least expensive, usually industry standard architecture, much smaller and are usually the best investment in price versus computing power although they still suffer from a relatively weak upper limit on their maximum computing capability.|
Obviously the supercomputer, mainframe, and minicomputer fall beyond the scope of the coverage of the course, but it is well worth noting that even some "barebones" minicomputers are actually more powerful than a "maxed out" PC-based server design and while still more expensive, the minicomputer can be upgraded to computing capacities far beyond the ability of any PC-based system. Here is a typical minicomputer: the IBM iSeries Model 9406-595:
- Dimensions: H:79.7" W:30.9" D:52.2"
- Weight: 2995lb (no integrated UPS)
- IBM advanced POWER5TM processors—the ninth and tenth generation of 64-bit processor technology
- Multi-platform operating environment with capability to support i5/OS, Linux, Microsoft Windows Server,
AIX 5L, plus application environments such as WebSphereTM and JavaTM
- On demand pricing options help match purchasing flexibility with technology and business needs
- Extensive portfolio of proven solutions available in all operating environments
- Enterprise-class dynamic logical partitioning for allocation and virtualization of resources—up to 10
partitions per processor
- Enhanced support for Web modernization of 5250 OLTP (On-line Transaction Processing) applications
- Flexible packaging options—Standard, Enterprise, High Availability and Capacity BackUp Editions deliver
outstanding value and provide growth options
- Capacity on Demand features to dynamically apply system resources for accommodating peak application
workloads — includes On/Off Capacity on Demand which allows you to pay only for the capacity you activate
- Includes i5/OS and can add Windows, Linux and AIX 5L operating systems
- 8/16-way, 16/32-way, 32/64-way and 4/32-way CPU offerings provide from 31500 CPW to 216000 CPW
(Commercial Processing Workload, these are 16, 32 and 64 CPU based parallel processing systems)
- 16 GB to 2 TB (2048 GB) memory
- Up to 2700 disk drives – up to 381 TB of capacity
- Up to 96 I/O expansion towers/drawers via High Speed Links
- Up to 1152 PCI-X slots, up to 160 LANs (160 independent Network Interface Controllers)
- Up to 60 Integrated xSeriesTM Servers (complete independent single box module server)
- Up to 57 Integrated xSeries Adapters
- Redundant, hot-plug components for additional reliability
The minicomputer's base hardware inventory including 16GB of RAM is larger than most PC's can be expanded up to, and having 16 microprocessors definitely outclasses anything that PC's can do, for now. It should be noted that the PC-Based system has the least upgradeability and the least powerful individual components in every category except one: the PC industry has the most powerful high-end CPU's. Unfortunately, only a few motherboards are available that can support up to a maximum of four of these CPU's. Since all server operating systems now feature SMP - Symmetric MultiProcessing, in which the OS can take full advantage of all of the installed processors, this limitation definitely makes the PC-based server the weakest amid the classes of general purpose computer.
PC-based servers come in the following physical form factor classes:
|Standard PC: Desktop and Tower form factors intended for use in regular PC configurations.|
|Rackmount: Standard telecommunications 19" rackmountable. The actual width of the device is 17.75" The standard telecommunications device modular rack system allows for the installation of 42U or height units. Each overall height unit is 1.75" so the standard maximum occupied height of the rack is 73.5" or 6 feet 1.5 inches. Rackmount server chassis are available in 1U (actually 1.719" height) to 27U and even larger.
Tour image taken from www.dell.com, clearly the interior of a 1U rackmount server is so cramped that
Typical 19" telecommunications rack occupied by Ethernet patch panels (top), switches (middle) and a hub (bottom). This rack also has "cable management," the plastic shells covering both sides and crossing in front of the network devices.
|Cabinet Chassis: While some of these are standalone "doublewide" towers, some are also rackmountable such as the DELL PowerEdge server already discussed in the Server Planning Phase.
The large standard 19" rackmount thumb screws are clearly visible on each side of this rackmountable cabinet chassis server that can also have casters installed on the bottom so it can be used standalone.
|Blade server: The next generation of compact server beyond the rackmount server is the blade server. No matter how thin the rackmount server is made, it will still occupy 1U of a telecommunications rack (1.75" height "footprint") Furthermore, a standard rackmount server must be 17.75" wide and typical modern rack systems allow for a maximum depth of 39.4" which is space that the server occupies that other units cannot occupy. To this end, manufacturers are designing blade enclosures that are rackmountable and into which individual blade servers are then mounted. This allows the form factor of the blade server to be even smaller and allows many more blade servers to be mounted into the rack than if each one actually occupied its own rack units of space. The IBM Bladecenter below occupies 9U but houses up to 14 blade servers which mount vertically side-by-side from the front, each blade server is even smaller than a 1U rackmount server meaning the form factors of the components are even smaller. Typical blade server systems may omit the power supply from the individual blade servers instead the blade server chassis may provide power to all of the blade servers installed in it.
|Custom server chassis: Companies may on occasion need to order custom chassis for their server(s). These can be any shape and size and can be made of any materials.
Despite their small form factors, PC-based rackmount servers and blade servers can be quite powerful due to the changes made in the standard component form factors installed in them. And it should be remembered that the PC-based server in general is more than adequate as a solution for any organization due to the separation of roles amongst groups of servers working together in the logical and physical server framework. Ultimately for a similar expenditure a group of PC-based servers may have more capabilities than a single minicomputer. That is, 14 blade servers and their chassis may price out to around $30,000, while the minicomputer may cost over $90,000. In the end, the 42 blade servers may have as a total on a well designed physical server framework better workload handling and throughput than the single minicomputer combined with the fact that the overall server framework would be far more available especially if the blade servers were installed at different geographical locations within the organization.
Aside from these typical physical categories or form factors of servers, it should be noted that servers on the network together as members of the server framework. Servers are then categorized according to their population and arrangement within the server framework:
|Standalone: Effectively the only server on the network. There are rare instances when a standalone server can be used within a larger LAN where more servers do exost, in this case the standalone, will not be aware of these other servers and will not cooperate or coordinate its activities with them.|
|Member server: Member servers do work together on larger LANs and are specifically installed and configured to cooperate and coordinate their activities on the network amongst each other. This server is part of the network infrastructure or the logical server framework.|
|Cluster server: Multiple PC-based systems are specifically designed and installed such that working together they form a single server on the network. This server might then be either a standalone server or a member server. This server can only exist if the individual nodes out of which it is built are connected by a dedicated physical server framework.|
Cluster servers are becoming much nore common as the hardware and the operating system support improves. Microsoft Windows 2000 Advanced Server included native support for connecting two nodes into a single cluster server. Windows Server 2003 included the ability to connect four nodes into a single cluster server. As this support continues to be developed and improved, cluster servers may well become the norm in enterprise level computing. Either way, servers either reside alone on the network or are configured to be aware of each other and to therefore cooperate and coordinate their activities amongst each other. While developing this configuration is beyond the scope of the course, the Server+ technician should be aware that this must be done in order to create a large LAN's logical server framework that will function properly and efficiently. For working with Microsoft Windows server products the recommended certification is the MCSE - Microsoft Certified Systems Engineer. For Linux there are many, some notable ones include: RHCE - Red Hat Certified Engineer, and SCNA - Sun Certified Network Administrator.
The physical architectural requirements of a server differ very slightly from the average PC, but the subtle differences can become significant as the workload demands on the server increase. Server capacity planning is definitely not an exact science, but there are some factors that can be taken into consideration:
Server Capacity Planning Factors
|Operating system requirements: The server operating system itself will have a minimum CPU/RAM/HDD space requirement. Microsoft, for example lists these minimum requirements for Server 2003:
The recommended RAM should be the minimum, and more RAM will be added based on other capacity requirements. This OS recommends 550Mhz CPU which is not a problem, and up to 8 CPU's. This would include virtual cores such as Intel's Hyperthreading technology as well as physical cores such as dual or quad core processors. Therefore a motherboard supporting two quad core processors has reached the maximum CPU capacity of Server 2003.
|Application requirements: Many individual applications offer recommendations for minimum and recommended RAM and sometimes CPU requirements as well. These values should be added onto the OS requirements.|
|Number of clients: This one is difficult to gauge. While the server might have 100 clients on the network, how many are actually engaged in tranasaction requests with the server at any given time? How many establish permanent sockets (TCP/IP connections) and how many do not establish permanent sockets (instead building, using and tearing down the socket with each transaction). Web browsing clients for example request transient sockets, as such they create much more traffic per request than LAN clients attaching toan SQL Server through a front-end client side application which will establish permanent sockets. The only real method of determining client traffic capacity requirements is to baseline them during the Pilot Phase.|
|Network throughput: This is application specific and the application may indicate its bandwidth. For example, a streaming media server delivers "256Kbps MPEG-4" therefore each client that is going to access this server's media content is going to require 256Kbps of network bandwidth. If it is not available then the quality of their connections will begin to degrade rapidly. The maximum number of concurrent requests to the server will have to be anticipated.|
|Storage: Even the smallest modern hard drives have sufficient storage capacity in todays computing industry, but their access speed is an entirely different story. The hard drive is the throughput bottleneck in any modern IT environment. Since the hard drive is orders of magnitude slower than everything else, and it holds all data including the operating system itself, it stands to reason that improving storage performance, if possible, must have the greatest effect on overall system performance. The highest speed interface would be the first choice, therefore 4Gbps fiberChannel SCSI (800MB/sec) would be the first choice. However, these controllers cost thousands as do each peripheral. A cost effective solution might be SCSI Ultra320 (320MB/sec) or even SATA-II (300MB/sec). In the end, the drive's random seek time and maximum platter-to-buffer data throughput however will form the upper limits of what any single physical hard drive can do when attached to these controllers. High RPM directly improves random seek and platter-to-buffer throughput. A typical relationship between capacity and speed was found in a random search through online hard drive vendors: 1000GB 7200RPM SATA-II drive was $305 and a 150GB 15000RPM U320 SCSI drive was $315. The SCSI drive is at least two times faster due to the RPM alone. Its superior bus (in more ways than pure speed) is also a factor to be considered. The slower drive is however over seven times larger. The final and possibly best method of maximizing hard drive performance is by implementing a RAID - Redundant Array of Inexpensive Disks. A RAID-0 with three hard drives reads and writes to all three simultaneously and therefore runs three times faster than any single drive. These are by far the most economical solutions and will be explored in depth in a later discussion.|
The physical architectural requirements of a server differ very slightly from the average PC, but the subtle differences can become significant as the workload demands on the server increase.
Server Physical Architecture Design Parameters
|Processing: Many server roles depend on maximized processing power. This indicates the demand for more physical processors and more cores per processor. Additionally, processors in general perform much better when they have multiple levels and large amounts of high speed cache. The next criteria for maximized processing capability is very high front-side bus throughput which in turn depends as much on the chipset as it does on the CPU architecture. Both Intel and AMD have processor and chipset products intended specifically for the server platform. These components concentrate specifically on high levels of raw computing or processing capacity. The Intel main product line intended for servers is the Xeon family and the AMD processors for servers are the Opterons.|
|RAM: Almost all server roles depend on handling large amounts of data simultaneously. In a perfect world, the server would only have to serve a single client, but it is certainly far from a perfect world and the demands on the the modern server are climbing exponentially. Todays server has to provide services to more clients per server than ever before. Because many client nodes are requesting the services of a single server node on the network, the servers workload, the server should be capable of caching into RAM all of the data that all of the clients need immediately, rather than have to fetch this information as needed from the disk drives. While Windows 2000 Terminal Server indicates a minimal requirement of an additional 128MB per client the optimal performance will probably be achieved with 256MB and this depends entirely on the specific applications being executed in the servers RAM space for the client. Since Terminal Services is the most processing intensive service this can be used as a "yardstick" for the RAM requirements per client. For simple services like small file processing (web server, mail server, etc. that handle small pieces of data/files at a time) probably 4 to 8 MB per client would be sufficient. For large processing intensive services like an SQL server working with a large database, the server itself is running the processing intensive application (database manager, i.e. SQL Server) and the minimum system requirements of that specific application should be exceeded as much as possible.|
|Storage: Servers intend to store the majority of the networks vital information. Workstations however will likely be expected to store their own operating system and possibly all of their applications. If this is the case, then the server will be expected to store its own OS, applications, and the network's data. Modern applications can generate enormous data files and since the server will be responsible for storing all user data, it will need large and upward scapable storage capacity. The server will also be expected to handle multiple concurrent user connections accessing this data, so the server should have high speed mass storage technologies in place.|
|Network Connectivity: Servers primary roles involve network accessibility and usually with multiple concurrent client connections. While clients may be able to function adequately with 100Mbps connectivity, the server may lag if it is accessed by multiple concurrent 100Mbps connections when it only has a single 100Mbps connection to the network itself. Considerations to alleviate network connectivity bandwidth limitations include using a 1000Mbps NIC attaching to the backbone fabric of the switch, or attaching the server to the switch with multiple adapters, or both.|
|Power requirements: Servers generally run more powerful CPU's with greater power requirements and often have multiple CPU motherboards as well. Furthermore, servers often have storage RAID's involving more than one hard drive and may include other high power consumption devices such as high capacity tape backup drives. Servers generally have far greater power consumption than end-user PC's and will need much larger power supplies.|
|Availability requirements: Servers have availability requirements that end-user systems simply do not address at all. This includes special hardware considerations including but certainly not limited to component redundancy and fault tolerant systems from the power supply to the hard drive and will significantly increase the design costs of available PC-based servers.|
The IT industry recognizes many specific roles of the modern server and it should be noted that these roles come and go very quickly like just about everything else within the high-tech PC industry sector. Server roles as has already been pointed out can be divided into the specific categories of
Within these general categories of role modern servers fall into what might be called "well defined" server roles which include but are certainly not limited to:
Local Hardware Share Server Roles
|Streaming Multimedia Servers: Also called Audio/Video servers provide the ability to broadcast or multicast multimedia content provided by the organization's Web site(s). Difficult to pidgeonhole since this is as much a web server as it is a storage server and will be categorized as both in this list.|
|FTP Servers: One of the oldest of the Internet services, File Transfer Protocol makes it possible to move one or more files securely between computers while providing file security and organization as well as transfer control. Another form of web service that is also functioning as a storage server.|
|File Servers: The fundamental storage server holds all relevant files of the network at a centralized, high capacity, manageable location. Almost considered archaic in the modern IT industry, this functionality aside from being native to almost all server operating systems, the role is usually combined with others such as the FTP server.|
|Print Servers: The fundamental local hardware share allows more than one client to use the same printer(s) and modern print servers introduce such additional features as fully configurable and manageable print queues and can manage user's usage of the printers. The role is usually combined with others such as the file server.|
Local Software Share Server Roles
|Application Servers: Also known as middleware systems, application servers range from database servers to terminal services servers and are becoming the single largest population of modern servers in use.|
|Chat Servers: Chat servers allow users to exchange information in real-time over the network. The users must connect to the chat server and its application interacts with each user and sorts all concurrent information, moving messages to and from the correct users.|
|Groupware Servers: Groupware servers run special software or special versions of existing software designed so that users to collaborate in a real-time virtual workspace environment across the Internet or corporate intranet regardless of their geographical locations.|
|IRC Servers: An older internetworking connectivity version of the chat server with real-time capabilities. Internet Relay Chat itself consists of various separate networks of servers that allow users to connect to each other via an IRC network over the Internet.|
|Mail Servers: Mail servers gather, store and transmit mail over corporate networks and across the Internet servicing both direct client connections for sending and retrieving their email as well as servicing other email servers by delivering bundles sent to users of that server and picking up the bundles from those servers intended for users of itself.|
|News Servers: News servers act as a distribution and delivery source for the thousands of public news groups currently accessible over the USENET news network.|
|Telnet Servers: A Telnet server allows users to log on to the server from any node engage in a remote console session able to perform any task on the server as if they were working from the server console directly. Windows allows Terminal Services to be installed in "Remote Administrative Mode" which is a GUI version of this same functionality.|
|Terminal Servers: Users log on to virtual remote desktops of the server. The client node only runs the Terminal Services Client application which carries the local keyboard and mouse actions to the Terminal Server, events occur within the application running entirely in the server's virtual desktop environment and changes to the screen are carried back to the client as compressed images which it then displays on the client node's screen. As server computing capacity increases, this role may also increase in popularity as it allows for the implementation of "disk-less workstations" or inexpensive client nodes with very low-end hardware configurations.|
|Web Servers: The web server is the most difficult basic server role to categorize. It is by definition an application server running the web server application. It is by definition a storage server holding documents that users want to see (the webpages) and it is a communications server often providing layer 2 gateway technologies.|
Communications Share Server Roles
|Fax Servers: A fax server reduces incoming and outgoing telephone resource usage while still fulfilling the need to fax actual documents. This server may well be heading toward deprecation as modern business will transform entirely to email which is far more capable than the fax technologies for exchanging documents.|
|Telnet Servers: A Telnet server generally provides access to the LAN through a dial-up connection and thereby acts as a layer 2 gateway between modem connectivity and local area network technologies.|
|Remote Access Servers: Also known as dial-up servers or RADIUS - Remote Authentication Dial-In User Service server. While the telnet server only allows a single user to establish a remote console to itself, RAS servers generally do not want a close connection to themselves and are devoted to passing through the remote modem users requests onto the rest of the LAN and are fully functional layer 2 gateways.|
|LAN Gateway: A specific role layer 2 gateway such as interconnecting two Ethernet LAN's or an Ethernet LAN and a Wireless LAN would be examples of using the server as a gateway. While there are dedicated network connectivity devices that can fulfill these roles, servers are sometimes prefered because of their ability to perform other roles and run more diverse software packages while being quite often the less expensive option.|
Network Service Provider Roles
|Proxy Servers: Proxy servers stand between a client program such as a Web browser and another server such as a server on the Internet and filter requests, improve performance, and share connections. Proxy servers are as much communication shares as they are network service providers.|
|DHCP Servers: Provide automated TCP/IP client configuration on the network and track "IP leases" so that they will not be duplicated on the network.|
|Name Servers: Provide automated IP-to-Hostname resolution services so that clients only need to remember user-friendly names of resources (like www.yahoo.com) rather than node IP addresses. DNS - Domain Name System is quickly becoming the standard for LAN's as it has been for the Internet from the beginning.|
In combining server roles certain guidelines should be followed. These guidelines are general in nature and certain specific circumstances can create situations in which they would not be followed. A prime example of this would be a small LAN using a primary standalone server. In this circumstance, the server provides all roles regardless of these general guidelines.
Network Service Provider Roles
|Combine similar/related roles: File servers and print servers are both local hardware share servers and are often combined into file and print servers. A natural role combination would be the remote access server and the proxy server, both are attached to the single Internet connection of the location and provide connection management at the single point of entry between the Internet and the LAN.|
|Assign one application role to each server: Application servers are in general running the most processor/data throughput intensive services on the network and should not be combined either with each other or with any other services.|
|Put high availability systems to work: If the system architecture plans to employ one or more backup or standby servers, change them from fail-over servers to load balancing servers. This increases their level of "OS rot" and therefore shortens their operational availability lifespan, but also does not waste their processing/data throughput capabilities either. In the situation where the plans were for a primary standalone and a backup primary standalone, this can be changed to a pair of member servers and split the services and roles between them. Or set them up as a primary standalone and a backup primary standalone in load balancing mode rather than fail-over mode.|
List and describe the four classes of general purpose computer:
Minicomputers are a superior class of general purpose computer to which other class?
Microcomputers have one superior component to the other classes of general purpose computer which is:
Aside from processing power, what other general features of PC-based microcomputers are lower than those of the minicomputer:
Given the same investment what class of server may actually be better than a minicomputer? In what ways?
List and describe the five categories of PC-based server form factors:
Which PC-based server form factors require custom components? What is the main reason these two types of server need custom components?
List and describe the three categories of PC-based server based on their relationships within the server framework:
One of the Planning Phase requisites in choosing the server operating system is to have trained and preferably certified staff. List three such certifications:
List and describe the five server capacity planning factors:
What is the method of determining the application server capacity planning factor:
What are the methods of improving the storage capabilities of the server capacity planning factor:
List and describe the six server architectural design parameters:
What are the four general types of server roles:
What are the methods of improving the network connectivity throughput server architectural design parameter:
Many server roles are difficult to classify because they combine more than one of the general types of role, list the describe the local hardware share types:
Many server roles are difficult to classify because they combine more than one of the general types of role, list the describe the local software share types:
Many server roles are difficult to classify because they combine more than one of the general types of role, list the describe the communication share types:
Many server roles are difficult to classify because they combine more than one of the general types of role, list the describe the network service provider types:
Of all possible server roles which type of server is the most common (greatest number exist)? Explain.
Of all possible server roles which type of server places the most demands on it by the clients. Explain why?
Of all possible server roles which type of server should be installed alone with no other roles on the server. Explain why?
If the server must undertake more than one role how should they be combined. Explain why. Give two examples of pairs of server roles that would do well if combined into one server.
What is one way to determine the number of network clients that the server can successfully support?
What is the Intel server processor product line? Go online and research the Intel website looking for the processor in this family with: the most cores, and the most onboard core speed cache. Then price it.
What is the AMD server processor product line? Go online and research the AMD website looking for the processor in this family with: the most cores, and the most onboard core speed cache. Then price it.
Research the Intel and AMD websites for server product chipsets. Determine which ones support two or more (SMP) of the CPU's you found above. Search the Internet for motherboards using these chipsets and offering two or more sockets/slots for these CPU's. Since they are not compatible, you will be finding two different motherboards. Price two of these motherboards, one for the Intel CPU's and one for the AMD CPU's and list the motherboard features. Note which features are open to the motherboard manufacturers to change and add as they choose.
Research the Internet to see if there are benchmark performance comparisons of the motherboards and processors that you chose in the preceding questions. Which was better? Was it better in all benchmark categories?
Copyright©2000-2008 Brian Robinson ALL RIGHTS RESERVED