If geeks love it, we’re on it

Servers made simple

Servers made simple


Servers made simple for business owners and managers

The word server has become a part of the language of business. Many think a server as an enigmatic object apparently every business needs, can’t do without and costs a lot of money. The reluctance to invest in a server stems from the lack of understanding of what it is and what it does. This is coupled with apprehension of cost. Those businesses with existing and aging servers question why the current hardware, that may have been a substantial investment years ago, is need of replacing.

Servers can also baffle. Groupware, scalability, storage needs, upgrade path, data management, load and balance are peculiar words that begin to creep into conversations. Mainstream marketing show monolithic blocks called servers and tell of flexible architecture, self-adjusting, self-correcting and self-managing infrastructure.

Whatever it is it sounds intimidating…and expensive. This article lays the foundation for understanding server technology and what benefit it can be. Servers are a valuable part of business and once the basics are understood then any business can proceed to increase productivity, communication, data integrity and, believe it or not, revenue without tipping the budget way over into the red.

A server simply is computer hardware and software that serves information to others. It doesn’t get more complicated than that for an explanation. Choosing the right server begins before the hardware and software is even considered. Making the correct choices and implementing proper solutions begins with assessing business needs and requirements now and into the foreseeable future.

opteron_pair_low_angle

Why use a server?

A server can be thought of as a device that delivers and stores data. If two PC users connect to a third for printer sharing or common files then that PC is a server. It serves the other two. The theory is that simple. Servers can streamline a business by providing a common area for files to be stored and shared. A server can provide for common access to other hardware; printers for example. A server can ensure all users are working on the same page, with the same information.

Centralizing common data and tasks has several benefits.

  • Users can access the most current data which can be maintained in a central location rather than multiple, independent locations thus ensuring the most up to date data is available.
  • Data integrity is maintained through redundant systems or backups. The single server can be backed up rather than multiple desktops.
  • Data can be made available to authorized users within the office and remotely.
  • Data access can be controlled. Privileges can be assigned to users.
  • A server can support a website.
  • A server can support communication applications such as email, faxes, online video or shared application meetings.
  • A server can make the same application available to multiple users.

Believe it or not a server can be a cost savings device and a revenue generating device all at the same time. If 10 users connect to a server to use one printer then that saves the cost of purchasing 10 printers.

Security is an immensely large issue for any business today. There are software updates, virus updates and data preservation issues to contend with. A server administrator can ensure that these issues are dealt with on the server level. User systems can be automatically updated when users log on. Inbound and outbound email can be scanned at the server level reducing downtime from virus infections. The time saved is better spent maintaining and optimizing the entire network to ensure that a business can continue to create revenue.

These are the most rudimentary of examples but they begin to reveal the benefits of a server.

mobo_ram_proc_in

What does your business look like?

Choosing the right server begins with assessing business needs now and into the foreseeable future from users both inside and outside of the office.

  • How many users will be connected to the server at any given time?
  • What will users access the server for?
  • File storage
  • Email
  • Applications
  • File sharing
  • Access to printers
  • Database support
  • Remote access or just in the office?
  • Will the users be sharing or interacting with large or small files and how often?
  • Will users access or share large or small files?
  • Will the serve be expected to support website then what type of website is it?
  • How many users at any given time for each of the tasks expected of the server?

There is no “I don’t know” excuse for any business when it comes to these questions. The answers are readily accessible and require no special IT knowledge and it’s forgivable if some answers are accurate while others may be approximations.

Example

If a business has 5 employees and the goal is double the size of the operation in 2 years then the server should be adequate to support the goal. The relevant numbers are 5 employees now and 10 employees later. If those employees require access to common files both inside the office and remotely then make note of that. Make a further note if they are to have email both in and out of the office and have access to files. If the business has a large database of products then that too is an important point to note.

This “what”, “when” and “how” of a business is essential information used to help define the path towards the correct server.

How is a server different from a desktop PC?

A server can be divided into two main areas; hardware and software. Server hardware and software are tailored to meet the demands put on a server which are uniquely different from that of a desktop PC.

A desktop PC is optimized to one user working with one program at any given time. Typically a program is opened, used and then shut down. If a program, such as an email client, is left open while another program is in use then the background program is exactly that; residing in the background. The computer allocates only the minimal amount of resources to the background program and turns the majority of its attention to the active program.

If a user runs several tasks simultaneously such as listening to music, burning a CD and surfing the web then limitations of a desktop PC materialize. For example, the computer becomes slower or the music stutters. Faster processor speeds have allowed for more to be done but it is only as a single user.

Servers handle multiple users simultaneously.

Remember that servers “serve”. Servers store, direct and deliver data to where it is needed. A server does not allocate its resources to displaying or assembling that data. It may deliver information or even part of a program but it is the desktop PC that will carry the ball the rest of the way.

Server software is written for specific tasks and can use different operating systems. These operating systems usually are geared towards the task of supporting installed software rather than fancy displays (GUI). There is software for fax servers, databases, file storage and so on.

This comes full circle to the question of “could a desktop PC be a server?” The answer is yes with the appropriate software but it will be limited in the amount of users or tasks it must support. This is important to remember as this helps debunk the cost myth.

The confusion lies in the comparison of a desktop PC costing only hundreds of dollars to that of a full fledged server costing thousands. The truth of the matter is that any desktop PC can be configured to act as a server but it may not be able to perform as well as a server in specific situations.

How much to spend?

Cost is top of mind with most businesses. Remember that a server, at this stage, is just computer hardware and software that “serves” information to others. Without a good understanding of needs and requirements then no one can say if a $500 budget is best or a $5000 budget. Assistance to choose the correct hardware and software and the installation, setup, configuration and optimization of it requires the training and experience found with an IT professional.

Unfortunately there is no fill-in-the-blank spreadsheet. There is no graph that compares processor speed to amount of email users. There is no formula for amount of RAM versus applications used beyond the minimum requirements stated by the manufacturer for the particular software used. The answers to questions like “how fast a processor” or “how much memory?” is one part budget, one part knowledge of current requirements, one part expected growth and one part IT professional.

The bottom line is preparation can result in matching a server correctly to the business. An office of five does not require a top end mainframe server to support email. They may well do just fine on hardware that costs less than $1000. The budget does increase when a bit more muscle is required but not as much as one would think.

What is wrong with the old server?

One of the great frustrations any business owner or manager encounters is questioning what is wrong with the old server when confronted by the IT person who is recommending a new server. The “old server” isn’t physically broken so why does it need replacing? There could be several reasons. The amount of users the server supports may have increased. The complexity of software has increased. The old server may just not be scalable.

Scalability is important to business. A scalable server holds promise that the hardware bought today won’t have to be replaced tomorrow or can be upgraded to meet increased demands at a fraction of the original purchase cost.

If, a year or three later, a simple upgrade of a processor will maintain server productivity then the common sense approach would be to upgrade the processor. It is far less expensive than replacing the entire server. A business does not want to find out that the processor is no longer made or isn’t available for their hardware, specifically the motherboard.

Scalability is the ability to upgrade components or add new components to grow the server with a business. Scalability should always be a primary consideration.

Understanding server performance

A server can be personified to get an idea of performance. Picture yourself as the mail person in an office of 20 people. You are in charge of distributing inbound mail to the correct recipient and processing the outbound mail. It isn’t too taxing a job and there are slow and busy times but nothing you can’t handle.

An energetic person would dash about the office delivering and picking up mail at a swift rate. The faster processors accomplish the tasks quicker.

The next level is an increase to 50 users. The job becomes more demanding. Then a new job is added of retrieving and filing data files from the basement. A faster processor will get the task moving through the system but it won’t solve all of the problems with delays. A second person is needed. Add to that the fact that the filing cabinets are older and the drawers sometimes stick. The bottlenecks begin to appear.

Without going too much further the puzzle begins to take shape. The sum of all the parts aids in determining server configuration. What the server is being used for can determine if the emphasis should be on processor speed, disk speed, RAM or all of it.

What server hardware?

Sometimes the best example is something to compare to. Everyone doesn’t launch into this in a blaze of glory. Sometimes it’s a cough, sputter and spurt. So was our own humble beginnings here at Short-Media.com. We had cobbled together enough bits and pieces from various systems to resurrect a working server. The word “working” was used loosely during the first 6 months. The server often rolled over onto its back due to some nameless writer “borking” it. Sometimes it was some stray piece of code that worked a month ago that has suddenly become evil only to lock up the entire database. Most times it was just by looking at the hardware the wrong way.

The server temporarily resided in the host provider’s office so the team could keep an eye on it. Yes it sat on a chair because many were afraid to move it when it finally was working. It was anything but pretty.

ws_server_chair

cu_server_chair

Short-Media.com resided on a server powered by single 1000 MHz PIII processor and 512 MB of RAM. The site database was stored on an EIDE hard drive. There was a steep learning curve and Short-Media.com evolved through various designs based on popular CMS software. The amount of traffic on the server soon surpassed the upper end of what the hardware could deliver. This is where the IT professionals paid off in spades. We bought ourselves some time by custom coding the entire site from the ground up. This “lean and mean” code provided a bit of headroom…as long as the traffic load did not increase any further.

Lesson one; don’t buy a server to meet your needs. Buy a server to exceed your needs.

No businessperson with a vested interest wants to break even, stay flat or just make budget. A success driven businessperson wants to grow the business to increase revenue. The Short-Media.com current hardware had reached its limit. There was no growth left in the hardware.

Our road to a new server began with assessing our needs and requirements now and for the future. At the time this article was written Short-Media.com saw over half a million visitors every month resulting in over 23 million hits to our server. Our goal was to triple the amount of visitors over the next year. We wanted to serve our current readership and forum members faster but we also wanted to ability to maintain that speed with increased traffic. Short-Media.com is built on CPU and disk intensive software. The database is under constant queries from users and a substantial load would create havoc with the old system. Technology changes had provided the performance increase but there was a new word that crept into our day to day language; scalability.

The Processors

opteron_850_pair

The choice was to build the server around the AMD Opteron processor. How an AMD Opteron processor is different from the competition involves technical debabblefication which is always a challenge. AMD Opteron processor technology is a extensive and a conversation better left to a gaggle of excitable geeks. Most business people prefer concise recommendations with supporting data at hand if required. The AMD Opteron processor made for an easy choice and here’s concisely why.

  • 32-bit/64-bit capable processor designed for servers. Something INTEL is yet to produce for the mainstream PC or server market.
  • Simultaneous 32- and 64-bit computing capabilities.
  • Offered in three series: the 100 series (1-way), the 200 series (Up to 2-way), and the 800 series (Up to 8-way).
  • Low-power processors in HE (55 Watt) and EE (30 Watt) versions available.
  • Can address up to 256 Terabytes of system memory in 64-bit mode. 32-bit servers typically have a ceiling of 4 GB of system memory. More software and data residing in memory equates to faster data delivery to the user.
  • Scalable up to 8-way servers and workstations.
  • Favorable price/performance ratio.
  • HyperTransport Technology
  • Integrated Memory Controller for decreased latency between system memory and processor. Less latency means fewer times the system is “waiting” for data and thus holding up users.

64-bit processing is here. 64-bit systems can address (use) more memory than 32-bit systems. The mathematical maximum addressable memory in a 64-bit system is 262,144 Gigabytes. The maximum addressable memory in a 32-bit system is 4 Gigabytes.

AMD 64-bit processors are also capable of computing in 32-bit and 64-bit modes simultaneously. This is another example of the word “scalability”. Software that isn’t yet available in 64-bit versions can run in 32-bit mode on a 32-bit or 64-bit operating system. The Short-Media.com site will be built on a 64-bit sever operating system and other 64-bit software will be integrated as it becomes available. The hardware platform grows with the support needs.

opteron_850_single

AMD Opteron models

X—indicates the maximum scalability of the processor. In other words:
100 Series = 1-way servers and workstations
200 Series = Up to 2-way servers and workstations
800 Series = Up to 8-way servers and workstations

YY—indicates relative performance within the series. In other words, an AMD Opteron Processor Model 244 is higher performing than an AMD Opteron Processor Model 242, etc.

Low-power AMD Opteron Processors are identified by a two character modifier, where:

HE—indicates a 55W processor within the series.

EE—indicates a 30W processor within the series.

cellpadding="2" width="450" align="center" border="4">
style="COLOR: white; font-weiht: bold"
width="40%" bgcolor="#006633">Series
style="COLOR: white; font-weiht: bold"
width="20%" bgcolor="#006633">
100 Series
style="COLOR: white; font-weiht: bold"
width="20%" bgcolor="#006633">
200 Series
style="COLOR: white; font-weiht: bold"
width="20%" bgcolor="#006633">
800 Series
Scalability
1-way
Up to 2-way
Up to 8-way
Performance
href="http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_8796_8800,00.html">AMD Opteron processor benchmarks
Frequency
Model Numbers
1.4GHz
Model 140
Model 240
Model 840
1.6GHz
Model 142
Model 242
Model 842
1.8GHz
Model 144
Model 244
Model 844
2.0GHz
Model 146
Model 246
Model 846
2.2GHz
Model 148
Model 248
Model 848
2.4GHz
Model 150
Model 250
Model 850
Low-Power Options
EE (30W) Frequency
Model Numbers
1.4GHz Model 140 EE Model 240 EE Model 840 EE
HE (55W) Frequency
Model Numbers
2.0GHz Model 146 HE Model 246 HE Model 846 HE
Integrated memory controller
Yes
Yes
Yes

Memory controller width

128-bit
128-bit
128-bit

ECC DRAM protection

Yes
Yes
Yes
HyperTransport™ Technology
Yes
Yes
Yes

HyperTransport Links (total/coherent)

3/0
3/1
3/3

HyperTransport Link width

16 bits x
16 bits
16 bits x
16 bits
16 bits x
16 bits

HyperTransport bus frequency

800MHz
800MHz
800MHz
AMD64
Yes
Yes
Yes

Simultaneous 32 & 64-bit computing

Yes
Yes
Yes

L1 Cache Size (data/instruction)

64KB/
64KB
64KB/
64KB
64KB/
64KB

L2 Cache Size

1MB
1MB
1MB

Pipeline stages (integer/floating point)

12/17
12/17
12/17

L1/L2 data cache protection

ECC
ECC
ECC

L1/L2 instruction cache protection

Parity
Parity
Parity

Global History Counter Entries

16K
16K
16K

L1 TLB entries (data/instruction)

40/40
40/40
40/40

L1 associativity (data/instruction)

Full/Full
Full/Full
Full/Full

L2 TLB entries (data/instruction)

512/512
512/512
512/512

L2 associativity (data/instruction)

4-way /
4-way
4-way /
4-way
4-way
/ 4-way
Direct Connect Architecture
Yes
Yes
Yes
Process
.13 micron SOI
.13 micron SOI
.13 micron SOI
Manufactured In
Fab 30, Dresden Germany
Fab 30, Dresden Germany
Fab 30, Dresden Germany

Source:

AMD Opteron™ Processor Model Numbers and Features Comparison

AMD implemented HyperTransport technology and integrated the memory controller on die. HyperTransport technology is, very simply, a very fast connection between two chips. Performance benefits have been most notable between system RAM and processor. HyperTransport links can also connect processor to processor. These high speed connections, designed to compliment the new PCI-Express standard, between processors and system RAM mean more users can get to the information they require faster.

A car on a highway would be a good analogy. Data moves around a processor core quite quickly. Intel may have their car wound up to a higher RPM but that’s only on it’s home track…inside the processor core. Once the data hits the outside world, the front size bus, the speed limit drops down to 800 MHz or less meanwhile AMD hits the road at up to 2.4 GHz. Everywhere else the Opteron provides for a higher speed limit on roads that have two to four times as many lanes for traffic.

opteron_pair_low_angle

Competitive Comparison

width="500" border="4" align="center" cellpadding="2" cellspacing="0" bordercolor="#6699CC" class="copysm">
cellspacing="0" cellpadding="0" width="100%"
border="1">
width="450" border="0" cellpadding="2" cellspacing="0" class="copysm">
color="#ffffff">Server System Comparison
color="#ffffff">AMD Opteron™
color="#ffffff">Intel Xeon1
color="#ffffff">Intel Xeon2
color="#ffffff">Intel Xeon MP3
color="#ffffff">Intel Itanium 24
colspan="6"> src="Competitive Comparison_files/1x1black.gif"
width="100%" align="left" border="0" />
class="copylg">Modular, glueless scalability align="middle" bgcolor="#ffffff">
Yes
align="middle"
bgcolor="#ffffff">Requires Northbridge
align="middle"
bgcolor="#ffffff">Requires Northbridge
align="middle"
bgcolor="#ffffff">Requires Northbridge
align="middle"
bgcolor="#ffffff">Requires Northbridge
bgcolor="#ffffff">SMP Capabilities align="middle" bgcolor="#ffffff">Up to 8-way align="middle" bgcolor="#ffffff">Up to 2-way align="middle" bgcolor="#ffffff">up to 2-way align="middle" bgcolor="#ffffff">Up to 4-way align="middle" bgcolor="#ffffff">Up to 4-way
bgcolor="#ffffff">Direct Connect Architecture bgcolor="#ffffff">Yes bgcolor="#ffffff">No bgcolor="#ffffff">No bgcolor="#ffffff">No bgcolor="#ffffff">No
High Performance 32-bit and 64-bit computing

align="middle"
bgcolor="#ffffff">AMD64
align="middle"
bgcolor="#ffffff">No
align="middle"
bgcolor="#ffffff">EM64T
align="middle"
bgcolor="#ffffff">No
align="middle"
bgcolor="#ffffff">No
class="copylg">HyperTransport™ technology

align="middle"
bgcolor="#ffffff">Yes
align="middle"
bgcolor="#ffffff">No
align="middle" bgcolor="#ffffff">No align="middle"
bgcolor="#ffffff">No
align="middle"
bgcolor="#ffffff">No
class="copylg">Integrated DDR memory controller

align="middle"
bgcolor="#ffffff">Yes
align="middle"
bgcolor="#ffffff">No
align="middle" bgcolor="#ffffff">No align="middle"
bgcolor="#ffffff">No
align="middle"
bgcolor="#ffffff">No
width="100%" align="left" border="0" />
class="copylg">Front Side Bus frequency
1.4 – 2.4 GHz†
533 MHz
800 MHz
400 MHz
400 MHz
class="copylg">Front Side Bus bandwidth
11.2 – 19.2 GB/s†
4.3 GB/s
6.4 GB/s
3.2 GB/s
6.4 GB/s
class="copylg">Maximum Inter-processor bandwidth
6.4 GB/s
4.3 GB/s
6.4 GB/s
3.2 GB/s
6.4 GB/s
class="copylg">Memory support
DDR266/333/400
DDR266
DDR333/DDR2-400
DDR200
DDR200
class="copylg">Memory Bandwidth 2P System
12.8 GB/s††
4.3 GB/s
6.4 GB/s
6.4 GB/s
6.4 GB/s
class="copylg">Memory Bandwidth 4P System
25.6 GB/s†††
N/A
N/A
6.4 GB/s
6.4 GB/s
colspan="6"> align="left" border="0" />
bgcolor="#ffffff">L1 cache size (max)

align="middle" bgcolor="#ffffff">128 KB align="middle" bgcolor="#ffffff">8KB + 12k mop align="middle" bgcolor="#ffffff">16KB + 12k mop align="middle" bgcolor="#ffffff">8KB + 12k mop align="middle" bgcolor="#ffffff">32 KB
L2 cache size (max)

align="middle" bgcolor="#ffffff">1 MB align="middle" bgcolor="#ffffff">512 KB align="middle" bgcolor="#ffffff">1 MB align="middle" bgcolor="#ffffff">512 KB align="middle" bgcolor="#ffffff">256 KB
L3 cache size (max)

align="middle"
bgcolor="#ffffff">N/A
align="middle" bgcolor="#ffffff">2 MB align="middle"
bgcolor="#ffffff">N/A
align="middle" bgcolor="#ffffff">4 MB align="middle" bgcolor="#ffffff">6 MB
class="copylg">Maximum I/O bandwidth 2P System
12.8 GB/s††
3.2 GB/s
12.3 GB/s
4.8 GB/s
6.4 GB/s
class="copylg">Maximum I/O bandwidth 4P System
25.6 GB/s†††
N/A
N/A
4.8 GB/s
6.4 GB/s
class="copylg">SIMD Instruction Set Support
SSE, SSE2
SSE, SSE2
SSE, SSE2, SSE3
SSE, SSE2
N/A
width="100%" align="left"
border="0" />
width="100%" border="0">
width="5%">† width="95%">The front side bus (interface to memory) of the AMD Opteron™ processor runs at the speed of the processor
†† AMD 2P System – AMD Opteron™ processor 200 series with 1 HyperTransport™ Inter-processor Bus and 2 HyperTransport™ I/O Buses with DDR400 memory
††† AMD 4P System – AMD Opteron™ processor 800 series with 4 HyperTransport™ Inter-processor Buses and 4 HyperTransport™ I/O Buses with DDR400 memory
1) With Intel E7501 Chipset ( href="http://developer.intel.com/design/chipsets/e7501/"
target="new">http://developer.intel.com/design/chipsets/e7501/)
2) With Intel E7520 Chipset ( href="http://developer.intel.com/design/chipsets/e7520_e7320/"
target="_blank">http://developer.intel.com/design/chipsets/e7520_e7320/)
3) With ServerWorks GC-HE chipset ( href="http://www.serverworks.com/products/GCHE.html"
target="new">http://www.serverworks.com/products/GCHE.html)
4) With Intel E8870 chipset ( href="http://developer.intel.com/design/chipsets/e8870/#features"
target="new">http://developer.intel.com/design/chipsets/e8870/#features)
border="0">
class="copysm" color="#000000">Dedicated Bandwidth
Shared Bandwidth

source:

AMD Opteron™ Processor Server Competitive Comparison

The Opteron processor is scalable for our needs. The Opteron 850 series allows for expansion to a 4P system. This would mean a purchase of a 4P motherboard but the 2P motherboard currently used by the two Opteron 850 processors could be refitted with two Opteron 2xx processors and used to support other sections of the site. Dividing tasks between multiple servers is required in high load environments and adapting hardware for new use makes financial sense.

From the OS up Short-Media is planned as 64-bit which is supported by the Opteron but will simultaneously operate any future software if it is still 32-bit. This backwards-forwards compatibility is extremely beneficial in the transition time between 32-bit and 64-bit software.

Opteron technology supports fast links and large bandwidth connections to the rest of the computer. This is on target with our own needs to handle a lot of high traffic situations where thousands or tens of thousands of users are requesting data near simultaneously.

Bottom line is that the Opteron processor provides the heart of a scalable system with room for growth.

System Memory

How much RAM is enough? System memory and the path between the system memory and the processor are much faster than the pathway between the hard drives and the rest of the system. The overall performance increases with the amount of data that can be pulled up into system memory. A server that supports a database, found in a website or information records company for example, would greatly benefit from more system memory. The users may be accessing different parts of the database but the database itself is changing very little if at all. Information will be retrieved at a significantly higher rate since the pathways between processor and RAM are much faster than looking to the disk for the database information. An IT professional may be able to advise if the emphasis should go to processor power or system RAM given the type of data retrieval or application support the server is required to do. If the server supports an application that requires a lot of data and then calculates new variables via user input then the emphasis may be on both.

ws_memory

Crucial memory is peace of mind. It is because Crucial is one of the most dependable brands in the world. It is dependably compatible. We’ve found that Crucial has worked where other brands have not. Crucial also comes with a limited lifetime warranty. If a module fails, for any reason during normal use, the module will be replaced for the original purchaser. Memory can have defects or fail. It is not a common occurrence but when it does happen it will be replaced at no cost except for the cost of shipping the failed module back to Crucial. 1 and 2 GB memory modules are an investment.

cu_crucial_sticker

The Platform

Server motherboards are different from desktop motherboards. A desktop motherboard is built around supporting the single user. It is an endpoint. A server motherboard is built around supporting many other users that will connect to it. A server motherboard will support multiple processors. It is not designed to support powerful gaming video cards but it will most likely have rudimentary onboard video capabilities. It may have onboard support for multiple drives or connections that support advanced storage controllers. The TYAN Thunder K8SR provided a stable and reliable motherboard platorm.

The Adaptec 7902W SCSI controller integrates dual channel hard drive support right onto the motherboard.

adaptec_aic_7902w_chip

A brief lesson in RAID storage

Storage is another main component in the server. It consists of the hard drives and the controllers. Server configurations can make use of a RAID system for hard drive support. RAID is a topic within itself but the most simple of explanations is two or more hard drives that work together or in support of one another. RAID means Redundant Array of independent Disks. There are many RAID configurations but are four are most common.

RAID 0 – two or more disks work together. Data is broken down into segments equal to the amount of disks in the RAID. RAID 0 systems provide for quick writing and reading of information by spreading the data across all the drives. RAID 0 would be useful in digital media environments speed of writing and reading large amounts of data is required.

RAID 1 – A secondary disk is the mirror image of the primary and is constantly update. RAID 1 is used in environments where data preservation is a priority. If the primary disk fails then the secondary can take over with no data loss. Financial institutions, accounting firms and other such data sensitive businesses would look to RAID 1. Data integrity would be further preserved with a secondary data backup system such a tape drive.

RAID 0+1 – Two RAID 0 arrays are used in a RAID 1 configuration. Disk access is maintained with redundancy. Website severs would be most suited to RAID 0+1 configurations. Disk access is not as fast as RAID 0 but faster that RAID 1. Data could be retrieved at a faster rate but there is the comfort of a mirror RAID if ever the primary should fail. Website traffic interruption would be minimal in the event of a disk failure.

JBOD - Just a Bunch of Disks. A JBOD setup may be used in an environment where many users store information on a network server but do not require high data access speed. A backup system such as a tape drive may provide for data preservation. The AIC-7902W could support up to 15 SCSI drives per channel. That’s two cables with 15 drives each for a total of 30 SCSI drives.

Server motherboards usually come with onboard video support. Most of the time the server is not driving a monitor. A monitor is only used during initial setup or to “monitor” the system periodically in a data centre. Onboard video means there is no requirement for a separate video card thus saving height. A host or data center charges by case height for rackmounted collocated servers and, of course, bandwidth usage.

The ATI RAGE XL is a lightweight compared to its gaming cousins tipping the graphics clock at only 83 MHz, 125 MHz for memory clock and a maximum of 8 MB of external memory. This is onboard server video and it does what it has to and no more.

ati_ragexl_video_chip

The most common rackmount server cases come in four height sizes measured in “U” (1U, 2U, 3U and 4U). A host will charge, in addition to bandwidth usage, by the height of the server. 1U server cases have the least amount of height at approximately 1.7 inches (43 mm.) 4U chassis are approximately 7 inches (178 mm.). 1U cases may be more practical for high density server applications including collocation. 2U cases may provide more thermal control better suited to servers that are I/O intensive in high density areas. A 3U chassis may be better suited for web servers, back office servers, firewall, proxy server and gateway applications. 4U servers provide more room for drive storage making them practical for video servers or large data storage.

The goal should be to determine what is expected of the server before choosing the components. A 1U server provides sufficient growth headroom at the most economical rack rate. It is our future plan to add a second server to share the website traffic load. The database could be run from one server and the code interface from the other. This would provide a better performance boost than an upgrade to the processors.

ws_empty_case

case_height

ws_case_full

Opteron 850 Server (New)

  • 2 x AMD Opteron 850 processors
  • Operating system: Red Hat Enterprise Linux 4.0 for AMD64
  • Seagate ST318453LW – 15KRPM U320 18gb HD x 2
  • Crucial PC3200 ECC REG 512mb
  • Tyan Thunder K8SR (S2881) motherboard
  • Custom low profile passive copper coolers
  • 1U Rackmount case
  • 400w 1U EPS PSU

DELL Poweredge 350 (Old)

  • Intel
    Pentium III 1Ghz (Coppermine)
  • 440BX motherboard
  • 512 MB PC133 ECC Registered RAM
  • Western Digital WD200BB-18DEA0 20 Gb ATA hard drive
  • Western Digital WD1200JB-75CRA0 100 Gb ATA hard drive
  • Operating System: Red Hat Linux 9 (2.4.20-18.9)
  • 1U Rackmount case

RAMSpeed Benchmarks

RAMSpeed is an open-source utility to measure cache & memory performance. Only the Integer and Floating Point benchmarks were run of the 8 benchmarks available and each was run 5 times. Note that RAMspeed defines a megabyte as 2^20 bytes (1048576 bytes), while
other benchmarking titles may define it as a million bytes, thus showing a
little “better” performance. More is better in the results.

RAMSpeed Integer Benchmarks

intmem_01

intmem_02

intmem_03

intmem_04

intmem_05

intmem_longrun

RAMSpeed Floating Point Benchmarks

floatmem_01

floatmem_02

floatmem_03

floatmem_04

floatmem_05

floatmem_longrun

Whetstone Double Precision Benchmark in C/C++

The Whetstone benchmark measures computing power.

The tests include simple addition, subtraction, multiplication and division of floating point numbers together with trigonometric functions such as sine, cosine, tangent and exponents. More is better.

whetstone

BYTE UNIX Benchmarks (Version 3.11)

BYTE UNIX Benchmarks (Version 3.11) measure operating-system and disk-subsystem performance as well as raw CPU power. More is better in the results.

byte_01

byte_02

byte_03

byte_04

byte_05

byte_06

Short-Media PHP Benchmarking Tool

The Short-Media PHP Benchmark Tool (SM Bench) was written to measure PHP database performance through a series of tests which create and populate a database with 1,000,000 rows of data then randomly query, truncate, delete and copy 100,000 rows of data. In layman’s terms it measures database performance (and thus the speed of PHP website data) if a “whole lot of users were doing a whole lot of things at once”. In all cases less is better.

smbench_01

smbench_02

smbench_03

smbench_04

Another way to interpret the graphs is to show the percentage that the new server is faster than the old.

BYTE UNIX Benchmarks (Version 3.11) RESULT: OLD RESULT: NEW % faster
Arithmetic Test (type = double) 265871.40 1213104.40 456.3%
Dhrystone 2 without register variables 1927990.30 7663850.60 397.5%
Execl Throughput Test 1915.70 0.00 0.0%
File Copy (30 seconds) 22580.00 72156.00 319.6%
Pipe-based Context Switching Test 264265.50 0.00 0.0%
Shell scripts (8 concurrent) 53.00 1057.00 1994.3%
RAMspeed (UNIX) v2.3.1 16Gb per pass mode 5-benchmark INTmem LongRun mode
Benchmark 01 Mb/s Mb/s % faster
INTEGER Copy 324.66 1814.3 558.8%
INTEGER Scale 326.47 1854.82 568.1%
INTEGER Add 320.86 2215.9 690.6%
INTEGER Triad 276.99 2149.67 776.1%
Integer Average 312.25 2008.67 643.3%
Benchmark 02 Mb/s Mb/s % faster
INTEGER Copy 324.39 1814.45 559.3%
INTEGER Scale 326.93 1859.51 568.8%
INTEGER Add 320.94 2211.2 689.0%
INTEGER Triad 276.87 2154.46 778.1%
Integer Average 312.28 2009.91 643.6%
Benchmark 03 Mb/s Mb/s % faster
INTEGER Copy 324.24 1820.08 561.3%
INTEGER Scale 326.71 1858.8 568.9%
INTEGER Add 321.23 2211.45 688.4%
INTEGER Triad 276.71 2151.04 777.4%
Integer Average 312.22 2010.34 643.9%
Benchmark 04 Mb/s Mb/s % faster
INTEGER Copy 324.46 1814.65 559.3%
INTEGER Scale 327.10 1865.47 570.3%
INTEGER Add 321.10 2209.84 688.2%
INTEGER Triad 276.95 2143.53 774.0%
Integer Average 312.40 2008.37 642.9%
Benchmark 05 Mb/s Mb/s % faster
INTEGER Copy 324.85 1814.89 558.7%
INTEGER Scale 326.26 1855.97 568.9%
INTEGER Add 320.72 2217.73 691.5%
INTEGER Triad 276.44 2143.7 775.5%
Integer Average 312.07 2008.07 643.5%
Mb/s Mb/s % faster
INTEGER LongRun Copy 324.52 1815.67 559.5%
INTEGER LongRun Scale 326.69 1858.91 569.0%
INTEGER LongRun Add 320.97 2213.22 689.5%
INTEGER LongRun Triad 276.79 2148.48 776.2%
INTEGER LongRun Average 312.24 2009.07 643.4%
RAM Speed (UNIX) v2.3.1 16Gb per pass mode 5-benchmark FLOATmem LongRun mode
Benchmark 01 Mb/s Mb/s % faster
FL-Point Copy 334.21 1809.2 541.3%
FL-Point Scale 334.19 1872.76 560.4%
FL-Point Add 383.78 2178.99 567.8%
FL-Point Triad 384.61 2173.01 565.0%
FL-Point Average 359.20 2008.49 559.2%
Benchmark 02 Mb/s Mb/s % faster
FL-Point Copy 334.47 1814 542.4%
FL-Point Scale 334.25 1877.99 561.9%
FL-Point Add 384.07 2186.4 569.3%
FL-Point Triad 384.34 2174.31 565.7%
FL-Point Average 359.28 2013.18 560.3%
Benchmark 03 Mb/s Mb/s % faster
FL-Point Copy 334.19 1812.22 542.3%
FL-Point Scale 333.82 1882.51 563.9%
FL-Point Add 382.93 2186.08 570.9%
FL-Point Triad 384.14 2173.88 565.9%
FL-Point Average 358.77 2013.67 561.3%
Benchmark 04 Mb/s Mb/s % faster
FL-Point Copy 334.37 1813.51 542.4%
FL-Point Scale 333.55 1873.84 561.8%
FL-Point Add 384.10 2193.76 571.1%
FL-Point Triad 384.98 2177.14 565.5%
FL-Point Average 359.25 2014.56 560.8%
Benchmark 05 Mb/s Mb/s % faster
FL-Point Copy 334.77 1809.2 540.4%
FL-Point Scale 334.44 1874.22 560.4%
FL-Point Add 383.58 2180.81 568.5%
FL-Point Triad 385.31 2173.62 564.1%
FL-Point Average 359.53 2009.46 558.9%
Mb/s Mb/s % faster
FL-Point LongRun Copy 334.40 1811.63 541.8%
FL-Point LongRun Scale 334.05 1876.26 561.7%
FL-Point LongRun Add 383.69 2185.21 569.5%
FL-Point LongRun Triad 384.68 2174.39 565.2%
FL-Point LongRun Average 359.20 2011.87 560.1%
Whetstone Double Precision Benchmark in C/C++
MWIPS 603.063 2062.109 341.9%
Mflops1 316.696 705.706 222.8%
Mflops2 266.313 666.505 250.3%
Mflops3 87.043 338.441 388.8%
Cosmops 22.418 50.246 224.1%
Expmops 11.968 38.306 320.1%
Fixpmops 660.634 2107.343 319.0%
Ifmops 152.944 2388.163 1561.5%
Eqmops 100.074 892.809 892.1%
Short-Media PHP Benchmark Seconds Seconds % faster
Database creation 0.0012 0.0051

2.35%

Test Table 1 creation 0.0017 0.0061 27.9%
Test Table 2 creation 0.0012 0.0161 7.5%
Table population of 1 Million Rows random data 1653.00 167.35 987.8%
Query of 100,000 random rows 197.5200 3.8147 5177.9%
Iteration of 100,000 random rows 266.8200 4.3258 6168.1%
Deletion of 100,000 random rows 39.4930 7.2645 543.6%
Copy of 100,000 random rows 89.7740 10.7940 831.7%
Deletion of all rows from table 1 0.3233 0.0184 1757.1%
Deletion of all rows from table 2 0.0044 0.0020 220.0%
Deletion of tables and database 0.0119 0.0005 2380.0%

Closing Thoughts

A server is another crucial component of today’s business world. The most important step is to assess what needs must be met; know the job before choosing the tool. IT professionals are highly recommended to assist in selecting and installing the proper hardware. They can be indispensable when installing and configuring the chosen operating system and applications.

The benchmark comparison isn’t as extreme a comparison in hardware as one might think. Often a server in a business is archaic by technology standards. The upgrade has been put off for years. The IT person has been in a manager’s office time and time again nearly begging and pleading for the budget to purchase. They’ve talked about scalable servers to reduce future costs while still meeting the demands that increased traffic will bring. They’ve used words like “better”, “faster” and “do more” which the present equipment is doing none of but there is a solution. A new server can boost performance not in the 10s of percentage points but often in the 100s if not 1000s of percentage points.

opteron_850_pair

A server is a tool; a tool that can be used to communicate with your customers. It’s hardware and software to put information at someone’s fingertips as fast as it can be typed. It connects person to person and products to people. It can increase the bottom line. Servers can be planned to meet the demands of today with upgrade paths to the future.

cu_crucial_sticker

A server is also a specific tool for a specific job. It’s important to do research or employ the services of an IT professional to ensure all of the hardware components are compatible with each other and the software applications to be used. A server is often misunderstood. It is not seen every day like a wristwatch or toothbrush yet rarely does a day go by where a server is not connected to, used or queried. A server can make money for a business. It can make money by saving money but, more importantly, it can make money by serving more customers.

For a business without a server there may not be the requirement for a large capacity, top of the line performance server. It may be using a sledgehammer and spike when a thumbtack would do. However, many businesses overtax their underpowered servers and time after time it results in downtime and that means lost revenue.

Servers are no longer an enigma. They are a tool that, with proper research, can meet today’s business demands and scale for the needs of tomorrow.

Our thanks to AMD and Crucial for
their support of this and many other sites.

Comments

  1. primesuspect
    primesuspect The benchmark differences are absolutely insane. I mean, in some tests, the new server literally is ten times faster than the old.

    Working on the dual opteron platform has been a dream. The P3 performed admirably considering what we put the poor dear through, but in the end, it's nice to have some leg room ;)

    GREAT article, Doug.
  2. Omega65
    Omega65 Nice Article - That's a serious upgrade from a P3.

    MM - you forgot about the Opteron x52: (2.6ghz & SSE3) and the Opteron x65 (1.8ghz), x70 (2.0ghz), x75 (2.2ghz) Dual Core Series. on pg4
  3. Shorty
    Shorty I love our Opterons :D
  4. MediaMan
    MediaMan
    Omega65 wrote:
    You forgot about the Opteron x52: (2.6ghz & SSE3) and the Opteron x65 (1.8ghz), x70 (2.0ghz), x75 (2.2ghz) Dual Core Series. on pg4

    I left Dual Core out as the product is still very new to the server world at the time of this article. Dual Core is a variable in itself and a lot of long term data isn't available. The focus of the article wasn't so much WHICH product to buy but to encourage adoption of server technology or an uprgrade to current.

    A lot of which comes from personal experience as a manager myself who is encouraging those above me to see beyond the capital costs to the benefits. Servers are business tools many are reluctant to invest in and even more reluctant to upgrade or optimize.

Howdy, Stranger!

You found the friendliest gaming & tech geeks around. Say hello!