Advertisment

How Much Load Can your Server Take?

author-image
PCQ Bureau
New Update

Are you planning to buy a couple of servers for your organization? Then hold your decision for a while and read this article. It might help you make a better decision. Well seriously servers are very costly preposition, so while buying servers; you have to check-out various performance parameters, apart from the server specs. Here, we will discuss those parameters and how these parameters can be your best friend, while making a purchase decision. Server Benchmarking is a way that can help to check the performance parameters of a server before buying them. To help you on this we want to share our test experiences, so that you can make the right decision while buying a server. Moreover, we'll also talk about some of the most common server benchmarks out there, what they mean, and how to interpret their results.

Advertisment

What is a server benchmark? 



While in the process of buying a server, you must have asked quotations from various vendors, where they send product brochures containing detail specs about their products. These brochures also contain server benchmarking scores claiming that they are number one in the market. And this brings confusion in to CIOs mind what to buy and what not. So here we will see what these benchmarks and their scores should mean to you and should you really bother to look at them or not.

TPC Benchmarks



Transaction processing and database benchmarks are used to evaluate and measure computing functions and operations. A transaction could refer to a set of operations, which perform disk read/writes, Operating System calls, or some form of data transfer from one subsystem to another.

Direct Hit!
Applies to: CIOs and IT Managers
USP:

Identify a server of your requirement
Primary Link:

www.tpc.org, www.varitest.com,

www.mercury.com 
Google keywords:

server benchmarking 
Advertisment

Typically these benchmarks can measure transaction processing and database performance in terms of how many transactions a given system and database can perform per unit of time, like transactions per second or transactions per minute. The overall results of particular benchmark comes in-terms of unit price per dollar, so lesser the price per transaction better the server performance. 

TPC benchmarking makes sense in the areas, where you are dealing with transactions like a commercial exchange of goods, services, or money. 

Therefore, organizations into inventory controls (goods), airline reservations (services), or banking (money) prefers to have TPC benchmarking before deploying any kind of servers.

Advertisment

TPC produces different kinds of benchmarks for various kinds of transaction areas and some of these are as follows.

TPC-APP: This benchmark is used to stress out Web services and simulates business activities such as a B2B transactional 



application server operating in a 24x7 environment. Its workload comprises commercially available application server products, messaging products, and databases to simulate a typical business activity. The results were given in two metrics, Web SIPS (Service Interactions Per Second) per Application Server system and Total SIPS, which is the total number of SIPS for the entire tested configuration. Both these metrics are associated with a certain number of dollars and the final scores are drawn in-terms of dollar per SIPS. So, lesser the dollar price per SIPS, better the performance. 

Advertisment

TPC-A: TPC-A benchmark is typically used in OLTP (On Line Transaction Processing) applications. It measures the performance in update-intensive database operations. And the final results are given in-terms of TPS (Transactions Per Seconds) associated with a dollar ($/TPS). Lesser the dollar betters the performance. 

TPC-B: TPC-B benchmark is a contrast to TPC-A. It's not an OLTP benchmark but it's a typical database stress test. It performs significant amount of input/output operation and also measures the system and application time. In addition it checks the transaction integrity. Final scores are given in ($/TPS). 

TPC-C: TPC-C is another OLTP benchmark like TPC-A, but this benchmark performs more complex transactions, which stress mainly the database and server hardware. 

Advertisment

It simulates a complete computing environment where a huge number of users execute transactions against a database. These transactions include entering and delivering orders, making payments, checking the status of orders, and monitoring the stock level at the warehouses. Here, the performance metrics are expressed in tpmC (transaction-per-minute) and is further associated with dollars. Lower the Dollar per tpmC ratio, better is the performance. 

TPC-D: This is a decision-support benchmark focusing on small and mid-sized customers. It basically evaluates performance of various decision-support systems by the execution of sets of queries against a standard database under given controlled conditions. The performance is evaluated in Query-Per-Hour Performance metric (QphD@Size) and like the other benchmarks it is also associated with Dollar

($/QphD@Size). 

Average Response Time

for Dell Power Edge 800

Response Time (ms)

Number of Clients

Notice how the server's response time increases as the number of clients increase. The effect is drastic after 19 clients

Advertisment

TPC-W: This is typically meant to test the Web performance on a particular server; it simulates workload in a controlled e-commerce environment and performs activities of a business oriented transactional Web server. This includes multiple online browser sessions, generating dynamic pages with data access, online transactions and simultaneous execution of multiple transaction types. It gives you scores in terms of the number of WIPS (Web Interactions Processed per Second) further associated with dollar($/WIPS). Like other TPC benchmarks, lower the Dollar to WIPS ratio, better the performance. 

TPC benchmarks are meant for evaluating fairly big-sized servers and setting up testing infrastructure with TPC costs millions of dollars. 

That's why you can't test the server in-house. So before buying big servers from the big names, ask them to show the TPC performance on your server live. 

Advertisment

Net Bench



Net Bench is a benchmark designed for testing the performance of a server as a file server. This is the benchmark which we use to test the mid-range servers in our labs. 

This benchmark is very nice breed between synthetic and real-life benchmark because to test the server, it generates a number of tasks from a set of client machines and starts copying and editing most commonly used files like MS Office documents, Photoshop and other graphic manipulation application files such as Coral, etc which you must be using in our organization. it then starts measuring the performance of the server in different stages. 

The main objective of NetBench is to measure how well the server handles the I/O requests generated by a huge number of 32-bit client machines and plots the performance in different stages. This gives us a fair idea that about the maximum amount of load (in terms of number of users) that the server can easily survive. 

Throughput for 



Dell Power Edge 800

Throughput (Mbps)

Number of Clients

The server manages to provide good throughput till 19 clients. After that, its throughput starts dropping after 150 Mbps

The setup required to run this benchmark is not very simple and only few places in our country (including the Cybermedia Labs) have the capability to run this tests in a controlled manner. 

So if the vendor of the server is not able to show the performance of the server you are planning to buy, then you know where to get it checked from. We do test this kind of servers frequently and publish the performance and results in our magazine. But if you are planning to buy one and the vendor is ready to test the servers using theses benchmarks, then you can write to us and we would do it for you. 

Web Bench



Web Bench is a cousin of Net Bench and is used to test the performance of a server as a Web server. Actually you can do two things with Web Bench. Either compare multiple Web servers with standard server hardware or compare a set of server hardware with a standard Web server. This is another benchmark which we use to test servers in our labs. 

Web Bench comes with standard test suites which can produce two overall scores for the server. The first one is the number of requests per second which the server can take and the second one is the throughput as measured in bytes per second. So for example if we see a major drop in the throughput of the server, we can immediately predict that the bottleneck is the network interface of the server and nothing else. 

Web Bench provides both static and dynamic test suites which executes applications that actually run on the server. Not only this, you can easily create your own test suites. No matter which test suites you use, your PC clients must be running 



Win 95/98/NT/2000/XP, and the controller must be running Win NT/2000/XP. On the other hand, the server can run any OS with any Web server. 

In our test bed, we generally use a standard Win 2003 installation and Apache Web server to test the server hardware.

LoadRunner 



Unlike the other benchmarks LoadRunner is more of a Web application benchmark than a hardware benchmark. But when you talk about benchmarking, both the application and the hardware are equally involved. So if the results are poor, either of them can have a problem and need optimization. 

LoadRunner is a benchmark which generates thousands of virtual users who access and overload any given application. 



Then the performance of the application is measured in different stages and finally a report is generated. Using this report, one can figure out the exact bottlenecks in the application and hardware and can then optimize them. 

The unique thing about LoadRunner is that it is very much customizable and can test any application developed including ERP/CRM packages from Oracle, SAP, PeopleSoft and Seibel. 

Anindya Roy and Sanjay Majumder

Dell Power Edge 800

Dell Power Edge 800 is a server class machine meant for small businesses and remote-offices operations. It comes with a single P4 processor running at 3.2 GHz, 800 MHz front size bus, 2 MB L2 cache. Plus it has 512 MB DDR-2 RAM. All these are placed on an Intel E7221 Chipset server board. In the storage, it has two 36 GB hot-plug Ultra 320 SCSI hard drives. For connectivity, it has one GbPS built-in network card. The server also comes with a deployment kit that lets you install OS on the machine. Its deployment CD is a Linux bootable CD, which lets you configure the machine's RAID and drivers. In addition, it asks you the OS installation you want and also asks you OS parameters such as CD key, machine name, time zone, etc. Then you will be asked to put in the OS CD, where the deployment CD copies the installation files and once it's done, it will ask for a reboot. After rebooting, it will start the final installation process. Dell provides you with detailed installation and troubleshoot guide with its box. 

We have tested this machine using 19 P-4 client machines all connected over a Gbe network and tried to overload the server using Net Bench. This benchmark is meant to test the file serving capability of a server class machine. You can see the graphs of the test results on the next page. One graph measures the server's response time as the number of clients increases, while the other measures its throughput. The former increases and the latter decreases with increased load. It may seem like the server is getting overloaded too quickly using just 19 clients, but please note that all the clients have a Gigabit interface. In the real world, the clients still use 10/100 Mbps, so the server would be able to handle much higher number of clients. This makes it a pretty good machine for the audience it's targeted at. 

Price: Rs 71,999, Meant for: Small Business 



Key specs:
Single P-4 3.2GHz, 512 MB DDR, Ultra 320 SCSI hard drive and one GBPS built-in Ethernet 



Pros:
Easy to set up using its deployment kit; Cons: None 



Contact:
Dell India, Banglore.



Tel:
25068269

Advertisment