Scalability Test Results for Windows Server 2012 DHCP Server

Article
12/15/2012

Introduction

This blog article provides data from tests conducted on server hardware (from different vendors and with different configurations) using Windows Server 2012 that depict how the different DHCP server operations scale with increasing hardware capacity. It can be used for guidance in assessing scalability of systems running Windows Server 2012 DHCP Server and choosing the appropriate hardware configuration for running the DHCP server based on the size of the deployment for a desired level of performance.

These scale testing results were executed at the Microsoft Enterprise Engineering Center. The EEC is a world-class facility which works with customers and partners to validate pre-release and in-market Windows Server and System Center product scenarios.

Background

DHCP service is an essential part of any network infrastructure. An enterprise network needs a robust, reliable and scalable DHCP service. While planning an enterprise scale DHCP server deployment, some of the questions that an administrator planning the deployment needs to answer are:

What configuration would be most suitable?

What would be the maximum lease capacity for this configuration?

How much time would it take to perform operations like restore/backup/export/import?

How much additional load (scopes, leases) will I be able to put on the system with the increase in CPU/memory?

It is to help answer these questions that this testing was undertaken. By conducting a range of tests on hardware from vendors like Dell and HP running Windows Server 2012, some key performance metrics were observed. The recorded observations for these performance metrics give a reasonable idea about the impact of varying hardware configuration. They can be used in not only deciding the configuration to opt for while deploying Windows 2012 DHCP Server but also in realizing the capacity of already installed systems.

Test Plan and Results

Test Setup

The RC (Release Candidate) build of Windows Server 2012 was used for conducting the tests. 5 Server models (3 from HP, 2 from Dell) with different configurations were selected for the hardware:

Manufacturer	Model	Core Count	Core Speed	Processor Model	Memory	Disk Configuration	Network Adapter
Dell	PowerEdge R810	12	2GHz	Xeon X5650	262GB	10k SAS	Intel® Ethernet Converged Network Adapter X520-DA2 Onboard 1G
HP	BL465cG1-2	4	2.6 GHz	2222SE	8GB	10k SAS	HP NC373i Multifunction 1Gb Adapter
Dell	PowerEdge R710	16	2.66GHz	Xeon X6550	98GB	15k SAS	Intel® Ethernet Converged Network Adapter X520-DA2 Onboard 1G
HP	BL460c-G1	8	2.6GHz	Xeon E5430	16GB	15k SAS	HP NC373i Multifunction 1Gb Adapter HP NC326i PCIe Dual Port 1Gb Adapter
HP	BL680-G5	24	2.4GHz	Xeon E7450	128GB	15k SAS	HP NC373i Multifunction 1Gb Adapter HP NC326i PCIe Dual Port 1Gb Adapter

Table 1. Configurations of servers selected for conducting the tests

The servers were managed using RSAT (Remote Server Administration Tool) running on a system with the below configuration:

Manufacturer

Model

Core Count

Core Speed

Processor Model

Memory

Disk Configuration

Network Adapter

BL460c-G1

2.6GHz

Xeon E5430

16GB

15k SAS

HP NC373i Multifunction 1Gb Adapter

HP NC326i PCIe Dual Port 1Gb Adapter

Table 2. Configuration of system running RSAT

As an enterprise deployment will require high availability of DHCP server, the tests were carried out with the new DHCP failover feature deployed between 2 DHCP servers with identical hardware configuration. The servers were setup with 15 failover relationships in hot standby mode. The complete setup along with the system running RSAT tool (DHCP MMC, PowerShell) has been shown in the figure below.

Fig. 1. The setup of the servers on which the tests were conducted

The purpose of the tests was to identify the number of scopes that can be deployed on a DHCP server at these different hardware configurations with the DHCP server being able to handle a certain minimum client transaction rate. The goal was also to measure how the DHCP server management operations done via DHCP MMC, PowerShell as well as operations like backup/restore, export/import would perform. With a number of tests being involved, the testing was conducted on each of the servers in three phases.

Note: To bring down the number of variable parameters, all the scopes deployed on the DHCP server were of size /24 and the number of leases in each scope was 100.

Test Results

In the first phase, the aim was to identify the number of scopes that the DHCP server would be able to support at a client transaction rate of at least 500/sec. A client transaction is a request-response exchange between the DHCP server and a client which results in either a new lease acquisition, renewal of existing lease or release of lease. The transactions were performed using a tool that simulated a multi-client environment. The client simulation test tool was configured to send these transaction is the ratio of 20 (new lease): 60 (renew lease): 20 (release). This ensured that the number of leases in the DHCP server database remained constant as the new lease and release transactions would set off each other. The duration for which each server was subjected to client transactions was kept at 30 minutes. The results of these first phase of tests are given below:

Manufacturer	Model	Number of scopes with failover	Transactions/sec with failover
Dell	PowerEdge R810	50,000	656
HP	BL465cG1-2	10,000	591
Dell	PowerEdge R710	50,000	684
HP	BL460c-G1	15,000	564
HP	BL680-G5	50,000	733

Table 3. Results of phase 1 tests

One of the inferences which can be drawn with the above results is the multi-threaded architecture of the DHCP server enables scaling up of the DHCP server performance as the number of cores in the system increases. It was also noted, while conducting the tests that the memory foot print of the DHCP server remained limited. Thus, any increase in system memory did not lead to an increase in scope capacity or transaction rate on DHCP server.

In the second phase, the servers were maintained at the number of scopes they were able to achieve in the first phase. They were then tested for their performance on the time it took to perform the following routine operations:

List all scopes in MMC
List all scopes in PowerShell
List all leases in a scope in PowerShell

The results of phase 2 are shown below:

Manufacturer	Model	Number of scopes with failover	Time taken to
Manufacturer	Model	Number of scopes with failover	List all scopes in DHCP MMC	List all scopes in PowerShell	List all leases in a scope (PowerShell)
Dell	PowerEdge R810	50,000	58 Secs	4 Mins 10 Secs	< 1 Sec
HP	BL465 cG1-2	10,000	15 Secs	28 Secs	< 1 Sec
Dell	PowerEdge R710	50,000	47 Secs	3 Min 06 Secs	< 1 Sec
HP	BL460c-G1	15,000	21 Secs	44 Secs	< 1 Sec
HP	BL680-G5	50,000	48 Secs	3 Min 12 Secs	< 1 Sec

Table 4. Results of phase 2 tests

In the third phase, the DHCP servers were maintained at the number of scopes (and leases) that they were able to achieve in the first phase and were tested for their performance on the time it took to perform the following operations:

Export with lease data
Import with lease data
Export without lease data
Import without lease data
Backup
Restore

The results of phase 3 are shown below:

Manufacturer	Model	Number of scopes with failover	Time taken to
Manufacturer	Model	Number of scopes with failover	Export with leases	Import with leases	Export without leases	Import without leases	Backup	Restore
Dell	PowerEdge R810	50,000	4 Hours 18 Mins	29 Hours 40 Mins	59 Mins 40 Secs	2 Hours 56 mins	1 Mins 11 Secs	9 Mins 18 Secs
HP	BL465cG1-2	10,000	40 Mins	5 Hours 21 Mins	8 Mins	30 Mins	40 Secs	3 Mins 12 Secs
Dell	PowerEdge R710	50,000	4 Hours 02 Mins	27 Hours 36 Mins	52 Mins 21 Secs	2 Hours 32 Mins	42 Secs	7 Mins 27 Secs
HP	BL460c-G1	15,000	1 Hour 9 Mins	8 Hours 9 Mins	14 Mins	47 Mins	58 Secs	4 Mins 1 Secs
HP	BL680-G5	50,000	4 Hours 4 Mins	29 Hours 20 Mins	56 Mins	2 Hours 47 Mins	1 Mins	8 Mins 57 Secs

Table 5. Results of phase 3 tests

As can be observed from the test data of phase 2 and phase 3, the management operations are fastest on the system with highest CPU speed. However, the management operations do not benefit from higher number of cores.

Concluding Notes/Remarks

This blog article intends to serve as a guide to DHCP server deployment planning. The results of the tests here depict performance of DHCP server with different hardware configuration. The performance has been measured in terms of a number of key performance metrics that include the number of scopes and the client transaction rate that can be supported and also the time taken to perform DHCP server management operations.

Disclaimer: The tests were conducted using the RC (Release Candidate) build of Windows Server 2012. The test results may differ slightly for the RTM (Release to Market) build.