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.
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.
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:
Intel® Ethernet Converged Network Adapter X520-DA2 Onboard 1G
HP NC373i Multifunction 1Gb Adapter
HP NC326i PCIe Dual Port 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:
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.
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:
Number of scopes with failover
Transactions/sec with failover
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:
The results of phase 2 are shown below:
Time taken to
List all scopes in DHCP MMC
List all scopes in PowerShell
List all leases in a scope (PowerShell)
4 Mins 10 Secs
< 1 Sec
3 Min 06 Secs
3 Min 12 Secs
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:
The results of phase 3 are shown below:
Export with leases
Import with leases
Export without leases
Import without leases
4 Hours 18 Mins
29 Hours 40 Mins
59 Mins 40 Secs
2 Hours 56 mins
1 Mins 11 Secs
9 Mins 18 Secs
5 Hours 21 Mins
3 Mins 12 Secs
4 Hours 02 Mins
27 Hours 36 Mins
52 Mins 21 Secs
2 Hours 32 Mins
7 Mins 27 Secs
1 Hour 9 Mins
8 Hours 9 Mins
4 Mins 1 Secs
4 Hours 4 Mins
29 Hours 20 Mins
2 Hours 47 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.
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.