Test Result Summary of Standalone Running on KVM Hypervisor

The standalone SBR Carrier server is qualified to support Kernel-based Virtual Machine (KVM) hypervisor on a Red Hat Enterprise Linux 7.3 machine. The results shown in this section are tested with guest RHEL 7.3 VMs hosted on a RHEL 7.3 KVM hypervisor with the following hardware configurations:

• The host has 128 GB RAM, 32 CPUs with 8 cores per CPU, and 1.5 TB storage capacity. SBR Carrier 8.4 is running on the clients.

KVM Hypervisor Running with One VM and Varying CPU Counts

KVM Hypervisor Running with One VM and Varying CPU Counts

Table 46 describes the results of tests run on the KVM hypervisor running with one VM for local accounting–start only (no SQL, proxy, or directed accounting methods). The following results are tested with a thread configuration of 500.

On a VM, if the CPU count is increased, the session rate and CPU utilization are also increased as shown in Table 46. With the local accounting-start only load, the RADIUS process size increases with an increase in number of sessions in the in-memory session database.

Figure 11: KVM Hypervisor Running with One VM—Sessions per Second Versus CPU Utilization

KVM Hypervisor Running with Six VMs

KVM Hypervisor Running with Six VMs

Table 47 describes the results of tests run on the KVM hypervisor running with six VMs for local accounting only. Each VM is configured with 16 GB RAM and 4 CPUs with 4 Cores per CPU.

Table 48 describes the results of tests run on the KVM hypervisor running with six VMs for local accounting only. Each VM is configured with 16 GB RAM and 4 CPUs with 8 cores.

Figure 12 compares the results of VMs configured with 4 CPUs x 4 cores and 4 CPUs x 8 cores. Doubling the number of cores per CPU increases the throughput without greatly increasing the CPU utilization.

Figure 12: KVM Hypervisor Running with Six VMs: 4 CPUs x 4 Cores Versus 4 CPUs x 8 Cores

Accounting Start Only for Proxy

Accounting Start Only for Proxy

Table 49 describes the results of tests run in an environment where the proxy SBR Carrier is running on a VM and the target SBR Carrier is running on a separate physical machine. The following results are tested with an accounting thread configuration of 200, RecordLocally=1, and Block=1 (with Block=1, the accounting thread processes the proxy transaction; a separate thread is used for the transaction when Block=0).

⁷TPS = (Total accounting requests - Dropped accounting packets) / 600 seconds

⁸Sessions are stored in the .hst file.

Table 50 describes the results of tests run in an environment where both the proxy and target are hosted as VMs on the same physical machine. The following results are tested with an accounting thread configuration of 200, test duration of 16 minutes, and Block=1.

Comparing results of proxy and target that are hosted as VMs on the same physical machine (Table 50) and proxy on a VM and target on a separate physical machine (Table 49) provides approximately the same performance.