Revision as of 00:39, May 12, 2017 by Sschlich (talk | contribs)
Jump to: navigation, search

GIR-GVP Port Capacity Tests

Genesys conducted performance tests of various Genesys Interactive Recording (GIR) and GVP Capacities. This section contains test profiles, test results, and analysis.


Performance Comparison of Physical Server and Virtual Machines

Single Hex Core

With a single hex core CPU, Genesys recommends 200 ports as a reasonable peak port capacity on a physical server with a single X5670, assuming that all criteria have been met. 300 ports can be achieved with a three-VMs configuration of the same hardware, with a single X5675 (performance is slightly better than X5670). The graph below compares overall CPU usage:

Figure 1: Comparison of System Usage between Physical Server and VM from Single Hex Core


Memory usage for MCP scales linearly against port capacity:

Figure 2: Comparison of MCP Memory Usage between Physical Server and VM from Single Hex Core


The two graphs below compare the 95th percentile value of Max Jitter Buffer and Max Delta, tracking audio quality from a sample RTP stream:

Figure 3: Comparison of Max Jitter between Physical Server and VM from Single Hex Core
Figure 4: Comparison of Max Delta between Physical Server and VM from Single Hex Core


A strong correlation exists between Max Jitter Buffer and Max Delta, regarding audio quality. A physical server can meet all criteria when its port capacity is 200 or below. Port capacity that is between 200 and 220 may impact audio quality, since both Max Jitter buffer and Max Delta are just slightly beyond the passing criteria. You can consider 220 as peak performance, if audio quality is not strictly required and can be waived. However, when port capacity reaches 230 or beyond, the two values become so big that there is apparent audio quality impact.

For VM configuration: Preferred/Recommended = 300 ports; Peak Port Capacity = 360 ports. With 390 ports, overall system CPU usage is 97%, close enough to 100% that it also observed audio quality impact.

Below are two tables of IOPS for the above two configurations:

Table 1: Disk IOPS of system level from a physical server with a single hex core

Ports Disk IOPS Physical Server
Total Reads Writes
60 11.13 0.001 11.13
120 21.82 0.001 21.82
180 32.03 0.001 32.03
200 34.95 0.001 34.95
210 36.53 0.001 36.53
220 37.76 0.001 37.76
230 39.48 0.001 39.48
240 43.33 0.002 43.33

Table 2: Disk IOPS of sum of all VMs of single hex core

Ports Disk IOPS VMs Overall
Total Reads Writes
120 20.68 0.101 20.58
240 36.29 0.070 36.22
270 41.39 0.087 41.30
300 45.57 0.065 45.50
330 48.85 0.000 48.85
360 51.69 0.000 51.69
390 57.82 0.002 57.82

Disk IOPS in Disk IOPS of sum of all VMs of single hex core table is the sum of Disk IOPS from all VMs. Also, IOPS is measured from each VM and then totaled, to determine overall IOPS. The same method is applied to all Disk IO calculations for VM environments in this series of tests.

Also in the above two tables, the IOPS Reads value is quite small because most of the operations are Writes.

The graph below compares the two IOPS tables above:

Figure 5: Comparison of System Disk IOPS Physical Server and VM from Single Hex Core
  • System level disk IOPS is scaling linearly against port capacity for both physical server and virtual machines.
  • SSD is only used on VM env as cache folder of MCP recording while SAS HDD drive is used to installed OS and MCP.

Dual Hex Cores

With a host of dual hex core CPUs (2x X5675@3.06GHz) with 32 GB RAM, we also compare the results from physical server and VM env. In VM env, on same hardware spec, 3 VMs are configured with 4 vCPU and 8 GB RAM assigned to each VM. Only one MCP installed on each VM and a SSD partition is used as cache folder for MCP recording.

The graph below depicts the overall system CPU usage:

Figure 6: Comparison of System Usage between Physical Server and VM from Dual Hex Cores


The next two graphs show 95 percentile values of Max Jitter and Max Delta from sample RTP stream quality analysis:

Figure 7: Comparison of Max Jitter between Physical Server and VM from Dual Hex Cores
Figure 8: Comparison of Max Delta between Physical Server and VM from Dual Hex Cores


The two tables below show:

  • Disk IOPS at system level on a physical server.
    and
  • Disk IOPS at system level on a VM environment.

Table 3: Disk IOPS at system level from physical server of dual hex cores

Ports Disk IOPS Physical Server
Total Reads Writes
50 9.069 0.000 9.07
100 18.587 0.000 18.59
150 28.598 0.001 28.60
200 37.460 0.001 37.46
240 41.290 0.003 41.29
280 49.031 0.020 49.01
330 53.373 0.001 53.37
350 53.150 0.001 53.15
380 61.456 0.001 61.46
400 64.123 0.001 64.12


Table 4: Disk IOPS of sum of all 3 VMs of dual hex cores

Ports Overall Disk IOPS
Total Reads Writes
120 22.38 0.024 22.35
240 38.99 0.012 38.97
300 48.60 0.017 48.59
360 56.05 0.047 56.00
390 60.24 0.002 60.24
420 64.59 0.028 64.57

The graph below compares the above two tables above:

Figure 9: Comparison of System Disk IOPS Physical Server and VM from Dual Hex Cores


Performance Comparison of Different Virtual Machines Configurations

Overall CPU usage on a physical server beyond peak port capacity is actually higher than overall CPU usage on virtual machines, while audio quality actually shows a quick downfall on a physical server. So the splitting the load into multiple MCPs in a VM environment will definitely take advantage of hardware resources and will achieve high port capacity with fewer audio quality concerns. There are three different VM configurations on the same hardware spec (counting the dual hex cores, 12 vCPUs in total) that are used for this purpose:

  • 3 VMs in total, 4 vCPU are assigned to each VM, only one MCP installed on one VM.
  • 4 VMs in total, 3 vCPU are assigned to each VM, only one MCP installed on one VM.
  • 6 VMs in total, 2 vCPU are assigned to each VM, only one MCP installed on one VM.

The graph below compares overall system CPU usage.

Figure 14: Comparison of System CPU Usage among different VMs configurations

Overall CPU usage scales linearly against port capacity, regardless of how many VMs are configured.


The two graphs below compare RTP stream quality related Max Jitter and Max Delta on these three different VM configurations:

Figure 15: Comparison of Max Jitter among different VM configurations
Figure 16: Comparison of Max Delta among different VM configurations


To achieve higher port capacity, configure more VMs and assign less vCPU to each VM. With audio quality criteria considered, Genesys recommends 600 ports as peak for six VM configurations. Six VMs with two vCPUs for each VM is the optimal configuration.

Below is the table of IOPS for 6 VM configurations:

Table: Disk IOPS of sum of all 6 VMs of dual hex cores, MP3 only

Ports Overall Disk IOPS (kbps)
Total Reads Writes
120 25.18 0.028 25.15
240 42.75 0.052 42.70
300 51.16 0.004 51.15
360 59.61 0.000 59.61
420 67.04 0.000 67.04
480 74.82 0.000 74.82
540 86.30 0.000 86.30
600 94.11 0.000 94.11
660 102.05 0.000 102.04
720 111.30 0.000 111.29


The graph below compares the two tables of IOPS (Table: Disk IOPS of sum of all 3 VMs of dual hex cores for 3 VMs and Table: Disk IOPS of sum of all 6 VMs of dual hex cores, MP3 only for 6 VMs), on the same hardware spec of dual hex cores:

Figure 18: Comparison of System Disk IOPS among different VMs.
  • System Disk IOPS scales linearly against port capacity, but not related for VM configurations.
  • We ran an additional test with only 1 vCPU assigned to each VM, on a single hex core server Hardware profile 2, with a 6-VMs in total on the one server. We could barely run beyond 150 ports—the single CPU cannot be linearly scaled—which compares with a 3-VMs configuration:
Figure 19: Comparison of System Usage for one vCPU vs two vCPUs VMs configuration


The two graphs below show that both Max Jitter and Max Delta jump significantly beyond 150 ports:

Figure 20: Comparison of Max Jitter for one vCPU vs two vCPUs VMs configuration
Figure 21: Comparison of Max Delta for one vCPU vs two vCPUs VMs configuration

The comparison indicates that MCP doesn’t perform well on a single vCPU VM.


Performance Comparison of MP3 only and MP3 + WAV

The graph below compares two test profiles (Profile 1 of MP3 only and Profile 2 of MP3 + WAV as dest2) on the same hardware spec with same 6 VM configurations of 2 vCPU per VM. Below is the CPU usage:

Figure 22: Comparison of System Usage for different test profiles

Overall CPU usage for Software Profile 2 (MP3 + WAV) is slightly higher than for Software Profile 1 (MP3 only).


The two graphs below compare audio quality criteria:

Figure 24: Comparison of Max Jitter among different test profiles
Figure 25: Comparison of Max Delta among different test profiles
  • For this test, applying Profile 2 to a 6 VMs configuration: Preferred/Recommended = 360 ports; Peak Port Capacity = 530 ports, if you can ignore some potential impact to audio quality.

The table below shows the IOPS of the sum of all 6 VMs for a test profile of MP3 + wav:

Table: IOPS of sum of all 6 VMs of dual hex cores, MP3 + wav

Ports Overall Disk IOPS (kbps)
Total Reads Writes
120 42.64 0.01 42.63
240 77.69 0.00 77.69
300 95.99 0.00 95.99
360 114.28 0.00 114.28
420 130.45 0.00 130.45
480 149.58 0.00 149.58
540 172.49 0.00 172.49
600 194.55 0.00 194.55
660 177.80 0.00 177.80

The graph below compares Table: IOPS of sum of all 6 VMs of dual hex cores, MP3 + wav with Table: Disk IOPS of sum of all 6 VMs of dual hex cores, MP3 only:

Figure 26: Comparison of System Disk IOPS for different test profiles on VMs


As we have cache folder on a different SSD drive, we can break down disk IOPS for each drive as below:

Table: Disk IOPS Break Down per Drive, Test Profile 1, MP3 only

Ports Overall Disk IOPS (kbps) SSD Drive E Disk IOPS (kbps) HDD Drive C Disk IOPS (kbps)
Total Reads Writes Total Reads Writes Total Reads Writes
120 25.18 0.03 25.15 20.88 0.00 20.88 4.30 0.03 4.28
240 42.75 0.05 42.70 36.96 0.00 36.96 5.79 0.05 5.74
300 51.16 0.00 51.15 44.63 0.00 44.63 6.53 0.00 6.53
360 59.61 0.00 59.61 52.80 0.00 52.80 6.81 0.00 6.81
420 67.04 0.00 67.04 60.31 0.00 60.31 6.74 0.00 6.74
480 74.82 0.00 74.82 67.85 0.00 67.85 6.97 0.00 6.97
540 86.30 0.00 86.30 79.31 0.00 79.31 6.99 0.00 6.99
600 94.11 0.00 94.11 87.31 0.00 87.31 6.80 0.00 6.80
660 102.05 0.00 102.04 95.12 0.00 95.12 6.92 0.00 6.92
720 111.30 0.00 111.29 104.30 0.00 104.30 6.99 0.00 6.99


Table: Disk IOPS Break Down per Drive, Test Profile 2, MP3 + wav

Ports Overall Disk IOPS (kbps) SSD Drive E Disk IOPS (kbps) HDD Drive C Disk IOPS (kbps)
Total Reads Writes Total Reads Writes Total Reads Writes
120 42.64 0.01 42.63 38.38 0.00 38.38 4.26 0.01 4.26
240 77.69 0.00 77.69 72.07 0.00 72.07 5.62 0.00 5.62
300 95.99 0.00 95.99 89.04 0.00 89.04 6.95 0.00 6.95
360 114.28 0.00 114.28 107.50 0.00 107.50 6.78 0.00 6.78
420 130.45 0.00 130.45 123.56 0.00 123.56 6.89 0.00 6.89
480 149.58 0.00 149.58 142.65 0.00 142.65 6.92 0.00 6.92
540 172.49 0.00 172.49 165.61 0.00 165.61 6.88 0.00 6.88
600 194.55 0.00 194.55 187.53 0.00 187.53 7.02 0.00 7.02

The two graphs below compare corresponding drives:

Figure 29: Comparison of cache folder of SSD Drive IOPS for different profiles

This SSD drive is used exclusively as the cache folder for MCP recording. The IOPS for Profile 2 (two dest2, MP3 + wav) is as double as Profile 1 (one dest. MP3 only).

Figure 30: Comparison of HDD Drive IOPS for different profiles


This HDD drive is used for all operations except the cache folder for MCP recording. IOPS is nearly constant at a regular load and below peak. Thus, the IOPS estimating formula can be:

IOPS1 = C + k * P (dest only)
IOPS2 = C + 2k * P (both dest + dest2)
Where P = ports, C = 7, k = 0.15


Performance Comparison between SAS HDDs and SSD

These tests compare performance between SAS HDDs and SSD for recording, using 6 VMs from the same hardware spec, and these four different HDD and SSD combinations:

  • 1 HDD: all 6 VMs on one 15 krpm SAS HDD drive.
  • 2 HDD: split 6 VMs on two 15 krpm SAS HDD drives, 3 VMs per drive.
  • 3 HDD: split 6 VMs on three 15 krpm SAS HDD drives, 2 VMs per drive.
  • SSD: all 6 VMs on one 15 krpm SAS HDD while a separate SSD drive used as cache folder only.

The testing was executed with Profile 1, MP3 only. Below is the overall system CPU usage:

Figure: Comparison of System Usage among different HDD/SSD drive combinations

The overall system CPU usage exhibits no significant different between HDD and SDD.

IOPS is almost the same for these 4 combinations, so these tests use the numbers in Table: Disk IOPS of sum of all 6 VMs of dual hex cores, MP3 only.


The graphs below compare max jitter and max delta for HDD/SSD drive combinations:

Figure 32: Comparison of Max Jitter among different HDD/SSD drive combinations
Figure 33: Comparison of Max Delta among different HDD/SSD drive combinations


This graph illustrates the average disk write queue for one drive:

Figure 34: Comparison of Avg Disk Write Queue among different HDD/SSD drive combinations
  • The queue starts to increase non linearly around 360 ports, which makes that number close to maximum port capacity of the hard drive.
  • In the three graphs above: with only one HDD drive, both max jitter and max delta started to increase dramatically from 330 ports and higher. Thus: Preferable/Recommended = 330 ports; Peak Port Capacity = 360 ports. In Table: Disk IOPS of sum of all 6 VMs of dual hex cores, MP3 only, IOPS is 51 for 330 ports; while IOPS is around 60 for 360 ports. Thus: Preferable/Recommended IOPS = 51; maximum IOPS for one 15 krpm SAS HDD = 60.
  • With multiple HDDs (2 or 3) to split the load, peak port capacity is nearly the same as SSD—660 ports since the load per drive would be 330 (for 2 HDD drives) and 220 (for 3 HDD drives). Max jitter does not exhibit big differences for these three configurations. But max delta shows a higher delay for 3 HDDs compared to SSD, and 2 HDDs compared to 3 HDDs. Thus: with strict audio quality required in these scenarios, fast media such as SSD will help improve latency and minimize any potential audio quality issues.


Data Throughput

These two formulas estimate data throughputs:

Formula 1 (for MP3 only):
MP3 bitrate * Ports / 8 = KB/sec
Or 32kbps * Ports / 8 =KB/sec if MP3 is 32kbps

Formula 2 (for MP3 + wav):
(MP3 bitrate + WAV bitrate) * Ports / 8 = KB/sec
Or (32 kbps + 128 kpbs) * Ports / 8 = 160 kbps / 8 = KB/sec if 32kpbs MP3 + wav

Six VM configurations, with SSD as the cache folder for MCP recording, produced the following measurements from testing for test SW Profile 1 (MP3 32 kbps only):

Table: Data Throughputs for MP3 32 kbps only

Ports Overall Disk (kbps) SSD Drive Disk (kbps)
Total Reads Writes Total Reads Writes
120 554.10 0.12 553.98 536.98 0.00 536.98
240 1075.70 0.50 1075.19 1053.19 0.00 1053.19
300 1332.61 0.06 1332.55 1308.69 0.00 1308.69
360 1601.09 0.00 1601.09 1575.02 0.00 1575.02
420 1847.91 0.00 1847.91 1822.30 0.00 1822.30
480 2109.57 0.00 2109.57 2082.49 0.00 2082.49
540 2461.25 0.00 2461.25 2434.04 0.00 2434.04
600 2728.83 0.00 2728.83 2702.57 0.00 2702.57
660 3010.07 0.00 3010.07 2982.84 0.00 2982.84
720 3310.64 0.00 3310.64 3280.45 0.00 3280.45

Apply Formula 1 to the 120-port and 600-port samples from the table above to achieve these results:

32 kpbs * 120 / 8 = 480 kb close to 534 in the table (in SSD)
32 kpbs * 600 / 8 = 2400 kb close to 2703 in the table (in SSD)

The measurements from real testing are slightly higher than calculations. Because other files such as metadata and JSON files are saved in the same cache folder, the formula might need adjusting.


Below is the table from testing measurement for SW Profile 2 (MP3 + wav) on the same 6 VM configuration with SSD as cache folder of MCP recording:

Table: Data Throughputs for MP3 32 kbps + wav

Ports Overall Disk (kbps) SSD Drive Disk (kbps)
Total Reads Writes Total Reads Writes
120 2444.255 0.25 2444.01 2427.025 0.000 2427.025
240 4881.163 0.01 4881.15 4859.951 0.000 4859.951
300 6083.649 0.00 6083.64 6058.294 0.000 6058.294
360 7380.491 0.00 7380.49 7354.970 0.000 7354.970
420 8547.663 0.00 8547.66 8522.034 0.000 8522.034
480 9828.785 0.00 9828.79 9802.991 0.000 9802.991
540 11093.931 0.00 11093.93 11067.838 0.000 11067.838
600 12335.993 0.01 12335.99 12309.182 0.000 12309.182

Apply Formula 2 to the 120-port and 600-port samples in the table above to achieve these results:

(32 kbps + 128 kbps) * 120 / 8 = 2400 kb close to 2427 in the table
(32 kbps + 128 kbps) * 600 / 8 = 12000 kb close to 12309 in the table


MCP IOPS

Important
IOPS is measured as it relates to MCP only, and includes IOPS for MP3, WAV and Log files (Interaction level verbosity).

A single HDD local hard drive was used for testing because the HDD itself could become the bottleneck. These tests focus on disk IOPS measurement and calculation, and certain real deployment scenarios require that a local drive not be used. Thus, the measurement for MCP IOPS is useful to calculate overall IO requirement. The three tables below offer three typical MCP IOPS configurations:

Table: MCP IOPS on physical server of single hex core, MP3 only

Ports Physical MCP IOPS (kbps)
Total Read Write
60 21.88 14.93 6.95
120 43.25 29.64 13.60
180 63.28 43.37 19.91
200 71.47 49.35 22.12
210 74.95 51.58 23.37
220 78.40 53.93 24.47
230 82.51 56.83 25.68
240 85.28 58.37 26.90


Table: MCP IOPS on 6 VMs of dual hex core, MP3 only

MP3 only
Ports 		Overall 6 VM MCP IOPS (kbps)
Total Read Write
120 43.672 29.679 13.993
240 87.477 59.280 28.197
300 109.292 73.848 35.445
360 130.965 88.917 42.047
420 151.706 103.095 48.611
480 171.290 116.053 55.237
540 194.772 132.867 61.905
600 215.101 146.882 68.219
660 236.654 161.990 74.664
720 259.100 177.279 81.820


Table: MCP IOPS on 6 VMs of dual hex core, MP3 + wav

MP3 + wav
Ports 		Overall 6 VM MCP IOPS (kbps)
Total Read Write
120 173.607 146.092 27.515
240 348.369 292.884 55.486
300 434.511 364.898 69.613
360 522.447 439.412 83.035
420 600.880 504.682 96.198
480 693.861 583.965 109.896
540 780.187 656.958 123.229
600 859.359 722.907 136.453
660 790.737 664.024 126.713


The graph below compares the two tables MCP IOPS on physical server of single hex core, MP3 only and MCP IOPS on 6 VMs of dual hex core, MP3 only:

Figure 40: MCP IOPS, Physical VS VMs


The graph below compares the two tables MCP IOPS on 6 VMs of dual hex core, MP3 only and MCP IOPS on 6 VMs of dual hex core, MP3 + wav:

Figure 41: MCP IOPS, MP3 only VS MP3 + WAV

The MCP IOPS is related to the test profile and ports, but on the same physical server and VMs.


MP3 16KBPS Bit Rate Compression

Support for MP3 16 kbps bit rate recording compression began with The GVP 8.5.1 release in December 2014. We tested performance on physical server and Virtual Machine (VM) environments, using Windows 2008 R2 x64.

Physical Server on Single Hex Core

Testing was performed on Hardware Profile 1: a physical server on a single hex core of Dell R410. The three graphs below compare system CPU usage and audio quality related metrics, max jitter and max delta.

Figure 42: Comparison of System CPU Usage, MP3 16kbps vs 32kbps on Physical Server
Figure 43: Comparison of Max Jitter, MP3 16kbps vs 32kbps on Physical Server
Figure 44: Comparison of Max Delta MP3 16kbps vs 32kbps on a Physical Server

MP3 16kbps consumes less CPU memory, which means higher port capacity. The two graphs above that compare Max Jitter with Max Delta also indicate the higher port capacity of MP3 16kbps. Recommended port capacity for MP3 16kbps: 240 ports (20% higher than the 200 recommended port capacity for MP3 32kbps). Peak port capacity: 270 ports (22.7% higher than the 220 peak port capacity for MP3 32kbps).

The table below lists the system disk IOPS:

Figure 45: System Disk IOPS on Physical Server, MP3 only 16 Kbps

Ports Physical Server Disk IOPS (kbps)
Total Reads Writes
60 14.66 0.036 14.62
120 24.00 0.041 23.95
180 33.42 0.029 33.39
210 37.65 0.030 37.62
240 42.21 0.029 42.18
270 47.18 0.036 47.14
300 51.44 0.011 51.43
330 55.81 0.006 55.81
360 60.99 0.002 60.99
390 67.12 0.003 67.11

The graph below compares Table: System Disk IOPS on Physical Server, MP3 only 16 Kbps with Table: Disk IOPS of system level from a physical server with a single hex core, both on a single hex core server:

Figure 46: Comparison of System Disk IOPS on Single Hex Core Physical Server, MP3 16 Kbps vs 32 Kbps

The system disk IOPS for MP3 16kbps and 32kbps are nearly identical to each other; reasonable since the disk IO operations should be the same, and at the same port capacity, no matter which MP3 bit rate is chosen.

MCP IOPS is listed here:

Figure 47: MCP IOPS on physical server of single hex core, MP3 only, 16 Kbps

Ports Physical Server MCP IOPS (kbps)
Total Reads Writes
60 14.56 7.53 7.04
120 28.64 14.92 13.72
180 42.54 22.29 20.25
210 49.42 25.93 23.48
240 56.41 29.64 26.76
270 63.38 33.34 30.04
300 70.36 36.92 33.44
330 77.53 40.79 36.74
360 85.52 44.46 41.06
390 94.68 48.14 46.54

The graph below compares Table: MCP IOPS on physical server of single hex core, MP3 only, 16 Kbps and Table: MCP IOPS on physical server of single hex core, MP3 only:

Figure 48: MCP IOPS on Single Hex Core Physical Server, MP3 16Kbps vs 32Kbps

MP3 16kbps uses less IOPS at the process level, probably be due to fewer network operations for MP3 16kbps.

VMs on Dual Hex Cores Server

The testing for MP3 16kbps was conducted on VM Profile 4 (based on Hardware Profile 4, which is a dual hex cores server). 6 VMs were configured, while only one MCP was installed on each Windows VM. The three graphs below compare overall CPU usage, audio quality related max jitter and max delta for MP3 16kbps vs 32kbps:

Figure 49: Comparison System CPU Usage of MP3 16kbps vs 32kbps on VM env
Figure 50: Comparison of Max Jitter, MP3 16kbps vs 32kbps on VM env
Figure 51: Comparison of Max Delta, MP3 16kbps vs 32kbps on VM environment

MP3 16kbps consumes less CPU memory, which matches test results on a physical server in Figure: Comparison of System CPU Usage, MP3 16kbps vs 32kbps on Physical Server. Both Max Jitter and Max Delta also show a higher port capacity for MP3 16kbps compression, which also matches test results on a physical server from Figure 4: Comparison of Max Jitter, MP3 16kbps vs 32kbps on Physical Server & Figure: Comparison of Max Delta MP3 16kbps vs 32kbps on a Physical Server. Preferred/Recommended port capacity for MP3 16 kbps: 720 ports (20% higher 600 ports for than MP3 32kbps). It’s the same increase as observed from a physical server. Peak port capacity for MP3 16kpbs can be as high as 840 ports (27.3% higher than 660 peak port capacity for MP3 32kbps.

The table below illustrates system disk IOPS:

Figure 52: Overall Disk IOPS on all 6 VMs of dual hex cores, MP3 only, 16 Kbps

Ports Overall 6 VMs Disk IOPS (kbps)
Total Reads Writes
120 26.57 0.13 26.44
360 63.47 0.13 63.34
480 80.66 0.15 80.51
600 93.73 0.04 93.69
660 109.53 0.14 109.39
720 118.76 0.13 118.62
780 126.15 0.07 126.08
840 134.12 0.04 134.09
900 142.21 0.09 142.12

The graph below compares overall disk IOPS of all 6 VMs for MP3 16kpbs against 32kbps in Table: Disk IOPS of sum of all 6 VMs of dual hex cores, MP3 only':

Figure 53: Comparison of Overall 6 VMs Disk IOPS MP3 16 kbps vs 32 kbps

The IOPS from both MP3 16kbps and 32kbps are inline with each other, as in the physical server tests.

Data throughput for MP3 16kbps is listed in following table:

Figure 54: Data Throughputs for MP3 only, 16 kbps

Ports Overall Disk (kbps) SSD Drive Disk (kbps)
Total Reads Writes Total Reads Writes
120 318.17 0.68 317.49 296.313 0.001 296.312
360 892.94 0.52 892.42 856.077 0.001 856.076
480 1175.63 0.79 1174.84 1132.997 0.001 1132.996
600 1537.43 0.19 1537.23 1510.543 0.000 1510.543
660 1729.45 0.58 1728.87 1680.374 0.003 1680.371
720 1890.48 0.58 1889.90 1837.492 0.000 1837.492
780 2045.34 0.35 2045.00 1995.239 0.004 1995.235
840 2191.98 0.15 2191.83 2142.373 0.002 2142.371
900 2349.18 0.75 2348.44 2298.426 0.004 2298.422

Using this formula:
MP3 bitrate * Ports / 8 = kbps
...where MP3 bitrate=16kbps and Ports = 120 and 720 from the table above,

The results...
16 kpbs * 120 / 8 = 240 kbps (compared to 296 in the table -- in SSD)
and
16 kpbs * 720 / 8 = 14400 kbps (compared to 1837 in the table -- in SSD)

...from real testing for MP3 16kbps are slightly higher than calculations predict, due to other files such as metadata and JSON files being saved in the same cache folder. So the formula still stands.

The following table lists MCP IOPS:

Figure 55: Overall MCP IOPS from 6 VMs of dual hex core, MP3 only, 16kbps

Ports Overall 6 VMs MCP IOPS (kbps)
Total Reads Writes
120 28.931 14.915 14.016
360 86.517 44.456 42.061
480 114.574 59.153 55.421
600 142.112 73.730 68.382
660 156.495 81.359 75.136
720 170.237 88.660 81.577
780 184.173 96.048 88.125
840 197.767 103.263 94.504
900 211.644 110.545 101.099

The graph below compares Overall MCP IOPS with MP3 32k MCP IOPS, and shows the same trend of physical server results that appeared in Figure: MCP IOPS on Single Hex Core Physical Server, MP3 16Kbps vs 32Kbps:

Figure 56: MCP IOPS on 6 VMs of Dual Hex Cores, MP3 16Kbps vs 32Kbps


MP3 16KBPS Bit Rate Compression with Encryption

We tested the MP3 16 kbps bit rate with encryption, using the dest2 physical server and Vitual Machine (VM) environments, which compares with results of non-encryption from MP3 16 kbps Bit Rate without Encryption. The OS remained Windows 2008 R2 x64.

Physical Server on Single Hex Core

These tests were performed on Hardware Profile 1: a physical server on a single hex core of Dell R410. The three graphs below compare system CPU usage and audio quality-related metrics, max jitter and max delta.

Figure 57: Comparison of Physical Server System CPU Usage of MP3 16kbps encryption vs non-encryption
Figure 58: Comparison of Physical Server Max Jitter of MP3 16kbps encryption vs non-encryption
Figure 59: Comparison of Physical Server Max Delta of MP3 16kbps encryption vs non-encryption

In the graphs above, encryption consumes slightly higher system CPU than does non-encryption. Max Jitter and Max Delta consume much more CPU with encryption, than without. If a slightly higher delay due to latency introduced by encryption is acceptable, then recommended and preferred port capacity would be 210 ports—only a 12.5% reduction from the peak capacity of 240 ports offered by non-encryption. If the audio quality strictly applies, then the recommended port capacity can be as low as 120 ports. Peak port capacity could be the same 270 ports as non-encryption, if the delay is acceptable.

The table below lists system disk IOPS:

Figure 60: IOPS on physical server of single hex core, MP3 only, 16 Kbps, encryption

Ports Physical Server Disk IOPS
Total Reads Writes
60 14.66 0.036 14.62
120 24.00 0.041 23.95
180 33.42 0.029 33.39
210 37.65 0.030 37.62
240 42.21 0.029 42.18
270 47.18 0.036 47.14
300 51.44 0.011 51.43
330 55.81 0.006 55.81
360 60.99 0.002 60.99
390 67.12 0.003 67.11

The graph below compares system disk IOPS on a physical server IOPS with non-encryption:

Figure 61: Comparison of System Disk IOPS on Single Hex Core Physical Server, MP3 16kbps encryption vs non-encryption

System disk IOPS is nearly the same for encryption and non-encryption; both increase slightly at a higher port capacity. Some of that can be attributed by other disk IO operations, such as encryption key files.

The table below lists MCP IOPS:

Figure 62: MCP IOPS on physical server of single hex core, MP3 only, 16 Kbps, encryption

Ports Physical Server MCP IOPS
Total Reads Writes
60 16.53 8.88 7.65
120 32.59 17.69 14.91
150 40.40 21.96 18.44
180 48.46 26.46 22.01
210 56.35 30.83 25.52
240 64.32 35.24 29.08
270 72.28 39.64 32.64
300 80.06 43.95 36.11
330 88.61 48.53 40.07
360 100.48 52.91 47.57

The graph below compares total MCP IOPS between encryption and non-encryption:

Figure 63: MCP IOPS on Single Hex Core Physical Server, MP3 16Kbps encryption vs non-encryption

MCP IOPS for encryption increases when port capacity increases. As seen in Figure: Comparison of System Disk IOPS on Single Hex Core Physical Server, MP3 16kbps encryption vs non-encryption, increase for disk IOPS is much smaller for encryption, so here the increase should be attributed to network IOs.

VMs on Dual Hex Cores Server

The testing for MP3 16kbps with encryption was conducted on the VM Profile 4 based on Hardware Profile 4 of a dual hex cores server, same as non-encryption in the 16knps tests VMs on Dual Hex Cores Server. Six VMs were configured while only one MCP was installed on each Windows VM. Below are three graphs comparing overall CPU usage, audio quality related max jitter and max delta for MP3 16kbps encryption vs non-encryption:

Figure 64: Comparison of Overall VMs CPU Usage of MP3 16kbps encryption vs non-encryption
Figure 65: Comparison of Overall VMs Max Jitter of MP3 16kbps encryption vs non-encryption
Figure 66: Comparison of Overall VMs Max Delta of MP3 16kbps encryption vs non-encryption

The VM environment exhibits a similar trend: slightly overall CPU usage for the encryption profile, and much higher for max jitter and max delta. Applying the same criteria from the physical server results, if a slightly higher delay (due to latency introduced by encryption) is acceptable, then the recommended and preferred port capacity could be 600 ports—only a 16.7% reduction of the peak 720 ports with non-encryption. If audio quality strictly applies, the recommended ports can be as low as 480 ports. And if some delay is acceptable, then the peak port capacity can be the same 840 ports as non-encryption.

The overall system disk IOPS for all 6 VMs is listed below:

Figure 67: Overall Disk IOPS on all 6 VMs of dual hex cores, MP3 only, 16 Kbps, encryption

Ports Overall 6 VMs Disk IOPS SSD Drive Disk IOPS
Total Reads Writes Total Reads Writes
120 28.70 0.004 28.69 21.881 0.000 21.881
360 67.46 0.004 67.46 56.238 0.000 56.238
480 87.56 0.026 87.54 74.903 0.000 74.903
600 108.01 0.015 107.99 93.647 0.000 93.647
660 119.49 0.005 119.48 104.304 0.000 104.304
720 128.76 0.020 128.74 114.441 0.000 114.441
780 137.68 0.015 137.66 123.210 0.002 123.209
840 146.99 0.009 146.98 132.646 0.002 132.644
900 154.68 0.025 154.66 140.145 0.002 140.143

The graph below compares system disk IOPS with encryption and with non-encryption, on the same VM environment:

Figure 68: Comparison of Overall 6 VMs Disk IOPS MP3 16 kbps encryption vs non-encryption

As with the physical server tests, encryption increases as port capacity increases. Also as with the physical server tests, some of that can be attributed to extra disk IO operations.

The table below lists Data throughputs for encryption:

Figure 69: Data Throughputs for MP3 only, 16 kbps, encryption

Ports Overall Disk KB/sec SSD Drive Disk KB/sec
Total Reads Writes Total Reads Writes
120 387.99 0.02 387.97 304.229 0.000 304.229
360 1096.82 22.54 1074.28 876.599 0.000 876.599
480 1344.60 107.95 1236.65 1191.403 0.006 1191.397
600 2187.50 348.40 1839.09 1532.171 0.000 1532.171
660 2024.16 35.09 1989.07 1652.232 0.000 1652.232
720 1955.33 99.81 1855.51 1803.207 0.006 1803.201
780 2572.79 205.15 2367.64 1982.733 0.024 1982.709
840 2534.97 28.65 2506.32 2097.871 0.043 2097.829
900 2851.85 119.47 2732.38 2297.264 0.007 2297.257

Using Formula 1...

MP3 bitrate * Ports / 8 = KB/sec
Or 16kbps * Ports / 8 =KB/sec if MP3 is 16kbps

...take two samples (120 & 720) from the above table above, and apply them to Formula 1:

16 kpbs * 120 / 8 = 240 kb close to 304 in the table (in SSD)
16 kpbs * 720 / 8 = 14400 kb close to 1803 in the table (in SSD)

The measurements from real testing for MP3 16kbps encryption are slightly higher than these calculations predict, due to other file, such as metadata and JSON files, being saved on the same cache folder.

The graph below compares overall data throughputs with no encryption:

Figure 70: Comparison of Overall 6 VMs Data Throughputs MP3 16 kbps encryption vs non-encryption

The data throughputs for encryption increase slightly when port capacity increases, matching a similar trend with system disk IOPS.

The table below lists overall MCP IOPS from all 6 VMs:

Figure 71: Overall MCP IOPS from 6 VMs of dual hex core, MP3 only, 16kbps, encryption

Ports Overall 6 VMs MCP IOPS
Total Reads Writes
120 34.874 17.638 17.236
360 102.624 52.900 49.724
480 130.285 70.377 59.909
600 168.849 87.766 81.083
660 186.175 96.882 89.293
720 193.248 105.171 88.077
780 219.395 114.398 104.997
840 235.730 123.009 112.720
900 252.198 131.682 120.516

The graph below compares performance of the same configuration, except with non-encryption:

Figure 72: MCP IOPS from 6 VMs of dual hex core, MP3 only, 16kbps, encryption vs non-encryption

MCP IOPS performance is affected slightly by encryption, similar to the trend expressed in the physical server results.

Comments or questions about this documentation? Contact us for support!