Scaling guidelines for statistics collection
Statistics collection
NFM-P provides the ability to collect statistics information from the network elements. This section provides guidelines that can be used to determine the extent to which statistics collection can be retrieved from the network.
Statistics collection definitions
Performance statistics: These statistics are associated with various network objects such as ports, interfaces, channels and network elements (routers). These statistics are retrieved by NFM-P using SNMP polling according to the MIB policies that are configured by the user.
Accounting statistics: These statistics are associated with Services, Subscribers, and Network Interfaces and contain data that can be used for accounting, billing and SLA management purposes. These statistics are collected on the 7x50 and retrieved by NFM-P via a file that is transferred via ftp/sftp.
Application Assurance Accounting statistics: These statistics are associated with Subscribers, SAPs, and spoke SDP bindings and contain data related to traffic flows that can be used for QoS and traffic management, and application aware reporting. These statistics are collected on the 7x50 ISA cards and retrieved by NFM-P via a file that is transferred via ftp/sftp.
Statistics Item: An individual statistics counter, such as RxOctets or TxFrames.
Statistics Record: A collection of statistics items which is retrieved from the router and stored in the NFM-P database as an atomic operations. In the various statistics forms on the NFM-P GUI client, a statistics record appears to the user as a single row which contains the collection or retrieval timestamp and a set of individual statistics items. In the case of performance statistics, a statistics record corresponds to a row in the MIB table.
Determining the number of statistics records that will be collected
Statistics can be collected and processed by the NFM-P server or by the NFM-P auxiliary statistics collector for dedicated statistics handling. The NFM-P auxiliary statistics collector provides a dedicated station for statistics collection. The following sections should be used to determine the maximum performance and accounting statistics for different hardware setups.
Performance statistics
See the NSP NFM-P Statistics Management Guide to find the steps required to configure NFM-P to retrieve and process performance statistics. Note that two steps are required to enable the collection of performance statistics from the network. First, a policy is defined which specifies a set of polling periods for various MIBs. Second, the policy is applied to a number of network elements.
In general, enabling the statistics collection of a MIB will result in one statistics record being collected, at the specified polling period, for each network object to which the MIB applies.
For example, consider a policy is created with only the rtr.L2AccessDhcpRelayCfgStats MIB enabled for collection at 15-minute intervals. That policy is assigned to only two network elements which each contain 500 L2 Access Interfaces. As a result of this action, NFM-P will collect 1000 statistics records from the network every 15 minutes.
The quantity of resources which are allocated to the retrieval and processing of performance statistics does not depend significantly on the number of CPU Cores available to the NFM-P server or auxiliary statistics collector software. The tables below show the maximum number of performance statistics that can be retrieved and processed by the NFM-P server and the NFM-P auxiliary statistics collector every 15 minutes.
Table 5-12: Maximum number of performance statistics records processed by an NFM-P server
|
Number of CPU cores on NFM-P server stations |
Maximum number of performance statistics records per 15-minute interval | |
|---|---|---|
|
Collocated configuration |
Distributed configuration | |
|
6 or greater |
50 000 |
150 000 |
Table 5-13: Maximum number of performance statistics records processed by an NFM-P statistics auxiliary
|
Number of active auxiliary statistics collectors |
Maximum number of performance statistics records per 15-minute interval | ||||||
|---|---|---|---|---|---|---|---|
|
Statistics collection with NFM-P database |
Statistics collection with single auxiliary database |
Statistics collection with three+ auxiliary database cluster |
logTofile only | ||||
|
8 CPU Cores, 32 GB RAM |
12 CPU Cores, 32 GB RAM |
8 CPU Cores, 32 GB RAM |
12 CPU Cores, 32 GB RAM |
12 CPU Cores, 32 GB RAM | |||
|
1 |
500 000 |
2 000 000 |
500 000 |
2 000 000 |
2 000 000 | ||
|
2 |
500 000 |
2 000 000 |
500 000 |
4 000 000 |
4 000 000 | ||
|
3 |
500 000 |
2 000 000 |
500 000 |
4 000 000 |
4 000 000 | ||
In situations where NFM-P is asked to collect more performance statistics than it can process in the specified polling period, the PollerDeadlineMissed alarms will start appearing. These alarms indicate to the user that the polling mechanisms within NFM-P cannot retrieve the requested information within the specified polling period. Should this situation arise, the polling period for statistics should be increased or the number of objects that are applied to Statistics Poller Policies should be reduced.
Performance statistics collection and network latency
NFM-P collection of performance statistics from a single network element may be limited due to the round trip delay caused by network and network element latency. NFM-P collects performance statistics records using SNMP. One record is collected at a time to limit the load on the network element. Therefore, round trip latency will directly impact the maximum number of performance statistics records collected. As an example, if the round trip latency is 100 ms, and we target a completion time of 66% of the collection interval (to allow for processing variances and other system impacts), the maximum number of performance statistics records that can be collected from one network element in a 15 minute interval would be 6000 records (66% of 900 seconds divided by 100 ms latency).
Accounting statistics
See the NSP NFM-P Statistics Management Guide to find the steps required to configure NFM-P to retrieve and process accounting statistics.
The quantity of resources which are allocated to the retrieval and processing of accounting statistics within the NFM-P server or auxiliary statistics collector are set at the installation time and depend on the number of CPU Core available to the NFM-P server or auxiliary statistics collector software. The number of CPU Cores available to the server depends on the number of CPU Cores on the station and whether the NFM-P database software is collocated with the NFM-P server software on the same station.
An accounting statistic record is the statistic for one queue for one SAP. For example, if 2 ingress and 2 egress queues are configured per SAP, the “Combined Ingress/Egress” statistic represents 4 NFM-P accounting statistic records.
It is recommended that the Accounting Policy Interval and the File Policy Interval be aligned to the same period. Misalignment of the policy periods can cause NFM-P resource contention for both performance and accounting statistics processing.
The following tables provide the maximum number of accounting statistics records that can be retrieved and processed by the NFM-P server or NFM-P auxiliary statistics collector in various situations.
To reach the peak accounting statistics collection from the NFM-P auxiliary statistics collector station, the NFM-P database station requires a customized configuration that can be obtained from Nokia personnel.
Table 5-14: Maximum number of accounting statistics records processed by an NFM-P server station
|
Number of CPU cores on NFM-P server stations |
Maximum number of accounting statistics records per 15-minute interval | |
|---|---|---|
|
Collocated configuration |
Distributed configuration | |
|
6 |
100 000 |
200 000 |
|
8 or greater |
200 000 |
400 000 |
Table 5-15: Maximum number of accounting statistics records processed by an NFM-P statistics auxiliary
|
Number of active auxiliary statistics collectors |
Maximum number of accounting statistics records per 15-minute interval | ||||||
|---|---|---|---|---|---|---|---|
|
Statistics collection with NFM-P database |
Statistics collection with single auxiliary database |
Statistics collection with three+ auxiliary database cluster |
logToFile only | ||||
|
8 CPU cores, 32 GB RAM |
12 CPU cores, 32 GB RAM |
8 CPU cores, 32 GB RAM |
12 CPU cores, 32 GB RAM |
12 CPU cores, 32 GB RAM | |||
|
1 |
10 000 000 |
10 000 000 |
5 000 000 |
20 000 000 |
20 000 000 | ||
|
2 |
10 000 000 |
10 000 000 |
5 000 000 |
40 000 000 |
40 000 000 | ||
|
3 |
10 000 000 |
10 000 000 |
5 000 000 |
60 000 000 |
60 000 000 | ||
In situations where NFM-P is asked to collect more accounting statistics records than it can process in the specified retrieval period, the extra statistics will not be retrieved from the network.
There are two methods to export accounting and performance statistics from NFM-P; registerLogToFile, and findToFile. The registerLogToFile method is the preferred method and is required for situations where more than 400 000 accounting statistics records are retrieved in 15 minutes or 500 000 performance statistics are retrieved in 15 minutes.
Application assurance accounting statistics
See the NSP NFM-P Statistics Management Guide to find the steps required to configure NFM-P to retrieve and process application assurance accounting statistics.
The quantity of resources which are allocated to the retrieval and processing of application assurance accounting statistics within the NFM-P server are set at the installation time and depend on the number of CPUs available to the NFM-P server software. The number of CPUs available to the NFM-P server depends on the number of CPUs on the station and whether the NFM-P database software is collocated with the NFM-P server software on the same station.
Scaling of application assurance collection is related to the number of objects configured for collection as opposed to the number of records collected per interval.
The following tables provide the maximum number of application assurance objects that can be configured for collection by the NFM-P server or NFM-P auxiliary statistics collector in various situations.
Table 5-16: Maximum number of application assurance accounting objects configured for collection by an NFM-P server station
|
Number of CPU cores on NFM-P server stations |
Maximum number of application assurance accounting objects configured for collection per 15-minute interval | |
|---|---|---|
|
Collocated configuration |
Distributed configuration | |
|
6 |
50 000 |
100 000 |
|
8 or greater |
100 000 |
200 000 |
Table 5-17: Maximum number of application assurance accounting objects configured for collection by an NFM-P statistics auxiliary
|
Number of active auxiliary statistics collectors |
Maximum number of application assurance accounting objects configured for collection per 15-minute interval | ||||||
|---|---|---|---|---|---|---|---|
|
Statistics collection with NFM-P database |
Statistics collection with single auxiliary database |
Statistics collection with three+ auxiliary database cluster | |||||
|
8 CPU Cores, 32 GB RAM |
12 CPU cores, 32 GB RAM |
8 CPU cores, 32 GB RAM |
12 CPU cores, 32 GB RAM | ||||
|
1 |
5 000 000 |
7 500 000 |
1 000 000 |
5 000 000 | |||
|
2 |
5 000 000 |
15 000 000 |
1 000 000 |
10 000 000 | |||
|
3 |
5 000 000 |
15 000 000 |
1 000 000 |
20 000 000 | |||
In situations where NFM-P is asked to collect more application assurance accounting records than it can process in the specified retrieval period, the extra statistics will not be retrieved from the network.
Exporting performance and accounting statistics records
There are two methods to export accounting and performance statistics from NFM-P; registerLogToFile, and findToFile. The registerLogToFile method is the preferred method and is required for situations where more than 400 000 accounting statistics records are retrieved in 15 minutes or 500 000 performance statistics are retrieved in 15 minutes. This recommendation also minimizes collection latency and reduces system load.
NFM-P database hardware platform requirements
To collect large numbers of statistics using the NFM-P database, there are RAM and storage I/O requirements for the NFM-P database station. The table below highlights these requirements.
Table 5-18: NFM-P database station hardware requirements for a distributed configuration
|
Maximum number of simultaneous statistics records per 15-minute interval |
NFM-P auxiliary statistics collector(s) |
Requires the following NFM-P database station setup | ||
|---|---|---|---|---|
|
Accounting statistics records |
Application assurance accounting objects configured for collection |
Performance statistics records | ||
|
400 000 |
0 |
0 |
No |
4 CPU cores, minimum 2.0GHz 1 4 disks (RAID 0) 32 GB RAM |
|
0 |
200 000 |
0 | ||
|
0 |
0 |
150 000 | ||
|
800 000 |
0 |
0 |
Yes |
4 CPU cores, minimum 2.0GHz 1 4 disks (RAID 0) 48 GB RAM |
|
0 |
400 000 |
0 | ||
|
0 |
0 |
200 000 | ||
|
10 000 000 |
0 |
500 000 |
Yes |
8 CPU cores, minimum 2.0GHz 1 6 disks (RAID 0) 64 GB RAM |
|
0 |
5 000 000 |
500 000 | ||
|
10 000 000 |
5 000 000 |
2 000 000 |
Yes |
12 CPU cores, minimum 2.0GHz 1 6 disks (RAID 0) 64 GB RAM |
|
0 |
15 000 000 |
0 | ||
Notes:
2.0GHz only supported on Skylake and newer CPU microarchitecture. Minimum speed on CPUs older than Skylake is 2.4GHz
Simultaneous collection of performance, application assurance accounting and accounting statistics records
NFM-P can collect performance, application assurance, and accounting statistics records simultaneously. However, it is important to consider that enabling the collection of one type of statistics will reduce the capability of NFM-P to collect and process the other type of statistics. It is therefore not possible to achieve the maximum stated limits for performance, application assurance, and accounting statistics records simultaneously, in certain configurations. Table 5-18, NFM-P database station hardware requirements for a distributed configuration shows an example of simultaneous collection.
Determining the number of performance and accounting statistics records being collected by NFM-P
To ensure the number of performance and accounting statistics records that NFM-P is asked to collect and process every 15 minutes remains below the stated scalability guidelines, it is important to carefully assess the impact of creating and assigning statistics policies. Review the number of objects that are assigned to statistics policies and ensure the polling and retrieval periods are set such that the numbers will remain below the stated guidelines.
Using NFM-P server performance statistics, NFM-P can assist in determining how many polled and accounting statistics are being collected.
NFM-P performance can be adversely affected by increasing the number of historical statistics entries recorded by the NFM-P. NFM-P system impacts include increased time listing log records from the GUI and XML API clients, increased database space, and increased database backups times.
Statistics record retention
The tables below shows the retention that is achievable depending upon the total number of records to retain, the statistic type, and the database used to retain the records:
Table 5-19: Maximum statistics interval retention - NFM-P database
|
Statistics type |
Total number of statistics records to be stored in the database |
Maximum number of retention intervals |
|---|---|---|
|
Performance |
<40M |
672 |
|
>40M |
96 | |
|
Accounting |
<40M |
672 |
|
>40M |
16 |
Table 5-20: Maximum statistics interval retention - auxiliary database
|
Statistics type |
Maximum number of retention intervals |
|---|---|
|
Performance |
35,040 |
|
Accounting |
35,040 |
When using the logToFile method only, for collection, the maximum retention of data on the file system is 600 minutes (10 hours).