Performance Server


The purpose of the Performance Server is to analyze network data in order to:

  • Determine if problems exist prior to their affecting services

  • Maximize network utilization

  • Pre-empt the occurrence of congestion

  • Demonstrate compliance with agreed SLAs

  • Indicate when extra network investment is needed (capacity planning)

The Performance Server faces into the network and receives asynchronously generated NE data. It can also proactively retrieve from MIBs data such as the level of bandwidth consumption. Performance data, for example, the number of IP packets received by an LER, may change very rapidly . For this reason, performance data is often automatically generated and emitted by NEs rather than being polled. The number of IP packets landing at an MPLS LER ingress interface may be hundreds of thousands per second. For data such as this, the NE generally periodically creates a data record and emits it to a listening application (in this case, the Performance Server).

In this discussion, we focus on the data issued by the NEs and the way this is processed and used by the Performance Server. Also, SLAs are introduced.

Figure 6-7 illustrates a possible structure for a Performance Server.

Figure 6-7. Performance Server components .

graphics/06fig07.gif

Mediation

We introduced mediation in this chapter and the previous one in the context of billing. It is a process of analyzing the raw data generated by NEs to produce standard format billing details. Mediation can also be applied to performance data to produce sanitized details for downstream use by third-party applications.

Aggregation

This is the process by which separate performance data records are combined. An example is an ATM SPVCC that spans a number of NEs. A given virtual circuit has certain performance parameters of interest:

  • Number of cells transported per second

  • Number of cells dropped due to excessive input traffic

  • Average bandwidth used by the cell traffic

  • Number of contract violations

Similarly, the performance details of other objects may be of interest:

  • Nodes

  • Interfaces

  • Links

  • LSPs

  • Multiservice cross connections ”Ethernet-to-MPLS, FR over ATM, and so on

Aggregation logically joins up the separate performance data records so that the relevant managed objects can be analyzed . An example is a count of all the cells transmitted by an ATM interface; at time T1 the number of cells may be x , and later at time T2 the number of cells may be x + y . Aggregation links this data together and stores it.

Correlation

The aggregated data is then correlated with the associated managed objects in readiness for reporting. Examples of correlated data are (for a single customer):

  • The number of IP packets forwarded by an LSP end to end

  • The number of Ethernet frames forwarded by an LSP end to end

  • The number of cells carried by an ATM PVC

  • The number of cells dropped by an ATM switch

These data should help give a clear picture of the performance of the underlying objects.

Reports

Reports are the means by which end users can view the performance details. Examples of reports are:

  • Utilization of managed objects, such as interfaces, links, nodes, and virtual circuits

  • The difference between actual and planned loads

  • Real-time and historical SLA conformance

Reports may be viewed by the network operator and may be accessible to customers via the Web. Important aspects of performance reports are:

  • Baseline measurements ”a realistic snapshot of the performance

  • Deviation from the baseline as network changes occur

  • Utilization of network resources

Baseline measures are essential for effective performance management. This is a set of readings taken from the network during normal operation. As the network changes ”for example, due to link failure, variations in traffic mix, or increased traffic ”the baseline changes with it. The extent of the deviation is important for planning changes or additions to the network. It may also help to pinpoint problems before they become service-affecting.

SLA Alerts

It is very important for enterprises to avoid violating SLA terms because there may be financial penalties ”particularly when the network management processes have been outsourced. SLA alerts can be issued based on ongoing analysis of trends in an effort to pre-empt violations before they actually occur. An example SLA is illustrated in Table 6-3.

Table 6-3. An IP SLA

Source IP Address

10.81.1.45

Source Port

444

Destination IP Address

10.81.2.89

Destination Port

445

Link Bandwidth

10Mbps

Interpacket Delay

1ms

Ordered Delivery

Yes

Packet Loss

0.0001%

Jitter

No

Round Trip Delay

30ms

This SLA indicates that IP traffic from 10.81.1.45 port 444 will land in the SP network on a 10Mbps link destined for 10.81.2.89 port 445. The interpacket arrival time is specified to be no more than 1 millisecond with no packets arriving out of order. A tunneling technology such as MPLS or L2TP could be employed to achieve the latter. An SLA alert might be raised in the Performance Server if the link bandwidth increased up to or beyond 9.9Mbps.

Topology Update

Performance Server data changes (such as detection of a congested link) will be of interest to clients , and it may be necessary for a topology update to occur after changes such as the following:

  • When congestion is imminent on a given link

  • When a virtual circuit is being presented with excessive traffic ”it may be necessary to add extra bandwidth to the circuit

These are important because, left unattended, they can have a serious impact on the network.

Performance Server Database Tables

Performance data can be related to the associated managed object; for example, there can be separate tables for each of the following:

  • Nodes

  • Interfaces

  • Links

  • Virtual connections

Each of these tables can have separate columns for the relevant performance data, such as:

  • Number of incoming and outgoing packets, cells, and frames

  • Bandwidth in use

  • SLA status

The rows and columns of these tables are populated by the components of the Performance Server. Alternatively, the other servers can share the above tables.



Network Management, MIBs and MPLS
Network Management, MIBs and MPLS: Principles, Design and Implementation
ISBN: 0131011138
EAN: 2147483647
Year: 2003
Pages: 150

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net