How our diagnostic tool for network analysis solves administrators' problems
Use of the Allegro Network Multimeter in the Data Centre
Actual situation: Overload in the data centre
Hot server rooms, a droning cooling system, a multitude of coloured cables and flashing lights: this is what a classic data centre looks like. How can discrepancies in the network be efficiently resolved in this environment?
A network must run in a stable manner. 24/7 even when the administrator is off duty. Servers with high data rates run continuously at high bandwidths with many different simultaneous services. There are always load peaks and packet losses. With this complexity, an administrator has to keep an eye on all network processes, be prepared for different error classes and be able to react quickly.
In addition, users may complain about slow networks, the poor quality of the VoIP system or server problems. The blame is often placed on the administrator, but there are many possible sources of error that come into question. Inconsistencies can occur on all network layers. Therefore, the Allegro Network Multimeter analyses and correlates all data across network layers 2 to 7.
Tips and important indicators that the Allegro Analyser uses to detect errors in seconds
Depending on the application, specific indicators provide important information about possible problem sources.
Storage system: For storage systems, one variable to be checked is data transfer speed. Special attention should be paid to whether server hardware, clients or other connected networks can support the specified speed. With data throughput over time, it is possible to find out which applications/IP addresses generate or receive particularly high traffic. The Allegro Network Multimeter detects precisely these applications/IPs. This information can reveal real-time collisions of particularly large data transfers, such as backups during office hours. It is also possible to find out which component is the bottleneck. An investigation of the TCP-Zero-Window parameter is helpful to detect receiver overload.
Web server: The error is not always within the administrator's sphere of influence. It is possible that the Web server cannot respond tp the requested Web content fast enough for internal and external use. To determine problems that can be traced back to the web server, the Allegro-Packets-tool analyses the requests to the server. Indicators are the number of requests to the server per second, the time of the request peaks and their cause, and the number and size of the transferred data. By passively measuring the response time, it is possible to determine whether the response times change over time and which services had to wait long periods for a response. Particularly in the case of unanswered requests, it should be checked which requests are affected and why. Even rare TCP retransmissions can lead to sporadic problems due to increasing latency of requests.
VoIP: VoIP data packets should always have priority over other data to ensure voice quality. In case of problems, the Allegro meter determines the following key figures to identify the cause: the number of calls made simultaneously, the bandwidth and codecs used, jitter, packet loss, MOS scores and failed calls. It should also always be considered if QoS is correctly configured in both directions.
Industrial server: Since production downtimes can be associated with considerable loss of revenue, frequent checks should be made to ensure if servers are working reliably, if there have been failures and if the server is being overloaded with additional, unneccessary tasks. Network traffic in the industrial sector is mostly relatively even. The Allegro system keeps an eye on outliers, failures, bandwidth and latency over time. Of course, the industry protocol-specific response times are also monitored.
In order to clarify a problem, different network classes or layers must be considered. Generic measurement tools suchs as the Allegro Network Multimeter are available which analyses layer by layer and presents the results clearly. With this method, both homemade problems and sporadic errors can be analysed. Furthermore, packet bursts become visible.
Access to data to assess the data centre condition
In order to assess the state of a data centre, a specific server can be monitored on site with the Allegro analysis tool. This provides only a brief insight into the current state of the network. If a specific error does not occur in a defined time window, it may not be recorded. It is therefore recommended to integrate the Allegro Network Multimeter in the network that can be accessed remotely by the network administrator. Simple access can be provided via managed switches. Managed switches allow the configuration of a mirror port. This can route all or parts of the traffic to a measurement system. This method can be used for continuous monitoring or over a specific time period. It is necessary to install the Allegro meter on the mirror port. The monitoring tool can receive, read and graphically process this data. Many switches support ERSPAN/GRE for forwarding network traffic to another IP addresses. The advantage is that the server to be monitored can be located at a different location than the measurement system and no connections changes are needed. However, the network must be able to cope with the additional traffic.
The most advantageous are measuring systems like the Allegro Network Multimeter which are remotely accessible via a web browser. You do not need to be at the remote site and you can be flexible in the use of end devices (desktop PC, smartphone) and multiple administrators can use one device to investigate different parameters. Valuable time can also be saved by monitoring virtual machines.
Virtualization is an everyday topic in data centres. There may be several physical servers running a large number of virtual machines. How can problems be analysed in such an environment? Virtual machines require additional precautions. If traffic goes into the physical network, the known routes can be used to route the traffic into the measurement system. However, internal traffic between several virtual machines may also need to be investigated. Problems there can lead to external interference. Here, ERSPAN can be used within a virtual machine to divert the traffic. It is also possible to set up a virtual measurement environment on the same VM host to receive the traffic without having to enter the physical network. The advantage is less network load.
Efficient troubleshooting with the Allegro Network Multimeter
High bandwidths of 10/25/40/100 GBit/s require dedicated measurement appliances such as the monitoring tools of the producer Allegro Packets. A conventional laptop can become overloaded with massive numbers of IP addresses and connections, or the traffic must be heavily filtered so that a Gigabit interface, for example, is sufficient. Recording raw data is essential; laptops cannot store high bandwidth network data as quickly and efficiently as a dedicated analysis appliance. Raw data enables a network manager to immerse themself deeper into solving a problem in a retrospective manner. The network troubleshooter from Allegro Packets can narrow down errors by extracting raw data for deeper analysis. This is particularly helpful for detecting sporadic errors, since it is not possible to record these in a targeted recording. Therefore, the Allegro Network Multimeter which can store data is indispensable. In this case, packet extraction saves a lot of time. In order to detect errors before they become a problem, as much live measurement data as possible should be available to obtain a complete overview and to enable a network administrator to deal with any issues. A prepared generic overview allows you to identify network status with a glance at the most important key figures.