Performance Optimization Guide: Difference between revisions
No edit summary |
|||
| Line 14: | Line 14: | ||
* Disk I/O | * Disk I/O | ||
Please note that there are queues that buffer messages between all hardware instances ( | Please note that there are queues that buffer messages between all hardware instances ( Interface chips, central processing unit, storage ) and threads which can be tuned. The following queues can be monitored to see where you need to change settings in the Allegro: | ||
=== NIC Hardware Queue === | === NIC Hardware Queue === | ||
| Line 20: | Line 20: | ||
This first queue is between the Network card and processing unit. Whenever the I/O threads are too slow to consume all packets from the built-in or extension ports, the '''hardware miss''' counter of the interface statistics will increase over time. | This first queue is between the Network card and processing unit. Whenever the I/O threads are too slow to consume all packets from the built-in or extension ports, the '''hardware miss''' counter of the interface statistics will increase over time. | ||
The load of the I/O threads can be checked at the '''Info''' → '''System Info''' → '''Load'''. | The load of the I/O threads can be checked at the '''Info''' → '''System Info''' → '''Load'''. | ||
If this counter increases regularly, you can try the following settings: | If this counter increases regularly, you can try the following settings: | ||
| Line 27: | Line 27: | ||
* increase the number of I/O threads: [[#More I/O Threads]] | * increase the number of I/O threads: [[#More I/O Threads]] | ||
=== Analyzer | === Analyzer queue === | ||
The Allegro has a packet queue between I/O threads and the analyzer threads. | The Allegro has a packet queue between I/O threads and the analyzer threads. There are 3 scenarios where the an queue overload can happen. | ||
==== Low analyzer load and drop ==== | |||
This happens if the Allegro is in a powersafe mode and a sudden peak of traffic is monitored. The solution is to use the [[#Sink Mode]] and to activate the '''Analyzer queue overcommit''' at '''Settings''' → '''Global settings''' → '''Expert settings'''. This allows to buffer sudden network bursts. | |||
[[File:Queue overcommit.png|600px]] | |||
==== High analyzer load ==== | |||
==== not-balanced Analyzer load ==== | |||
Please note that the increasing counters at the initial startup of the Allegro can be normal under high load scenarios for a few seconds. | Please note that the increasing counters at the initial startup of the Allegro can be normal under high load scenarios for a few seconds. | ||
=== DB Queue === | === DB Queue === | ||
== Packet Drops in Queue == | == Packet Drops in Queue == | ||
Revision as of 13:08, 5 May 2020
About
This guide is about performance optimization of the Allegro for specific use cases. By default, the Allegro runs in a configuration that fits for the majority of users and you do not need to change any parameter of the configuration.
High Level Allegro System Layout
The Allegro has various different units that can be optimized. These units are:
- I/O Threads: responsible for all packet I/O operations
- Analyzer Threads: responsible for decoding and most of the DB operations
- DB Threads: optional threads offloading memory intensive database operations, see DB mode
- Disk I/O
Please note that there are queues that buffer messages between all hardware instances ( Interface chips, central processing unit, storage ) and threads which can be tuned. The following queues can be monitored to see where you need to change settings in the Allegro:
NIC Hardware Queue
This first queue is between the Network card and processing unit. Whenever the I/O threads are too slow to consume all packets from the built-in or extension ports, the hardware miss counter of the interface statistics will increase over time.
The load of the I/O threads can be checked at the Info → System Info → Load.
If this counter increases regularly, you can try the following settings:
- use sink mode instead of bridge mode, see below #Sink Mode
- increase the number of I/O threads: #More I/O Threads
Analyzer queue
The Allegro has a packet queue between I/O threads and the analyzer threads. There are 3 scenarios where the an queue overload can happen.
Low analyzer load and drop
This happens if the Allegro is in a powersafe mode and a sudden peak of traffic is monitored. The solution is to use the #Sink Mode and to activate the Analyzer queue overcommit at Settings → Global settings → Expert settings. This allows to buffer sudden network bursts.
High analyzer load
not-balanced Analyzer load
Please note that the increasing counters at the initial startup of the Allegro can be normal under high load scenarios for a few seconds.
DB Queue
Packet Drops in Queue
The Allegro Network Multimeter will inform you if a packet loss has been happened. The Allegro has certain hardware and software queues that can drop packet due overload.
Optimizations
Sink Mode
For configuration, please see: Global_settings#Packet_processing_mode
The bridge mode requires approximately 10 % more load on the I/O Threads as this threads have to sent the traffic to the corresponding network interface for forwarding. Allegro recommends to use