NET.performance issue: MTU for qr-*, qg-*, tap-* interfaces in network namespaces of hdcp and l3 agents should be adjusted to external and private MTU
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Fuel for OpenStack |
Fix Released
|
High
|
Bartłomiej Piotrowski | ||
6.0.x |
Won't Fix
|
High
|
MOS Maintenance | ||
6.1.x |
Fix Released
|
High
|
Bartłomiej Piotrowski |
Bug Description
DHCP and L3 agents create qr-*, qg-*, tap-* interfaces with default (1500) MTU.
After increasing MTU inside instances, instance will emit network packets up to 9000 payload size. If that packets has ho DF flags we will have segmentation and network performance (VM to outside world) degradation. Packets with DF flags will be dropped.
Incoming (from outside world to, for example, floating IP) traffic will be emitted from namespace to VM as packets with up to 1500 payload size and may me segmented to. In any FLOATING-IP cases increasing MTU to 9000 on VM useless.
Neutron has 'network_
We should use this option for adjusting MTU on agents interfaces.
no longer affects: | fuel/7.0.x |
tags: | added: l23network |
Changed in fuel: | |
assignee: | Bartlomiej Piotrowski (bpiotrowski) → Sergey Vasilenko (xenolog) |
tags: | added: on-verification |
tags: | removed: on-verification |
Sergey told me that Sergey Kolekonov is implementing a function for getting base (phys) interface for any endpoint. After that it'll be sensible to deal with this issue.