If you notice that network interfaces for Metacloud Hypervisors (MHVs) or Metacloud Control Planes (MCPs) are dropping packets, this is not necessarily a sign of performance problems in your cloud, especially if the interfaces are bonded in active-backup mode.
Packet loss is the failure of data packets to reach a destination. Depending on other network conditions, such as traffic surges, a high ratio of packet loss to overall packet traffic volume can signify performance issues, especially with data such as live-streaming video. In the case of text files or Web pages, even the loss of a single packet can compromise the delivered content if there is no reliable protocol for retransmission.
Why Packet Loss is Expected with Active-Backup Bonding
Bonding is the aggregation of multiple network interface cards (NICs), combining their bandwidth into a single connection. Benefits of bonding include redundant links, fault tolerance, load-balancing of traffic, and high availability.
Different, configurable bonding "modes" address different use cases. Active-backup, which is the default mode of Metacloud deployments, incorporates one active "slave" and one inactive backup slave. If the active slave fails, the backup becomes active. The primary use case for this mode is fault tolerance.
The backup slave is set to inactive in the bonding configuration but receives all broadcast traffic. Consistent with normal Linux network activity, it drops broadcast packets because of the inactive status. This is unlikely to cause any performance issues. See the following section for steps to determine if a backup slave is dropping packets.
Viewing Network Performance
You can view network performance statistics for MCPs and MHVs, including the volume of packet traffic and percentage of dropped packets, by using the Live Stats display. You must be a Metacloud Administrator to access Live Stats:
- Log into the Dashboard and click Controllers or Hypervisors under the Admin tab, depending on what you want to see.
- Click View Live Stats for the MCP or MHV you want to see.
- Click the Network tab on the Live Stats page.
The Dashboard shows activity for bonds and for individual NICs, but does not explicitly indicate which two NICs are in a bond. You can infer that by which two NICs have network traffic.
In the following screen shot, eth1 and eth0 show traffic, while the other four NICs do not, so it is likely that eth1 and eth0 are bonded and the others are not in use. Since eth1 is not dropping packets, it is probably the active slave. eth0, which accumulates drops, is probably the backup.
Noticing Other Issues
While packet loss is expected with active-backup bonding, it may be caused by other factors. Some observable issues, such as the following, could indicate that packet loss has a more serious cause:
- packet loss for both slaves in a bond
- extraordinarily high percentage of packet loss
- performance problems, such as failed data transmission or delivery of content
- other network issues, such as traffic surges
If you notice these or similar issues, submit a Support request. Also, Support has access to bonding configurations for MCPs and MHVs and can confirm the active and backup slaves in a given bond.