cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
622
Views
0
Helpful
1
Replies

QoS Drop Packet Normal?

PolarPanda
Level 1
Level 1

Hi there,

          I have a policy map applies to an interface and it has multiple class under it, like default, voice, video etc. I don't see any drops in any class except class default. The drop is very little, like 0.011%, and it's stable(not increasing or decreasing over time). We don't have any drops in the queue either. So we're experiencing microburst packet loss. I'm wondering if this is normal range? Do I need to pursue 0% drops? If we increase queue-limit value, will it cause more problem? Thank you. 

1 Accepted Solution

Accepted Solutions

Joseph W. Doherty
Hall of Fame
Hall of Fame
The general rule of thumb is not to be concerned until packet loss is 1% or greater. However, I believe that "rule" goes way back before we had "bulk" traffic, and if you have a 1% loss on every hop, it might rapidly add up. Something like microbursts might also mean you have a much, much higher "average" drop rate when they are happening, but if the microbursts are infrequent, your overall drop average is low. I.e. occasionally you have yucking performance.

So, IMO, any drops require analysis, but such analysis may show they are acceptable for your SLA needs. That though, is the real question, are your drops adverse, or not, to your needs.

As to whether extending the queue size will decrease drops, or cause other problems, to the first question, maybe, maybe not, or it might even increase overall drops. To the second question, also maybe or maybe not. The most common issue increasing queue size creates it can increase overall latency and/or jitter. Again, for the latter, there are other adverse possibilities too.

View solution in original post

1 Reply 1

Joseph W. Doherty
Hall of Fame
Hall of Fame
The general rule of thumb is not to be concerned until packet loss is 1% or greater. However, I believe that "rule" goes way back before we had "bulk" traffic, and if you have a 1% loss on every hop, it might rapidly add up. Something like microbursts might also mean you have a much, much higher "average" drop rate when they are happening, but if the microbursts are infrequent, your overall drop average is low. I.e. occasionally you have yucking performance.

So, IMO, any drops require analysis, but such analysis may show they are acceptable for your SLA needs. That though, is the real question, are your drops adverse, or not, to your needs.

As to whether extending the queue size will decrease drops, or cause other problems, to the first question, maybe, maybe not, or it might even increase overall drops. To the second question, also maybe or maybe not. The most common issue increasing queue size creates it can increase overall latency and/or jitter. Again, for the latter, there are other adverse possibilities too.
Review Cisco Networking for a $25 gift card