04-02-2025 06:17 PM - edited 04-02-2025 06:18 PM
Sorry to take up some of your precious time but I have a small question. We have a 1gbps dedicated internet (Fiber) AT&T link and I'm trying to determine if the total load (inbound + outbound) is a little high for the link. Since this is full duplex should I assume the total capacity of inbound and outbound should be 2gbps? IF I look at the below utilization graph the peak would maybe get above 70% utilization which is a little on the high side however I would not worry much if the total capacity is actually 2gbps for a duplex link. Inbound is blue and outbound is red. Feedback is appreciated.
04-02-2025 06:27 PM
Yes, typically a duplex link's bandwidth is its bandwidth in either direction. So, yes, you should have two gig, a gig in both directions.
As to whether to be concerned about a particular load percentage, a low value isn't always good and conversely a high value isn't always bad; you need additional information to make such a determination.
04-02-2025 07:39 PM
I understand and agree. Your input helps me better to make that determination. thanks again.
04-03-2025 02:33 AM
". . . you need additional information to make such a determination."
Both @Ramblin Tech and @Giuseppe Larosa have provided some of the information for aquiring information for making a determination whether you have sufficient bandwidth but the true key is your goals for performance.
For example if doing a bulk data transfer, and if link is over subscribed, ideally link should use all 100% of its bandwidth for the duration of the transfer. Further, in something like classical TCP, queuing and drops are to be expected, although within certain bounds.
For some kinds of traffic, like VoIP, they too can work just fine at 100% but for such, the moment you exceed 100%, the app is impaired, although degree of impairment depends on latency and drops.
Bandwidth utilization, alone, is almost worthless for determining whether you have adequate bandwidth.
Further, additional stats can be useless too if you don't understand how to analyze them to what a certain app considers acceptable. For example, FTP and VoIP have very different bandwidth needs yet both can run fine at 100% or have major problems at 1%, going on just typical link bandwidth utilization measurements.
04-03-2025 09:43 AM
Oh, laugh, I haven't make my usual statement on the importance of QoS.
Basically, if you want predictable performance, either you need to have no oversubscription end-to-end or you need QoS. However, using QoS, effectively, pretty much overlaps the analysis you, ideally, need to do to determine what's the necessary amount of bandwidth. However, effective usage of QoS, also requires understanding how to use it to accomplish service levels, with the necessary amount of bandwidth. I.e. figuring out how much bandwidth app X needs is one problem, knowing how to guarantee that, with QoS, is the subsequent problem.
Commonly, to avoid all the forgoing, often more and more bandwidth is acquired, until users stop complaining about poor network performance, since, generally, oversubscription of bandwidth is the norm, users continue to complain (even when bandwidth utilization graphs, show "low" utilization).
One story I like to tell, years ago, had a remote site with two E3s, that during business hours, each E3 was constantly pegged at 100% utilization. Across those links, we ran everything, from VoIP, video conferencing, video streaming, Enterprise transactional apps, email, client backups, etc. All worked fine, but QoS managed the bandwidth.
04-02-2025 07:35 PM - edited 04-02-2025 07:37 PM
A packet communications channel has two states: busy sending a packet (100% utilized) and idle (0% utilized). To say that a channel had 70% utilization is to say that during the measurement interval, the channel was 100% busy sending packets 70% of the time. Is 70% utilization of a channel too high? Probably, but possibly not: what if the measurement interval was only 10 millisecond long while preceding and subsequent measurement intervals were idle? What if the utilization was only 20% during a 10 minute interval, meaning that it was 100% utilized for a total of 2 minutes. Conceivably, it could have been 100% utilized for 120 consecutive seconds. All that to say: link utilization stats are interesting, but do not necessarily tell you if a link is over-utilized or not.
Keep in mind that packet traffic is bursty; packets tend to arrive in bursts rather than being uniformly spaced out (if all packets had uniform inter-arrival rates, then 99% utilization would not be too high). Any packets that arrive at the egress interface for transmission over that channel when it was busy have to be queued until the channel is idle. The implication of this is that queueing is a key metric to monitor to determine if a channel is over-utilized. Except that queues tend to fill and drain so quickly that monitoring them directly is virtually impossible. What can you measure then? You can measure the effects of queueing.
What effects of queueing can you directly measure? Queue drops and latency. When a queue fills up, any additional packets arriving have no buffer space and are dropped. Switches & routers have counters for packets dropped from queues, so measure those with CLI/SNMP/telemetry. Packets not dropped from an over-utilized channel will be delayed as they work their way to the head of queue for transmission. That queueing time adds latency that can be measured. The best way to measure latency is via synthetic traffic generated and monitored by performance tools like IPSLA/TWAMP/ThousandEyes/Accedian. These tools can also measure loss of the synthetic traffic to corroborate your queue drop measurements.
BTW… yes, the link is full-duplex, so its capacity is 1Gbps in each direction.
04-02-2025 07:43 PM
very helpful, thanks.
04-02-2025 11:27 PM
Hello @hmc2500 ,
it depends on the device involved some software based routers have GE interfaces but are not able to support full line rate bidirectional.
In your case looking at the traffic loads graphs your device is likely capable to support 1 Gbps per direction. As noted by @Ramblin Tech the graphs are produced using SNMP queries every X seconds where X by default is 300 seconds . The values that are visualized are actually the difference between byte counters at time N*X and time (N-1)*X then multiplied by 8 ( bits/byte).
The 64 bit versions of the byte counters are recommended in their OID path there is "HC".
Hope to help
Giuseppe
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide