PredictLinear usage is unclear

I’m struggling in using the predictLinear. What I want to do is to graph the current throughput for one of our services, and predict how that will trend in the future.

Hey Dustin, what are you struggling with specifically?

1 Like

@dustin.pauze It would also be helpful to have further details so that I can get one of our support team members to help also. Thanks :slight_smile:

Here’s what I’m trying to achieve:

Another option would be a simple trend line like this:

Here’s what I’ve attempted:

SELECT predictLinear(*, 30 days) as ‘Throughput’ FROM Transaction WHERE appName = ‘myApp’ TIMESERIES 30 days

That clearly doesn’t work because one must select a specific attribute.
Even when I try and select the transaction duration, I still don’t see any prediction:

SELECT predictLinear(duration, 14 days) FROM Transaction WHERE appName = ‘myApp’ TIMESERIES 60 minutes

It’s entirely possible I’m misusing the ‘predictLinear’ function and it’s supposed to only give me a singular value, but that feels like a missed opportunity imo.

2 Likes

Yea, I don’t think this is a use suited for predictlinear(). It’s really meant for things that grow over time. Throughput doesn’t not. From the docs it says its good for things like disk usage and its not recommended for timeseries or things that do not increase in a linear way. Sort of predictable from the name I guess :wink:

Sure would be great to predict throughput out 14 days.

I accept this might not be a good application of this function, but I don’t completely agree that throughput doesn’t grow over time. I agree that generally when looking at a short window of time, throughput probably doesn’t grow predictably, but over longer periods of time or when onboarding customers, there is a predictable trend to throughput.

This does beg the question : How does one use this function correctly?

1 Like

I think as it suggests its good with things that grow over time. Throughput more often than not goes up and down. On average your throughput can grow over time.

So are you just interested in seeing of the machine can predict your increase projections? Like this month you do 10MM requests and last month you did 8MM so maybe next month you will do 12MM?

There are a lot of factors. I like to use come of the compare options to show month over month to look for obvious trends.

1 Like

Looks like you all are working this out. If you need further support around this please let me know :slight_smile:

So are you just interested in seeing of the machine can predict your increase projections? Like this month you do 10MM requests and last month you did 8MM so maybe next month you will do 12MM?

Yes, that sums this up. It would also be helpful to visualize this prediction. We can use a month over month view, but what we’re really after is to see the slope of the prediction and be able to pick arbitrary points along that slope to gauge the predicted value at that point.

Yea, that sure would be nice. Charts really aren’t keeping up with the other innovations. So we are also a customer of Datadog and I do recognize the graphic you have above :slight_smile:

@6MM not sure how to feel about the DD reference but its this type of discussion that is informative to our team :slight_smile:

Well, we sure wish New Relic would have better dashboarding. We enjoy our DD dashboards for so many reasons. NR has been working hard to get NR1 out the door and cleaned up… and its’ been years in the making. I’m sympathetic. Maybe empathetic. I still want better widgets and ways to visualize data from New Relic. I’d be thrilled beyond belief if NR just came up to parity with DD on dashboarding. If you want feedback in a thread feel free to have the team responsible for the widgets etc post a thread. I’ll fill it :wink:

1 Like

That’s good to know! @6MM :star_struck: