by sramuschkat
Hi,
for server monitoring I have to define thresholds ... do you have any plans for more intelligence in your product like anomaly detection or outlier detection so I do not have to define thresholds anymore?
Sven
Like (4)
Reply
Replies (11)
Hello Sven,
You had asked this at a right time. We just released Anomaly Detection in Beta for our Premier pack users.
This platform will initially detect anomaly based on seasonal values. This will use last 4 weeks seasonal data for determining anomaly. Currently it just pushes the anomaly information to events of the monitor. We yet to bring in alerting support for this feature. We will do it after we stabilize things in this.
We will have our other packs to support Anomaly Detection very soon.
Please find few screenshots of Anomaly Reports.
I would look forward to here any suggestion or use case you would like to solve using this platform.
Raghavan
Attachments
screencapture-pre-site24x7-app-client-1476794174230.png
Size: 293.83 KB
Like (0)
Reply
by sramuschkat
Hi,
we have it now in our Account and it is collecting data and already shows up Anomalies ... which at this stage have no meaning because auf to less data ...
Now I have these questions:
we have it now in our Account and it is collecting data and already shows up Anomalies ... which at this stage have no meaning because auf to less data ...
Now I have these questions:
- if there is an anomaly it currently shows up in the anomaly module in the table. But will this also listed somewhere inside the monitor module .. so I can see on the monitor level all anomalies in the past for that specific monitor?
- In your documentation you speak about 4 weeks of data to define the baseline. Is the anomaly is just detected based on statistics (as an outlier) or is there also machine learning behind the scenes ... what happens e.g. when for a specific server every Tuesday from 10.00 am to 10:10 am there is >95% CPU consumption because of a scheduled job ... is this an anomoly or is the algorthim learning that this for that specific date&time is a normal state?
Sven
Like (0)
Reply
Hi Sven,
1. Yes anomalies will also be displayed under individual monitors details page. We are working on an Events timeline chart, where anomalies along with outages will be shown as events. This will be live by this week. So you can see monitor level all anomalies in the past for a specific monitor.
2. Coming to your second question, as of now Anomaly is detected based on statistics and no machine learning is involved. We will be comparing the present data collected with last four weeks data of same timeslot. For example, if there is a >90% CPU consumption for a server on Wednesday 11am to 11:30am, then this will be compared with the data of 11am to 12pm of last four week's Wednesdays. If there is any hike in the present value when compared to baseline value, then it will be detected as Anomaly. So in your case, if a server is having >95% of CPU Consumption every Tuesday from 10am to 10:10am, we will compare this value with data from time 10am to 11am of last four week's Tuesdays. Then this spike is considered as normal in that time period and is not detected as Anomaly since every Tuesday 10am to 10:10am, we are having higher values.
Ramya
1. Yes anomalies will also be displayed under individual monitors details page. We are working on an Events timeline chart, where anomalies along with outages will be shown as events. This will be live by this week. So you can see monitor level all anomalies in the past for a specific monitor.
2. Coming to your second question, as of now Anomaly is detected based on statistics and no machine learning is involved. We will be comparing the present data collected with last four weeks data of same timeslot. For example, if there is a >90% CPU consumption for a server on Wednesday 11am to 11:30am, then this will be compared with the data of 11am to 12pm of last four week's Wednesdays. If there is any hike in the present value when compared to baseline value, then it will be detected as Anomaly. So in your case, if a server is having >95% of CPU Consumption every Tuesday from 10am to 10:10am, we will compare this value with data from time 10am to 11am of last four week's Tuesdays. Then this spike is considered as normal in that time period and is not detected as Anomaly since every Tuesday 10am to 10:10am, we are having higher values.
Ramya
Like (0)
Reply
by sramuschkat
Hi Ramya,
thanks for that clarification.
By the way ... I love your tool and actually for me it's one of the best in the market ... and I really love the speed of improving and releasing new stuff in your tool over the last couple of months ... so well done Site24x7 Team.
Sven
Like (0)
Reply
Hi Sven,
Thanks for your appreciation. We always welcome you to provide us your feedback and suggestions to improve the functionality of Site24x7.
Keep going.
Raghavan
Like (0)
Reply
by sramuschkat
Does Annomly Detection cover all Metrics available in your System?
- APM Metrics?
- EC2 / RDS Metrics?
- SQLServer Stuff?
- etc?
Is there is list which is supported and which not?
Sven
Like (0)
Reply
Hi Sven,
As of now, Anomaly detection covers few metrics - "Response Time" for Internet Services monitors and "CPU Used percent, Memory Used Percent" for Server monitor.
We will enable other attributes of different monitors based on customer suggestions in future.
Ramya
As of now, Anomaly detection covers few metrics - "Response Time" for Internet Services monitors and "CPU Used percent, Memory Used Percent" for Server monitor.
We will enable other attributes of different monitors based on customer suggestions in future.
Ramya
Like (0)
Reply
Hi,
makes sense for these ...
for the Upcoming AWS ELB support in your Product Latency and number of requests is on my wishlist ...
Sven
makes sense for these ...
for the Upcoming AWS ELB support in your Product Latency and number of requests is on my wishlist ...
Sven
Like (0)
Reply
by sramuschkat
Hi,
can you please give me an update on this:
"We are working on an Events timeline chart, where anomalies along with outages will be shown as events. This will be live by this week. So you can see monitor level all anomalies in the past for a specific monitor."
And also when beta of Anomaly will be over.
Sven
Like (0)
Reply
Hi Sven,
Events Timeline is live now. Anomalies occurred will be shown in the events timeline in individual monitor details and also monitor groups details page.
Also, we are going to support Anomaly detection for other monitor types like AWS ELB. It will be enabled by this week. We are working on enhancements like alerting for Anomaly detection. Will get back to you about these enhancements once it is stabilized.
Thanks,
Ramya
Like (0)
Reply