Forum Discussion

Stuart_Weenig's avatar
10 months ago

Can you compare and contrast Dexda's capabilities with PagerDuty's alert grouping feature

We currently send alerts to PagerDuty because it has really good ML based alert grouping. From there it goes on to our ticketing system. Looks like this might be able to replace that capability (which would simplify things greatly on our side).

It seems like grouped alerts are called “episodes”. Will episodes have the ability to be routed similar to the way alerts are routed through alert rules? 

In PD, in order to train the ML engine that some alerts belong together, all we have to do is combine two “incidents” (what PD calls a Dexda episode). That not only groups all the alerts from both groups into one group, but it also trains the ML to do better next time. Will Dexda have a similar capability?

In PD, in order to train the ML engine that some alerts DON’T belong together, all we have to do is split one or more alerts from one “incident” (what PD calls a Dexda episode) into multiple incidents. This not only ungroups the alerts into separate groups, but also trains the ML to do better next time. Will Dexda have a similar capability?

Dexda will automatically re-cluster alerts when it identifies a more optimal clustering option” - does this mean it will change the grouping of alerts that it has already grouped?

How is multi-tenancy handled? There’s a issue with the tenant id currently that’s making it undefined for all our alerts. I don’t mind using tenant ID as long as we can get that issue fixed.

9 Replies

  • FYI, that sounds exactly like what Pager Duty provides for us, except that pager duty calls it an incident instead of an insight. Other than that, so far the alert grouping is the same.

    Is Dexda a separate product, or is it something that's going to be part of the platform?

  • ...

    Is Dexda a separate product, or is it something that's going to be part of the platform?

    Hi @Ranjan , could you answer this question, please.

  • Can you use Dexda without service now?

    Hi @Ranjan , could you answer this question as well, please.

  • Ranjan's avatar
    Ranjan
    Icon for Product Manager rankProduct Manager

    Dexda works quite differently than the workflow you described for PagerDuty.

    In Dexda Alerts are correlated together based on OOB or customer ML models into “Insights”.

    Insights can be manually or through pre-configured setting raised as “Incident” into ServiceNow.

    ML Models can be altered by the users to improve the model efficiency across any type of Alerts where these ML Models are applicable.

    Yes, there are rules and workflows on Insights that can be configured in Dexda.

    This is a summary of operations inside Dexda with more details available in the Dexda support docs.

  • Sorry, I’m out. It only works for ServiceNow. Doesn’t actually use ML, you have to build rules. Also, how will it be priced? I’ve heard rumors that it’s going to be priced by event. 

    HMU when it actually does.

  • Ranjan's avatar
    Ranjan
    Icon for Product Manager rankProduct Manager

    @Stuart Weenig bit of half truths in your statements above.

    Dexda works with ServiceNow (both CMDB and ITSM) for now. Other integrations would come later.

    Dexda leverages built-in ML-models for Alert de-dup and correlations. Those models are open and customizable. Additionally, it leverages NLP for summarization.

    Anyone in LM Community if interested please reach out to LM sales/CSM team for further details.

  • Dexda leverages built-in ML-models for Alert de-dup and correlations. Those models are open and customizable. Additionally, it leverages NLP for summarization.

    Great! So why would i have to build models based on CI properties matching with a specified percentage? I don’t have to do that with PagerDuty, which is how i get this functionality now. I just feed in the alerts, tell it which field in the alert to use as tenant id for multi-tenancy, and it just groups “similar” alerts. I train it after the fact if it gets it wrong by combining alerts or splitting grouped alerts manually. However, that’s just adding to the ML, not building out matching models/rules.

  • Dexda leverages built-in ML-models for Alert de-dup and correlations. Those models are open and customizable. Additionally, it leverages NLP for summarization.

    Great! So why would i have to build models based on CI properties matching with a specified percentage? I don’t have to do that with PagerDuty, which is how i get this functionality now. I just feed in the alerts, tell it which field in the alert to use as tenant id for multi-tenancy, and it just groups “similar” alerts. I train it after the fact if it gets it wrong by combining alerts or splitting grouped alerts manually. However, that’s just adding to the ML, not building out matching models/rules.

    Hey @Stuart Weenig working to provide you some additional details/answers for this one!