[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[LDM #ANS-486206]: Optimizing LDM configuration



Carrie,

> Another quick question for you folks.  I am working on getting our LDM 
> configuration optimized. I am using the plots generated by plotMetrics and I 
> wanted to verify I am interpreting the plots correctly.  The plots I have the 
> greatest questions are plot 1, plot 2 , and plot 8. I have gone through each 
> of the plots and gave a brief description of what I see and what I think we 
> should be looking for. Please correct me where I am wrong.   I have also 
> attached .gifs of the 3 plots and the output from the ldmadmin config.
> 
> Plot 1 - Queue Size - Number of Bytes Used in Queue Vs. Time
> We want to ensure we have a large enough queue so we do not have any data 
> loss. It appears the traces are hitting the ceiling at the 3GB mark which I 
> assume means I need a larger queue.  How large of queue is to large?

You want your queue to be large enough to tide you over any planned or 
unplanned outages in the distribution network.  We recommend that the oldest 
product in an operational queue be at least one hour old, so you also need to 
check the "Age" plot.

> Plot 2 - Product Count - Number of Products in Queue vs. Time
> The number of slots is set to default. The number of slots  will limit the 
> number of products, do not let the count hit a fixed ceiling, it they do you 
> will need to change the slot size in the config.

This is similar to the "Queue Size" plot: it, too, must be interpreted in light 
of the "Age" plot.

The queue is always going to be limited by something: either the number of 
slots or the size of the data section. 

> Plot 3 - LDM Connections - LDM Connections vs. Time
> Our current plot is 0, we have no one connected to our LDM, we are only 
> receiving data, not sending.

Just FYI, we call such LDM-s a "leaf node".

> Plot 4 - Space - Space vs. Time
> Watch the Used Memory+Swap for memory leak, you will notice an steady 
> increase in space if a leak exists.

Yup.

> Plot 5 - CPU-Modes - CPU-Modes vs. Time
> Watch for high I/O waits.

This plot, basically, tells you if your machine is powerful enough to do what 
you want and, if not, where the bottleneck is.

> Plot 6 - Context Switches - CPU Context Switch Rate vs. Time
> ? More important for developers, dealing with threads, what do I need to be 
> looking for?

It's a plot of how many things the CPU has to do, so it's kinda like Plot 5.  
If it ramps up and you start falling behind in the processing of the data 
product, then you need to process less or get a faster system.

> Plot 7 - Load-Average - Load-Average vs. Time
> Watch for an increasing or abnormal loads.

Yup.

> Plot 8 - Age - Age of Oldest Product in Queue vs. Time
> Want the age of the products to be greater than 1 hour to avoid data loss.  
> Can you explain this a bit more for me?

By default, the LDM will ask its upstream LDM-s for data from the last hour 
upon reconnection and if the product-queue is empty.  You risk not receiving 
all possible data if the upstream LDM-s don't have all of the last hour's worth.

This is more of an issue for relay nodes (as opposed to leaf nodes).

> Many thanks for all your help,
> Carrie

Regards,
Steve Emmerson

Ticket Details
===================
Ticket ID: ANS-486206
Department: Support LDM
Priority: Normal
Status: Closed


NOTE: All email exchanges with Unidata User Support are recorded in the Unidata inquiry tracking system and then made publicly available through the web. If you do not want to have your interactions made available in this way, you must let us know in each email you send to us.