Malcom,
First of all you should determine which process takes 2GB. se_trn, se_sm?
How many buffers do you use for se_sm?
What kind of memory does it use? Virtual or physical? Can you send top
output for that process?
Ivan
Hello Ivan,****
>
> ** **
>
> Thank you very much for the response. ****
>
> ** **
>
> We do a bulk data load, and then incremental updates that occur in batch
> mode. In some systems, the bulk load can be as much as 3 GB, with
> incremental batch updates of 20-30 MB.****
>
> ** **
>
> We prevent access to the system during bulk and batch updates, so no
> queries occur. ****
>
> ** **
>
> The problem is that we noticed one of the Sedna systems use 2 G of ram
> memory, which basically dragged down an Amazon micro instance and made it
> unusable. (We stopped & started the instance and it worked correctly)****
>
> ** **
>
> We have not discovered the root cause of the memory problem, and are still
> investigating other concerns. (The OS and Sedna are the only thing we have
> running on the micro instance)****
>
> ** **
>
> We will start profiling Sedna request in an attempt discover root cause of
> some of issues.****
>
> ** **
>
> Thanks again for the quick response,****
>
> Malcolm****
>
> ** **
>
> ** **
>
> ** **
>
> ** **
>
> *From:* Ivan Shcheklein [mailto:shchekl...@gmail.com]
> *Sent:* Thursday, July 07, 2011 2:07 PM
> *To:* Malcolm Davis
> *Cc:* sedna-discussion@lists.sourceforge.net
> *Subject:* Re: [Sedna-discussion] Sedna memory requirement and
> configuration****
>
> ** **
>
> Hi Malcolm,****
>
> ** **
>
> I understand the problem of determining memory requirements is more
> complicated than just volume. There are the number of nodes to consider,
> and the number and type of the indexes.****
>
> ****
>
> I was curious if any formulas existed I could utilize to determine minimum
> memory requirements.****
>
> ** **
>
> No, there is no such formula. Apart from the data structure and indexes
> there is also one important factor - workload - i.e. queries/updates you
> run. Besides, what does enough mean? 100MB is *physically* enough to run any
> query on any data.****
>
> ** **
>
> I believe the only really effective approach to analyze queries. Run them
> and look how many blocks they read/write (this information is available in
> event log after session is closed). ****
>
> ** **
>
> Ivan Shcheklein,****
>
> Sedna Team****
>
------------------------------------------------------------------------------
All of the data generated in your IT infrastructure is seriously valuable.
Why? It contains a definitive record of application performance, security
threats, fraudulent activity, and more. Splunk takes this data and makes
sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-d2d-c2
_______________________________________________
Sedna-discussion mailing list
Sedna-discussion@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/sedna-discussion
