RE: Unknown Exadata Cell Parameter - "_cell_fc_toresilver_limit_chdrs"

I have no idea.  

IF in fact my interpretation is right at all, I still don’t know whether this is regarding the trigger to start “resilvering” or the amount of “resilvering” operations per unit time (or something else I have not thought of).  

Lots of “resilvering” could quite possibly take place without affecting data delivery to users at all.  

Probably you need to seek the detailed explanation of this “hidden” parameter with Oracle development or someone formerly with Oracle development. It is entirely possible that it is not even used for anything and is just an artifact of something someone was thinking of doing or testing.  

Good luck, and remember, when someone writes OCI I think of Oracle Call Interface before context might lead me to consider that they are writing about time sharing.  

mwf  

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Osman DINC Sent: Friday, December 15, 2023 7:16 AM
To: Mark W. Farnham
Cc: oracle-l_at_freelists.org
Subject: Re: Unknown Exadata Cell Parameter - "_cell_fc_toresilver_limit_chdrs"  

Hi Mark,  

Thanks for your valuable feedback. Its(_cell_fc_toresilver_limit_chdrs) default value is 390000.  

Please correct me if i got you wrong, I interpret changing it to 6000000 as increasing the change directives required to start resilvering as 15 times more, thus Resilvering will occur 15 times less frequently.

Less resilvering means less Internal IO.  

But what can be the trade off of decreasing resilvering frequency?

As far as i know, resilvering should occur in situations like flash card change, a bad mirror detection on scrubbing. On standard daily activities, there should be no resilvering operation.  

Thanks. Regards.  

Osman DİNÇ              

Mark W. Farnham <mwf_at_rsiz.com>, 15 Ara 2023 Cum, 02:16 tarihinde şunu yazdı:

I don’t have any documentation on that, but reading the alphacrypt, I’d say that if you have a bad “plex” (that is old school called a mirror, as in image, even though it doggone well better not be reversed), and oracle is rebuilding it from one or two other still valid images, this is some limit at the cell and fiber channel? level of the rates from change directives? to process.  

It could be something entirely different, but that matches your comment. Unless you have something being rebuilt at the media image level, this shouldn’t matter (if I have the meaning anything like correct.)  

The old notion was that if someone shut Oracle down to rebuild plexes on the physical media, then it was pedal to the metal. But if you were up and running (with either two or one remaining good plexes), then Oracle would be rebuilding the broken plexes but not be piggy and starve the users for i/o.  

I could be completely wrong.  

mwf  

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Laurentiu Oprea Sent: Thursday, December 14, 2023 4:55 PM To: dinch.osman_at_gmail.com
Cc: ORACLE-L (oracle-l_at_freelists.org) Subject: Re: Unknown Exadata Cell Parameter - "_cell_fc_toresilver_limit_chdrs"  

Out of curiosity, did you tried to measure the performance (defined as duration) of a set queries with and without this parameter set?  

Thanks  

On Thu, Dec 14, 2023, 22:22 Osman DINC <dinch.osman_at_gmail.com> wrote:

Hi all Oracle enthusiasts,  

Do you have any information about the "_cell_fc_toresilver_limit_chdrs" parameter?

It is set on my environment (X7-2) cellinit.ora. It is an undocumented and non-default configuration.  

I could not find any information about it, but it is commented as "It is set to reduce Internal IO on exadata cells". We do not have prior knowledge about it.  

Also I have screenshots of AWR reports - Exadata Statistics - Top IO Reasons by MB section.

(Before and after parameter change)  

Before _cell_fc_toresilver_limit_chdrs parameter is set:

https://drive.google.com/file/d/1PEcjUYAuG8SNTq1vfXLl0fyoeoffSLhq/view?usp=drive_link  

After "_cell_fc_toresilver_limit_chdrs"=6000000

https://drive.google.com/file/d/12tDsDvt22oFyEaQtRdMsXcCYduw9992A/view?usp=drive_link  

According to the screenshots, It reduces Internal IO done by storage servers dramatically, but there may be a trade-off which i do not know.  

I want to clear this parameter as it is a non-default configuration. It was set on image version 20.1.5.0.0.201209 and it has been retained for years.  

Exadata image version is 22.1.17 now.

If anybody can share more information about what this parameter does, I will be glad.    

Regards,

Osman DİNÇ.    

--
http://www.freelists.org/webpage/oracle-l