Hello,
Reading release notes, 9.16.25 is the point release where the default
memory allocator was switched:
https://downloads.isc.org/isc/bind9/9.16.32/doc/arm/html/notes.html#notes-for-bind-9-16-25
The issue linked https://gitlab.isc.org/isc-projects/bind9/-/issues/2398
is giving lots of interesting informations.
The only thing explaining your observation is that you test machine is
very low cpu count powered and that the trades of made by the new
allocator is negative (from mem consumption point of view) in this
specific artificial corner case : very low powered machine with very
high zone/record count. Useless in real world scenario.
You could do the test yourself with command-line option -M internal with
>= 9.16.25
For the other parameter I recalled from memory, it was
--with-tuning=large/--with-tuning=small but was large since 9.16.0.
Emmanuel.
Le 08/08/2022 à 12:29, Dmitri Pavlov a écrit :
Hi again Ondřej, Emmanuel,
I hope you enjoyed the weekend ?!
Just trying to summarize where we after running a few tests against "zoo of machines" and
comparing to your lab results. Please, see the attached "Symmary.jpg"
What we see is the RAM utilization before "<" 9.16.25 (regardless of kernel version
and GCC) lower than in >= 9.16.25 including 9.18.5 and 9.19.3. Pre 9.16.x results are without
jemalloc. And the results are pretty much consistent before 9.16.25 and after. It is 2 GB higher
in 9.18.x with ~ 100 mil records data set. I understand your previous notes about ridiculousness
of the given scenario, patch levels, bugs in earlier releases etc But anyway, sorry , the
question is still open for us -> is there really a change in 9.18.x codebase that allows lower
RAM utilization as compared vs pre 9.16.25 versions
https://kb.isc.org/docs/bind-memory-consumption-explained? Or maybe particularly in this scenario,
when we adding 100 mil records into to the cache, does not fall under KB scope?
Your response about the KB correctness will help to deliver a better optimized
business decision.
Thank you very much in advice,
Dmitri.
-Original Message-
From: bind-users On Behalf Of Ondrej Surý
Sent: Thursday, August 4, 2022 7:48 PM
To: Emmanuel Fusté
Cc: bind-users@lists.isc.org
Subject: Re: High memory consumption in bind 9.18.2
What Emmanuel said…
--
Ondřej Surý — ISC (He/Him)
My working hours and your working hours may be different. Please do not feel
obligated to reply outside your normal working hours.
On 4. 8. 2022, at 19:15, Emmanuel Fusté wrote:
Le 04/08/2022 à 17:48, Dmitri Pavlov a écrit
Therefore, a very small request. Would it be possible on your side to run the same
experiment as with (BIND 9.16.32 / BIND 9.18.6 / BIND 9.19.4) one more time but
with BIND 9.16.21 (or any other version in 9.16.x <25 range )?
Why not the opposite ? Why do you insist to run obsolete/inferior patch level
version ? Who want to run something older than the latest patch release of one
maintained version and even more a more than ten patch level apart ?
The memory consomption diff is not an argument as it is simply ridiculous vs
the used scenario.
Reproduce the 9.16.32 scenario, and if it reproduce Ondřej result, the
conclusion will be evident : bugs in the older patch level as you clearly
reproduced the 9.18 usage which you could surely reproduce with the 9.19 series
too.
Do you really prefer to run buggy but less memory hungry version ?
I understand that you want to have answer to you questions. Simply do the
complete exercise and you will have answers. Don't ask people to do them for
you.
Emmanuel.
PS: there where a switch on the default runtime config switch to "big server"
mode sometimes during the 9.16 series if I recall correctly. It perhaps explain the diff.
--
Visit
https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.isc.org%2Fmailman%2Flistinfo%2Fbind-users&data=05%7C01%7Cdpavlov%40perforce.com%7Cbf656a835be74993002e08da76391910%7C95b666d19a7549ab95a38969fbcdc08c%7C0%7C0%7C637952284930827181%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=Kirvq3EkQe239keUeA18xvG%2FOxY8tAhs2yF9AxaoR0Y%3D&reserved=0
to unsubscribe from this list
ISC funds the development of this software with paid support subscriptions. Contact us at
https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.isc.org%2Fcontact%2F&data=05%7C01%7Cdpavlov%40perforce.com%7Cbf656a835be74993002e08da76391910%7C95b666d19a7549ab95a38969fbcdc08c%7C0%7C0%7C637952284930827181%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=V1HTBX29PPdDrjG6wHaJ6YU5BrHNymn1oNoaiidh6dk%3D&reserved=0
for more information.
bind-users mailing list
bind-users@lists.isc.org
https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.isc.org%2Fmailman%2Flistinfo%2Fbind-users&data=05%7C01%7Cdpavlov%40perforce.com%7Cbf656a835be74993002e08d