Request to remove blacklisted configuration settings

[nvader]

coming from "another" NAS product. I was getting 110+mb /sec to/from my server.
I upgraded to much faster hardware and 3x the ram & moved to TN Scale.
my smb transfer speeds were 40mb/sec.
I reconfigured smb with the same settings i used on my old nas.
I know there is debate about what is and isn't effective, but these are the settings that had worked for me.
I added these and got to around 60 mb/sec after adding the settings and restarting.

strict locking = no
read raw = yes
write raw = yes
oplocks = yes
max xmit = 65536
deadtime = 15
getwd cache = yes
aio read size = 16384
aio write size = 16384

Then I tried to add these

socket options = TCP_NODELAY IPTOS_LOWDELAY SO_SNDBUF=131072 SO_RCVBUF=131072

To which I received the following message.

socket options is a blacklisted auxiliary parameter. Changes to this parameter are not permitted

before I go any further, I would kindly like to request this restriction be removed. tweaking of a parameter should be up to the end user and not prevented by the host software. Even if it's under the advanced your on your own if you do this check box. it really should still be up to me if I want to add those parameters or not.
Having said that, I now needed to find a way to make those settings active. to which I did.
My result was that I was now getting over 110 mb/sec again.
Again I know there is debate about the effectiveness of those settings. but in my case, they clearly made a huge difference.
I went from 40 to 60 to 110 by only adding those additional parameters.
Thanks!

 

[Redcoat]

before I go any further, I would kindly like to request this restriction be removed. tweaking of a parameter should be up to the end user and not prevented by the host software. Even if it's under the advanced your on your own if you do this check box. it really should still be up to me if I want to add those parameters or not.

Improvement/Feature requests are made on Jira through the Report a Bug link on the masthead. Log in and use the create link to select your option.

 

[nvader]

Improvement/Feature requests are made on Jira through the Report a Bug link on the masthead. Log in and use the create link to select your option.

View attachment 55151

OK since it wasn't an actual bug, I wasn't sure of the process. thanks for the info.

 

[morganL]

coming from "another" NAS product. I was getting 110+mb /sec to/from my server.
I upgraded to much faster hardware and 3x the ram & moved to TN Scale.
my smb transfer speeds were 40mb/sec.
I reconfigured smb with the same settings i used on my old nas.
I know there is debate about what is and isn't effective, but these are the settings that had worked for me.
I added these and got to around 60 mb/sec after adding the settings and restarting.

Then I tried to add these

To which I received the following message.

before I go any further, I would kindly like to request this restriction be removed. tweaking of a parameter should be up to the end user and not prevented by the host software. Even if it's under the advanced your on your own if you do this check box. it really should still be up to me if I want to add those parameters or not.
Having said that, I now needed to find a way to make those settings active. to which I did.
My result was that I was now getting over 110 mb/sec again.
Again I know there is debate about the effectiveness of those settings. but in my case, they clearly made a huge difference.
I went from 40 to 60 to 110 by only adding those additional parameters.
Thanks!

Thanks for the write-up... it would be useful to know which of these parameters have the most impact. Was it TCPNOEDELAY?

Sometimes there are reason for these restrictions... eg the system can be less reliable. Or the parameters need to be set in another place because they impact other protocols. Its very useful to review these.. as there are Linux/FreeBSD differences.

 

[Volts]

socket options = TCP_NODELAY is a Samba default since forever.

Many of the other options are Samba defaults too - read raw, write raw, oplocks, getwd cache.

The changed values for aio read size and aio write size are outdated:

aio read size (S)
...
The only reasonable values for this parameter are 0 (no async I/O)
and 1 (always do async I/O).
...
Default: aio read size = 1
...

If SO_SNDBUF and SO_RCVBUF are set, modern Linux will disable TCP autotuning ... that's likely to make things worse, but I'm very curious to hear otherwise. Those seem like the changes most likely to have any impact. Did you test them in isolation, by any chance?

You might share more information about your config and how you're testing speed. It's possible you can get advice that will help accomplish your speed goals, or explain the current behavior.

 

[nvader]

socket options = TCP_NODELAY is a Samba default since forever.

Many of the other options are Samba defaults too - read raw, write raw, oplocks, getwd cache.

The changed values for aio read size and aio write size are outdated:



If SO_SNDBUF and SO_RCVBUF are set, modern Linux will disable TCP autotuning ... that's likely to make things worse, but I'm very curious to hear otherwise. Those seem like the changes most likely to have any impact. Did you test them in isolation, by any chance?

You might share more information about your config and how you're testing speed. It's possible you can get advice that will help accomplish your speed goals, or explain the current behavior.

Thanks for the reply @morganL @Volts after searching for answer, I do see some people say xxx settings are dated, not helpful. default. etc. I don't claim to know the specifics (yet) however the results do speak for themselves. add those settings to a brand new system and my speed jump 60+ mb/sec.
I am currently migrating many terabytes of data from my old server to the new. After that is done, I will see if I can isolate what settings are causing the speed gains and will report back my findings for analysis.