DSM-G600, DNS-3xx and NSA-220 Hack Forum

uksparky wrote:

Hi jcard and everyone else for that matter,

I recently posted that i was experiencing slower network speeds than i hoped, particularly download speeds (around 60% of my maximum). I finally decided that there must be some issue causing this since some wgets were ok from sources in the same country, however other speeds were struggling.

I therefore followed the tcp tweaks at http://wwwx.cs.unc.edu/~sparkst/howto/n … tuning.php and now am experiencing much better download speeds ~(95-100% of my maximum). I'm not recommending to do these changes without testing/backing up settings and without knowledge of what they do and how they work. However I believe this might be something that jcard wants to investigate further, in order to improve overall network performance of the dns-323.

Regards,
sparky

Hi,

Yes, there are several tuning (or tweaks) than can be applied, but be aware, some are old, and linux is a moving target. Currently linux adapts itself to the environment, so most tweaks might not always work as expected.

Anyhow, I think that users could post their tweaks, with a short comment on what they are intended for and what improvements did they notice.
After some independent confirmation I could put these tweaks, commented, in Alt-F /etc/sysctl.conf, with the user comment on what they are intended for. It would then be up to the user to un-comment and use them.

Do you mind sharing your findings?
The two/three lines comment should say if they are intended for NFS, or cifs, for Gbps nets, for long distance connections, etc, and what are the noticed improvements.

Thanks.

oxygen wrote:

these steps are usually not useful, they polute the few ram with buffers which are not required for transfers with a low bandwidth - delay product (e.g. lan transfers).

they might improve performance download speed for the internet but that's not the general use case. i'd generally discourage increasing the tcp buffer sizes. this will have a negative impact on general performance.

You re right (and most tweaks are for desktop computers with several GB of main memory), but who knows if uksparky is not doing a secondary backup from his Malaysian office to his Hungarian branch? He reports almost a two fold increase in his download bandwidth. It looks like for him it works (at the expense of what we don't know)

That's why I suggested a two lines text explaining the situation at hand.

I'm pretty sure that is someone publish a working samba tweak for the DNS that improves cifs bandwith at the expense of a degraded NFS and ftp and rsync and wget and whatever, then most users wouldn't mind using it, as I believe that most users are using samba and not those "exotic" things.

I don't use tweaks for myself, none give me full satisfaction

BTW, 9K jumbo frames are a lost of memory with a very small performance improvement over 7.5K jumbo frames. That is a fact. http://code.google.com/p/alt-f/issues/d … &can=1 (max MTU=9000 bytes, allocated 3*4096=12288 bytes, gives 3288 unused bytes, times 130 reserved slots by the driver totals 427440 of unused and precious memory!)

I think there is a common case where people have just got their device set up and want to copy 2TB of data to it where even relatively small speed improvements can shave hours off the time it takes for the initial upload.  Using more buffers while one transfer protocol is moving data would probably be a fair trade off...