I just joined the Clear Linux band wagon for its speed.Just made afirst test install on my dev desktop, and it is feeling good.
My ultimate goal is to install and configure Clear Linux on our production server. I absolutely do not care about desk top operability. i believe there are other OSes for that.
I am happy that there finally is a Linux geared towards server performance. And I am seeking bare-metal, ultra-low latency.
I will do my homework and go through the tutorials, etc.
However, for now if anybody can point me in the right direction for:
Customising/configuring the kernel for ultra low latency
building such a kernel (if it is different from the normal buikd process)
And last, of course, how to load the newly built kernel.
The next post/question will be about āportingā centOS/rHEL 3rd-party 9no source) librairies/packages.
Since you donāt know what parameters to tune, donāt know how to build it, and donāt know how to load it, probably you should just give up trying thatā¦
However, there are plenty of tunable such as (RCUs, HZ, AudIT, etc.), just to name a few, and they can all be configured one way or another, with different latency results.
āUltraā low latency and Determinism are difficult to obtain.
Clearly, Clear Linux, (LoL), is optimised for performance, and therefore will already have a fewo of these tunable built in a certain way. However, performance is usually related to throughput. Although it tends to benefit, to a small degree, to latency; but latency and throughput are āalmostā mutually exclusive, in many use cases.
Therefore, a kernel built for ultra low latency will be configured differently than when āpure performanceā (in general) is the goal.
So, I think thatās specific enough to stir the convo in the right direction, donāt you think?
You want someone else endeavors to find a set of kernel parameters to reliably reduce latency, which could takes weeks of work and still not guaranteed to work out well.
You want someone else to give you a detailed guide on how to build the kernel and how to load it.
Cheers 0xA1B2,
it canāt get any barer metal than this. Very interesting concept; I will certainly look into this. The challenge will be to get the 3rd party drivers to work with this. We have an fpga and a specialised NIC that are going to be tough to integrate. Iām on it, the challenge is on.
It does also open an other venue to investigate: the uni-kernels.The concept looks good on paper, but the implementations I came across were a bit too āprimitiveā (for lack of better words.)
Iām not sure what latency your referring to but Pulse Audio has latency that Ubuntu Studio dreams of.
Iām converting 16 bit audio to 32 bit, running it through a 32 band stereo EQ so 64 channels total and an Autogain and converting back to 16 bit in 140ms.
If you want faster get an A4000 Amiga with SCSI HDD, a Retina 24 bit graphics card and ZIII ram. Load your OS and software into a RAD drive. Make sure you get all the OS replacement software thatās been rewritten in assembly. Then unsolder the clock chip from the MB and solder in a socket so you can overclock it. Software loads faster than the video card can display it.