SphinUX OS Claims To Be ~150% Faster Than GNU/Linux
Phoronix: SphinUX OS Claims To Be ~150% Faster Than GNU/Linux
SphinUX OS is an open-source POSIX-compatible operating system developed under the GPLv3 and running the Egyptian LSX Kernel Architecture. This open-source operating system claims to be much faster than Linux and that its memory usage can even be 3x less! This is an operating system with some of the most wildest performance claims we have ever seen...
If it supports only POSIX and not the range of APIs Linux supports - it's only interesting to folks who already use fringe OSes like BeOS, DOS etc.
I'm pretty sure we're presented with the usual situation: the size of the OS is generally inversely proportional to its speed. When Linux was much smaller there were also astonishing benchmark results, when it matured to support a much wider array of APIs, drivers and hw it did so generally at the expense of speed.
Not really sure WTF these guys are trying to pull, but what I see on their sourceforge page is one directory with the linux kernel, one with gnu mach, and one with minix with a fake name.
Is there any (technical) reference to the LSX kernel?
Maybe I'm dumb, but I cannot find it!
I didn't hear such bullshit since a long time. BeOS, Haiku, DOS are much slower than bigger operating systems. It's because of architecture not because of being small or not. Furthermore, Linux doesn't use 90% of its features on a common PC, so your point is invalid.
Originally Posted by mark45
According to this video: http://www.youtube.com/watch?v=AqupsTFyuGE ... it boots and is recognized as a 2.6.32 Linux kernel (look on the boot sequence, at 17:06 ), so it looks to me that this SphinuX is more or less a modularizing effort inside the Linux kernel that may reduce in some use-cases the IO and/or maybe some kernel modules.
I think that the 100% and 300% kind of speedups refer just to the use-cases when this new infrastructure reduces the IO (so in an I/O kind of benchmark it may work like 2-3 times faster) and the memory is again related with the minimalist kind of loading of the drivers.
As it looks, it uses standard compilers (like GCC) and similarly standard file systems (Ext3 and Ext4), so the Phoronix like benchmarking, which stresses the user-space I think it will behave (more or less) like a 2.6.32 kernel with some patches on it. Also, the source doesn't say 100% or things like this performance speedup against what. I think that is in case of Ext4 with encryption vs Ext4 with Xor++ (their encryption scheme).
So based on what is written, I can expect the following:
- some boot scenarios would boot faster on a similar package selection and using a bit less kernel memory
- for embedded cases, the kernel can be "up-to" 300% smaller memory
- in low memory systems, where it will be more IO, the smaller kernel (Sphinx) may work a bit better
- on disk encrypted disks, a faster encryption scheme, which may be "good enough" may be faster than Ext4 than the default Ubuntu with Encryption
What would not happen:
- Sphinux would run the same CPU bound codes (which are computed in user-space or on video card, but not on kernel) to be more than 5% faster/slower than 2.6.20+ kernels, so Michael will be able to "bust" the myth
- Database benchmarks or IO tests without encryption on Ext4 should work similarly as a 2.6.32 kernel
- Loading KDE or whatever environment with similar package selection will look and feel as any bare-metal Debian with the same package selection. SphinuX would have more bugs, but other than this, will feel very "upstreamy". Both memory wise and performance
But Michael is right, the specifics should be told, writing "100% speedup" or "300% less memory" without putting some strings attached is a bit misleading (to say it kindly) and it would be great if the value numbers come with a methodology about how it can be this way.
What the hell are you smoking?
Originally Posted by kraftman
Anyone who has tried BeOS, Haiku or FreeDOS will instantly tell you about how fast they are compared to other OS's.
Methinks you haven't tried any of these, or at least haven't tried trolling before and giving it a shot for the first time.
It depends a lot on what you mean by "faster". Those OS's will certainly boot faster. Whether they'll run benchmarks faster is unlikely, but i'm sure it depends on the benchmark.
Originally Posted by intellivision
It depends on the benchmark, too. Typical desktop users will "feel" latency much faster than other things. BeOS might be snappier (less latency) at typical tasks but maybe file copying or the like is significantly slower.
Originally Posted by smitty3268
You see the same thing in Linux with the competing CPU/IO schedulers. An algorithm that maintains maximum file transfer bandwidth on spinning media is pretty different than an algorithm for minimimizing file access on an SSD, after all.
Fast is in many ways relative and even can be measured, all of these OSes are not as fast and let me say why:
Originally Posted by intellivision
- Haiku and BeOS are micro-kernels that work with Servers that offer Services in the OS. The intercommunication between these systems is slower than a hybrid OS like Windows NT, OS X or Linux. All these OSes merged the subsystems where it makes sense, and loads as modules parts that are not that important. So a kernel heavy benchmark (with multiple context switches) will not work (that) well on BeOS/Haiku
- Haiku had in the past problems compiling with newer GCCs (and it still doesn't compile as far as I know with the 4.7 release) so the generated code on these platforms matches the compiler capabilities. So people using newer compilers will get all the optimizations of these newer compilers
- bigger OSes offer optimized DLLs/libSOs for the target instructions set, and a compatibility one that is fast enough. Look if you use Windows XP for example that you have p3.sys, or athlon.sys as parts of the kernel that are loaded (based on the machine's instruction set). Similarly LLVM (which is in fact is used real time compiler) can optimize shaders for machines that do not support specific operations in OS X/Linux (as part of Gallium).
- BeOS, and Haiku support 32 bit code, and big OSes that support 64 bit offer the capability of 64 bit processing, which again in (mathematical) benchmarks at least, are faster code
If you mean about: how fast it boots, the older OSes many times do recognize much less hardware and they need to load less services because they don't support much more of the functionality the new OSes support. BeOS has similar features with Windows 2000, and this OS would load fairly fast by today's standard, but it would not offer anti-aliasing on fonts (I know that Haiku offers this, but Haiku is still a much less featured than modern Linux/Windows), many features that many application offer (and they have to be loaded on disk), including some that are just for the sake of open standards and easier to be debugged (like configuration files written in Xml format). All these features slow the OS starting time, but 1 minute to boot was true from Windows 95 era, and the spinning disks at least did not increase by as many orders of magnitude the access time (compared with disk space).
FreeDOS itself boots very fast as it doesn't: detect and assign a hardwared mapping and device initialization to all devices in your system. The application have to do this (like using the sound card!). Based on this, the "fast" is all about offering nothing. If you try to add stuff in FreeDOS you will find only what it doesn't support, and if you load everything, it loads/runs much slower: you don't have video acceleration, 64 bit processing, you don't have a TCP stack, so you have to write your own, and as you add more functionality, at the end the OS, the applications will run as a Frankenstein. Amazingly, there is a very new GCC for FreeDOS, the DJGPP distribution (4.7.3), which is a great achievement if you would ask me.