2009/1/15 Matt Sealey
On Thu, Jan 15, 2009 at 5:55 AM, Rob OpenSuSE
wrote: 2009/1/15
:
It's not about "turning them off by default so nobody can use them", it's about turning off services which can be CONFIRMED as unused on the current system.
When you & Larry complained, without making specific examples, via generalised comments, I actually looked into the differences between 10.3 & 11.1. I see reasonable choices, the fact is, I cannot file Bugs that something is loaded and activated on my system when I use it. It would be hugely unfair on developers reputation to allow such unsubstantiated claims go unchallenged. They are in many cases actively working to reduce memory consumption, whilst actually increasing functionality. Actually I was interested to investigating why you had problems, but it may be Power specific, as the systems I tested on, were not showing the problems you saw. In private (I think) you then mentioned an 80KiB wastage thanks to device mapper, and a claimed 0.5s increase on boot time. Ask yourself if that is going to cause performance problems on a typical desktop system, where even small machines, have 512 MiB RAM. Furthermore, and you really deserve a huge flame for this, when the tuneable that frees up memory was suggested to you, which will free up several MB of pages, you dismiss that, despite going after KiB reductions in areas you understand. There seems to be a fixed mindset based on what is a significant performance issue is on an embedded system, despite the fact that openSUSE is a general desktop and server OS. Unless you can come up with specific examples of performance issues, it appears that you're just hand waving and ranting about bloat etc. It is clear that there has been work done in the past to speed up boot times, and folk agree that that is a benefit. However being abusive to ppl who look into what you are saying, and report what they find, suggest that you actually seek sycophants, rather than real information or trouble shooting.
Initrd building - a discussion you're also chipping into on this and another list - is one example of where you build something which only supports the desired drivers on the system. If your system uses pata_via and uhci, then these are put into the initrd so they can be loaded on early boot. This is determined by the state of the running system, and what udev did the last time it booted.
I agree and have pointed out, before in a bug report that this is may prove to be a problem. It is an issue that I have wanted to discuss, rather than file a specific bug report, but so far I have not seen evidence of it causing many problems, which makes it harder to build a solid case. Specific examples are much better than shrill handwaving.
Now, if you do not have LVM2 setup, dm-mod is still loaded and dm-mod is still started on boot. There is no need for this. The service can be deactivated without negative effect until someone needs to set up an LVM2 system.
I am using LVM, so I cannot investigate why device mapper would be erroneously loaded.
rpcbind/portmap is started, even though xinetd is not started and no NFS clients are installed. No other services depend on rpcbind/portmap on my Automatic Configuration-installed system. So why is it still enabled?
When I install nfs-kernel-server and enable xinetd for some service, rpcbind should be pulled in by those as and when it is needed, for the best efficiency of resource usage.
These are good questions, but these services are not the cause of the performance problems, that you complained of. For reasons that have been explained previously. Privately I advised using the possible security undesirability of having these services active (installed) when unecessary, rather than hammer the performance angle.
And while a few years ago one of the benefits of Linux was that it ran on any old hardware, where Windows was not an option, this is not the case any more.
openSUSE is not Linux, it has always had suggested hardware that was much greater than typical embedded systems, because it is a general purpose server and desktop OS.
It's a testament to how great openSUSE is that it DOES boot and DOES support our 128MB 400MHz board, but the trend seems to be that a generic server or desktop will have many gigabytes of RAM and many thousands of MHz of CPU and extremely fast peripherals. This is not born out by the requirements to install the system - 256MB of RAM (on the Efika we have to build a swap file on a USB key or activate an already existing swap partition).
We realise we are on the shitty end of the stick here, but we've noticed that the trend from 10.x to 11.x is that installation is fine, but the installed system has been gaining more and more hard dependencies on things that the vast majority of people may not even have activated (LVM2 etc.)
Are you filing bug reports, on these specifics? What I have been saying to you is, that on a range of "typical" systems, even quite old Intel ones, they are not causing end user problems. The example of the 128 MiB board, cannot be typical, where a desktop, or FF3 itself is likely to use that amount of RAM alone, despite work by developers made to reduce memory requirements, comparing 10.3->11.1 as I have, when you complain. Don't confuse non-agreement, with not listening. I've actually been the person who investigated claims because I took some interest, as I do test new software pre-release on "small" 256MiB systems and ones with slow CPUs. -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-factory+help@opensuse.org