From mboxrd@z Thu Jan 1 00:00:00 1970 From: danielhilst@gmail.com (Daniel.) Date: Tue, 25 Oct 2016 16:03:02 -0200 Subject: What subsystem!? In-Reply-To: <65708.1477410153@turing-police.cc.vt.edu> References: <65708.1477410153@turing-police.cc.vt.edu> Message-ID: To: kernelnewbies@lists.kernelnewbies.org List-Id: kernelnewbies.lists.kernelnewbies.org Vladis, Thank you so much for your anwer. I had never thought that way, and make totally sense. Challenge accepted, I'll look for interesting applications and see how they interact with kernel. Thank you so much!!! Best regards!!! 2016-10-25 13:42 GMT-02:00 : > On Tue, 25 Oct 2016 10:33:47 -0200, "Daniel." said: >> I've been studying linux for some time now. I still see my self as an >> begginer, but that's not a bad thing. I struggling to decide what >> subsystem to focus right know. > > I've said it before, and I'll say it again: > > If you have to ask others which part of the kernel you should study, > maybe you shouldn't be doing kernel work. It's like a beginning > author asking other people if they should write a mystery or a western. > If you're only writing a murder mystery because somebody suggested it, > it's probably not going to be very good. > > Note that for most systems all the *real* magic (and hard work) aren't in > the kernel, they're in userspace. > > Let's look at the Internet of Things as an example. Yes, networking is > going to be important - but the kernel just gives userspace (for instance) > TCP over IPv6. All the code to actually *do* something with it is out in > userspace. > > Just the other day, a large part of the Internet fell apart because of > a DDoS attack on some important DNS servers. The root cause? Default > passwords and other userspace security issues on a lot of DVRs and > security cameras. A userspace issue. > > I'm looking at buying some LED-based light bulbs that can change color on > command. Kernel support for the networking? Get a packet, send a packet. > All the code that says "during early evening, use a 5K color temperature, > and slowly roll it back to 3.2K around midnight, but an hour later in the > kitchen"? Userspace. > > You mention database tuning. Another case where you *really* need to > understand what userspace is doing before you can do much in the kernel. > Linux kernel hacking for tuning our Oracle servers? Zip. Zero. None. > We just create yet another VM with X amount of RAM and Y amount of > disk on a fiberchannel-connected EMC VNX storage unit. However, our > DBAs spend a *lot* of time tweaking Oracle parameters in userspace, and > my co-worker who administers that storage keeps busy tuning it - but > he has a nice userspace GUI that handles all the hard work so he can > concentrate on higher-level issues like proper tiering of a high-activity > disk. > > My current job is building petabyte-scale file systems for HPC clusters - > and although I end up tweaking kernel parameters a fair amount, there's > actually not much kernel hacking involved. What *is* needed is the ability > to talk to the computational scientists, and find out things like what > blocksize their software wants to use, whether it's sequential or random > access, what degree of parallelism is needed, and so on... > > You hopefully notice a pattern here... Find an application that *you* think > is interesting - and first learn the userspace end of it. You may not need > to know very much about the kernel at all - and if you *do*, you'll have > a better idea of which part of the kernel and what you need to know.... -- "Do or do not. There is no try" Yoda Master