Hi Greg, > > The only data I am missing now is the memory used by each additional > > sysfs file we create. We need to know, as Hans objected that too many > > sysfs files could have a negative impact on memory consumption. I dug > > down the sysfs code yeterday evening to find out, but didn't find what > > I was looking for yet. I hope to get the answer this evening. > > sysfs files are very light. The "large" memory structures for a ram > based file system are the dentry and inode structures. sysfs now > creates them on the fly when they are needed, and if we have memory > pressure on our internal kernel caches, they are freed. > > So in short, don't worry about creating new sysfs files, it's not an > issue. The people running 20,000 disks on a 31bit s390 system have > already done the hard work for you :) Thanks for your enlightened comment on this. Now, this still raises the question of how much the dentry and inode structures take. You say that they are created on the fly, but let's imagine a hardware monitoring utility polling the sysfs files on a regular basis, I guess that these structures could be considered as "permanently allocated" for this set of sysfs files, so the dentry and inode sizes would start to matter. >From /proc/slabinfo, I get the following sizes: x86: dentry 124 bytes, inode 336 bytes x86_64: dentry 200 bytes, inode 608 bytes For 33 additional files, counting one of each structure per file (which might not be correct, I'm really ignorant of how it actually works), it's around 17 kB on x86 and 29 kB on x86_64. While still not unacceptable (IMHO at least), this is much more than my first estimation. Thanks, -- Jean Delvare
