Linux 2.6.30's best five features

Windows and Mac OS updates every few years. Windows 7 arrives on October 22nd and Apple's Snow Leopard will show up in September. The Linux kernel, the heart of Linux distributions, however, gets updated every few months.

What this means for you is that Windows and Mac OS are taking large, slow steps, while Linux is constantly evolving. Thus, Linux's changes may not be as big from version to version, but they tend to be more thoroughly tested and stable. What most users will like in this distribution starts with a faster boot-up for Linux.

1. Fast boot. Older versions of Linux spend a lot of time scanning for hard drives and other storage devices and then partitions on each of them. This eats up a lot of milliseconds because it looks for them one at a time. With the 2.6.30 boot-up, however, instead of waiting for this to get done the rest of the kernel continues to boot-up. At the same time, the storage devices are being checked in parallel, two or more at a time, to further improve the system's boot speed.

There are other efforts afoot to speed up Linux's boot times. The upshot of all this work will be to keep Linux the fastest booting operating system well into the future.

2. Storage improvements. Speaking of storage devices, there's a long, laundry list of file system improvements. I won't go into most of those in detail. Suffice it to say that no matter what file system you use either locally or on a network, chances are that it's performance and stability has been improved. For a high-level view of these changes see the Linux Kernel Newbie 2.6.30 reference page.

I will mention one issue though simply because, as Jonathan Corbet, Linux kernel developer and journalist put it, "Long, highly-technical, and animated discussion threads are certainly not unheard of on the linux-kernel mailing list. Even by linux-kernel standards, though, the thread that followed the 2.6.29 announcement was impressive." You can say that again.

The argument... ah discussion was over how file systems and block I/O (input/output) using the fsync() function in Linux should work. The really simple version of this discussion is that fsync has defaulted to forcing a system to write file system journal and related file data to be written immediately. Most I/O schedulers though push reads over writes. On a non-journaling file system, that's not a big deal. But, a journal write has to go through immediately and it can take up a lot of time while it's doing it.

On Ext3, probably the most widely used Linux file system, the result is that Ext3 is very stable, because it makes sure those journal writes go through, but at the same time it's very slow, once more because of those journal writes. You can argue almost endlessly over how to handle this problem, or even that Ext3 fsync function runs perfectly fine. Linus Torvalds, however, finally came down on the side of making the writes faster.

The arguments continue though on how to handle fsync(). And, in addition, side discussions on how to handle file reads, writes and creation continue on. For users most of this doesn't matter, developers who get down and dirty with file-systems though should continue to pay close attention.

3) Ext4 tuning. Linux's new Ext4 file system has been in the works for several years now. It's now being used in major Linux distributions like Ubuntu 9.04, and it's working well. That said, Ext4 has gotten numerous minor changes to improve its stability and performance.

I've been switching my Linux systems to Ext4 over the last few months. If you've been considering making the switch, wait until your distribution adopts the 2.6.30 kernel, and give it a try. I think you'll be pleased.

4) Kernel Integrity Management. Linux is more secure than most other operating systems. Notice, though, that I say it's more secure. I don't say, and I'd be an idiot if I did, that it's completely secure. Nothing is in this world. The operating system took a big step forward in making it harder for any would be cracker to break it though with the introduction of Integrity Management.

This is an old idea that's finally made it into the kernel. What it boils down to is that checks the integrity of files and their metadata when they're called by the operating system using an EVM (extended verification module) code. If a file appears to have been tampered with, the system can lock down the its use and notify the administrator that mischief is afoot.

While SE-Linux (Security Enhanced Linux) is far more useful for protecting most users, I can see Integrity Management being very handy for Linux devices that don't get a lot of maintenance such as Wi-Fi routers. Attacks on devices are begining to happen and a simple way to lock them down if their files have been changed strikes me as a really handy feature.

5) Network file system caching. How do you speed up a hard drive, or anything else with a file system on it for that matter? You use a cache. Now, with the adoption of FS-Cache, you can use caching with networked file systems.

Right now it only works with NFS (Network File System) and AFS (Andrew File System). These network file systems tend to be used in Unix and Linux-only shops, but there's no reason why you can't use FS-Cache on top of any file system that's network accessible.

I tend to be suspicious of network caching since it's all too easy to lose a network connection, which means you can be left with a real mess between what the server thinks has been changed, added, and saved and what your local cache thinks has been saved. FS-Cache addresses this problem of cache coherency by using journaling on the cache so you can bring the local and remote file systems back into agreement.

While 2.6.30 may not be the most exciting Linux kernel release, it does include several very solid and important improvements. Personally, I plan on switching my servers over to 2.6.30-based distributions as soon as they become available. If your concerns are mostly with the Linux desktop though I wouldn't be in that much of a hurry, most of the updates are more important for server administrators than desktop users.