Marc's Public Blog - Linux Hacking

Let's start with the dreaded network manager: well, believe it or not, but after 2Y+ of being broken for me, it finally worked. It still had a few rough edges, but it finally would manage my interfaces without crashing.
It still has a few rough edges, and things I had to do were:

It's not perfect but I can now have it seemlessly switch between my wired and wireless networks while keeping the same IP (like I used to do on debian years before networkmanager even existed).

Then comes upstart. It's an overdue parallel init implementation (my cubemate Richard Gooch also wrote one years ago that never caught on unfortunately). It obviously has a few warts: some boots still hang for me, exim isn't supported (booh!), but it should hopefully be fully supported by all soon.

pulseaudio, well, what can I say? It still does not work as shipped. I tried to make it work, but it still randomly hung mplayer and caused other random problems. If it's all this trouble just to get per application volume levels which I can't get in alsa I'm told, and moving audio to other destinations while it's play, which I don't need, it's definitely not worth it for me.
My fix was:
dpkg --purge pulseaudio pulseaudio-esound-compat pulseaudio-module-udev pulseaudio-module-gconf pulseaudio-module-x11.
See this thread on phoronix, this comment sums what I think:
I've never seen such a buggy thing like pulseaudio on linux regular distros.
It's fine people is debugging this stuff, but I think they should not release it till it's stable on 99% of configurations.

Acpid is still kind of broken as shipped for thinkpads, especially if you don't run gnome power manager, which I couldn't find a good reason to run: it's pretty featureless and braindead. Thankfully the acpid author works at my company too and I finally figured out with him that debian killed the logging in acpid, which explains why I couldn't see any output from it anymore when trying to make it work. The good news is that after fixing that I was able to create the right events to support all my Fn keys, and the only thing that really changed in the last years is that I can now call scripts from pm-utils to sleep and hibernate as opposed to doing it all in shell in acpid. Not too bad.

For the rest, it looked like a pretty smooth upgrade: radeonhd still does not work reliably on my Z61p's M56GL [Mobility FireGL V5200] (crashes on 3D, known problem), but radeon still works fine.

Anyway, the point of this post was to talk about benchmarks. I was curious to know how much faster switching from a 5400RPM drive to a 7200RPM drive would be (I have a 500GB drive, SSD in that size is not too much an option yet, and I'm not really hurting for speed anyway).
The next thing was to see how much dmcrypt was slowing me down.

Ready for it?
dmcrypt gave me virtually no slowdown on a kernel build and a mere 4-5 seconds to get Enlightenment started from boot (about 40 seconds from lilo prompt to Enlightenment loaded and opening its screen with Eterms already loaded). So, I really don't care about 4 seconds of boot time if it makes virtually no difference once the system is running. The best part was that switching from a 160GB 7200 RPM drive without dmcrypt to a 500GB 7200 RPM drive with dmcrypt, opening a maildir folder with 50,000 messages actually went DOWN from 50-ish seconds to 16-ish seconds!

7200 vs 5400RPM gave a modest boot speed improvement (8 seconds) and considering I boot once a month on average, it's really not something I care much for. Opening a maildir folder is faster by a few seconds (15%-ish) but nothing earth shattering. Building a kernel is maybe 20 seconds faster. All this to say that 7200 RPM is not something I would kill or pay load of extra money to get.

Anyway, switching from my old 5400RPM drive with dmcrypt to a 7200RPM drive with dmcrypt (all using ext4 on a freshly layed out filesystem).

Oh, the other test I did quickly was to see whether HIMEM64G (enabling PAE and a 3rd level indirection required for me to see 64GB instead of 3GB), and indeed HIMEM64G is a tad slower than running without HIMEM4G as I had feared, but not by a whole lot (looks like less than 5%), so I'm willing to ignore it.

Raw data below. It shows time from lilo prompt to X starting, and E fully launched as well as how long it takes to open my rcvd and snd maildir folders on a cold cache. Then, I build a full featured kernels with 4 concurrent threads.

5400 RPM 160GB seagate, no dmcrypt, HIMEM64G

boot1: 27 X start / 50 E ready / 0:58 rcvds / 0:52 snd

kernel build #1:
real    23m24.829s
user    39m59.454s
sys     4m16.530s

kernel build #2:
real    22m47.524s
user    39m52.953s
sys     4m16.569s

boot 2: 23 X start / 41 E ready / 0:58 rcvd / 0:34 snd

kernel build #1:
real    23m3.003s
user    40m2.183s
sys     4m17.200s

kernel build #2:
real    22m38.859s
user    40m0.710s
sys     4m21.401s

boot 3: 25 X start / 46 E ready / 0:58 rcvd / 0:34 snd

kernel build #1:
real    24m19.828s
user    40m1.283s
sys     4m35.624s

kernel build #2:
real    23m51.198s
user    39m58.137s
sys     4m37.391s

boot 4: 23 X start / 42 E ready / 0:60 rcvd / 0:33 snd

----------------------------------------------------------------------------

7200 RPM 160GB seagate, no dmcrypt, HIMEM64G

boot 1: 21 X start / 36 E ready / 0:48 rcvd / 0:25 snd

kernel build #1:
real    23m22.448s
user    39m49.538s
sys     4m13.604s

boot 2: 21 X start / 34 E ready / 1:03 rcvd / 0:21 snd

kernel build #1:
real    22m58.988s
user    39m53.308s
sys     4m12.700s

kernel build #2:
real    22m43.217s
user    39m50.925s
sys     4m14.976s

boot 3: 21 X start / 35 E ready / 0:39 rcvd / 0:20 snd

----------------------------------------------------------------------------
7200 RPM 500GB hitachi, dmcrypt, HIMEM64G

boot 1: 27 X start / 40 E ready / 0:18 rcvd / 0:19 snd
(cryptsetup unlock takes a few seconds)

kernel build #1:
real    23m14.759s
user    39m56.628s
sys     4m11.470s

kernel build #2:
real    22m46.385s
user    39m52.201s
sys     4m11.900s

boot 2: 26 X start / 39 ready / 0:14 rcvd / 0:18 snd

kernel build #1:
real    23m5.225s
user    39m59.785s
sys     4m11.592s

real    22m47.208s
user    39m52.411s
sys     4m11.155s

----------------------------------------------------------------------------
7200 RPM 500GB hitachi, dmcrypt, HIMEM4G
boot 1: 24 X start / 36 ready / 0:14 rcvd / 0:20 snd

build #1:
real    23m0.698s
user    39m45.571s
sys     4m3.892s

build #2:
real    22m39.361s
user    39m42.988s
sys     4m3.924s