So Yesterday I did some testing with different distributions running live on CD or DVD. I was looking for performance while running Live. This may not matter too some distributions as running Live is seen as a form of testing and then a method of install. Depending on what you’re testing application performance may not matter. For Synergy Linux it does because we have an option to install Live to SSD media, or for that matter just run off a CD and user a USB key as a permanent home instead of a machine’s hard drive. I launched multiple applications but focused primary on Firefox, as it’s a known memory hog and takes at least a few seconds to come up on most normal installs. This gave me something to measure. Now I know this isn’t a entirely accurate way of getting dependable results, but it gives a starting point.Here are the results with an explanation below:
Fedora 16 (latest rpms, CD 654M)
Boot 2:45
Launching Firefox 28s
Install 5:07
Mandriva 2011 (DVD 1.7G)
Boot 2:50
Launching Firefox 22s
Install 44:01
Mageia (CD 700M)
Boot 2:41
Launch Firefox 21s
Install 15:10
Chakra (CD 700M)
Boot 1:42
Launch Firefox (not on CD)
Install 9:48
Pardus (DVD 1.3G)
Boot 1:38
Launch Firefox 14s
Install N/A
This was done on a 2nd Generation Intel Mac Book Pro with 3G or Ram and a Intel Core 2 T7400 (2.16GHz). Now I know this is no where near a 800Mhz EeePC with 512M of ram, but these initial tests (I would like to do more on an EeePC with 512M) were on a machine that is now somewhat common. So think of this as a semi-control.
Explanation of 3 Tests:
Boot: was times from the selection of the boot option till X was initialized (not KDM necessarily though all of these were KDE based).
Launch Firefox: The time at which the Firefox icon was double clicked to the window being fully loaded (this was done after the desktop completely loaded).
Install: This was how long the installer took to copy files to the installation media. This does not include partition, user or system management options in the installer.
Firefox versions.
For Fedora Firefox 6 was used. The other distributions all ran Firefox 5 with the exception of Mageia, that ran 4. What’s interesting though is there was supposed to be launch time improvements to 5 so I expect Mandriva to beat Mageia just for that fact, however as you can see that was not the case. If we were to just go off Firefox launch time then Firefox 6 is very slow or Fedora LiveCD performance is lacking, coming in at almost 30 seconds.
Performance.
Unfortunately Fedora was rather unimpressive with performance, but came in on top with install because of their installation method. I would like to do more tests to see if their solution of create a 3G Ext4 disk image and then compressing it causes issues. Most distributions squashfs a folder instead of a file. The add overhead of using a loopback to mount a diskimage along with the unneeded journalling in Ext4 could be why it was rather slow with booting and launching applications.
Pardus goes a little bit in the opposite direction. Although they come in second as well with install time, they also have the fastest launch times with applications (not just firefox) and the difference it noticeable. All around Pardus seems like the best contender. However there’s at least one definite reason why.. Squashfs compression.. Pardus for their 2011 release switched to LZO compression instead of using LZMA. While you get no where near the compression ratios (smaller size) with LZO that LZMA offers (Larger ISO Size) you get faster (way faster) decompression time. Meaning boot up would be faster, file launch faster and install faster (if they offered it) then those using similar methods, like Mandriva or Mageia. Fedora however may already be at a disadvantage with their squashfs method plus the usage of LZMA.
Conclusion.
The reason Pardus switched to LZO is because they switched media from CD to DVD. If I was to switch to Synergy Linux to LZO that would mean a larger download size, and ISO size which may defeat the purpose of running a compressed install on small HD space devices like the EeePC Surf (2G). So there’s obviously reasons to use LZMA and keep a smaller ISO footprint. That’s a make of break deal. However more testing needs to go into Fedora’s method and reasoning for the use of the squashfsing as diskimage instead of just the filesystem itself. Does using a diskimage create another layer and a slow down in performance? Leave your thought comments and considerations on the forum.