Minutes

BLUG MEET - 23rd April 2004

Hi Folks,

This is Surjo Das here again presenting the Minutes of the Meet for the BLUG April 2004 Meet. As usual, the meet was held at the Shantala Hall of Ashraya Hotel on Infantry Road. The theme for this month's meet was File Systems in Linux. For the first time this year, we had less than 50 people turning up for the meet. Only 44 people turned up for the meet. The meet should have started at 6 pm. But due to a low turnout and expectation of more people to attend the meet, it finally took off at 6:30 pm. Not new for our meets to start late ! Anyway, once about 35 people gathered at the hall, the BLUG Co-ordinator Kartik kicked off the meet with the usual welcome address and introduced the first speaker of the meet, Arun Raghavan whose talk was about 'An Introduction of File Systems'.

Arun's talk was a neat and detailed one. He covered the various file systems supported in Linux and their advantages and disadvantages. He also came up with suggestions as to what file system to use under various circumstances. Arun started off his talk by mentioning that over 20 years users have had the same view for any file system. Basically starting with the root directory and spanning downwards. This view has not changed at all. The various file systems in use today include, ext2, ext3, ReiserFS 3, ReiserFS 4, JFS, XFS, NTFS. All these file systems provide ACLs. JFS was derived from the database concepts. It maintains transaction logs. Its pretty consistent, far less likely for corruption and its startup time after a crash is very low. Gentoo Linux supports ext2, ext3, ReiserFS 3, XFS & JFS. Slackware supports ext2, ext3 and ReiserFS 3. Red Hat and Fedora Core support ext2 and ext3. Mandrake supports ext2, ext3, ReiserFS 3 and JFS.

After mentioning the various file systems supported in various distros, Arun then moved on with his talk explaining the properties of various file systems with their advantages and disadvantages. Ext2 is about 6 years old. It is the default file system on all major distros. It's a simple file system, very stable, not a journalled file system and running fsck on ext2 takes ages to complete depending on the level of damage done to it. Arun recommends Ext3 over ext2. Ext3 file system is ext2 with journaling capability. Migrating from ext2 to ext3 is uncomplicated. Using the tune2fs command and making the relevant changes in /etc/fstab does the trick. Ext3 is a simple and robust file system. However, Arun doesn't recommend it for high performance machines. Then moving on to Resier FS 3. This file system was written from scratch. Arun strongly recommended using this file system. It is used widely in Gentoo, Suse and Lindows. It uses enhanced B+ trees and tail packing. Tail packing basically helps save space. However Reiser FS is not recommended for servers.

Then Arun's talk moved on to JFS. It was developed by IBM for AIX and then ported to OS/2 which was made open source. It supports different block sizes. Directories are stored in 2 ways, viz., small directories in blocks and large directories in B+ trees. It uses extents for large files. The other features of JFS include dynamic inode allocation, online partition resizing, which is ideal when used with Logical Volume Management and it is also 64-bit ready. Then a small discussion followed between Arun and the audience on the question of the 64-bit readiness of JFS. Arun dispelled the doubts that other file systems were not 64-bit ready, the reference was made to the fact that 64-bit readiness of JFS meant that this file system can be used on Sparc, Itanium and Opteron based machines as well. Arun's talk then moved on to XFS. It was written by SGI for Irix. XFS main advantage is its industrial strength file system, its high performance which is ideal for big servers. Doesn't seem to work with Kernel 2.6.5 as per Arun's experience and not as reliable as ext3. The next file system discussed was ReiserFS 4 which is currently in beta and is expected to be in Kernel 2.6. Its much faster than its predecessor ReiserFS 3. ReiserFS 4 supports layering and plugins. And finally Arun came to WinFS which is believed to be the default file system for the next release of M$ Windows whenever that happens.

Arun then revealed some benchmarking test results on his Home PC using various file systems. The benchmark test involved copying the kernel tar file ~245 MB to the hard disk, untarring it, listing it recursively and deleting the kernel directory recursively. The benchmark test revealed that ext3 performed better across all the benchmark tests and is highly recommended. ReiserFS was not far behind. FAT32 performed miserably. Arun then concluded his talk and threw open the meet for a brief Q & A session. A good discussion followed among the audience and Arun about the way metadata is handled differently in different file systems, viz., ReiserFS & ext3, using tune2fs to convert from ext2 to ext3, results of running the benchmark test on SCSI and whether it would have concluded differently and recovering deleted files which is done much better in XFS & JFS. XFS is recommended on a high-end server using Samba with ACLs. The discussions were very lively.

After Arun's time at the limelight and his moments of glory were over, Kartik then announced the 2nd and last talk of the evening which was given by Sreekiran on CD based file systems. This happened to be the first meet being attended by Sreekiran and he gets to give a talk as well during his first meet. Congrats to Sreekiran ! May we have more talks from you in future meets. Sreekiran's talk was a brief one and provided some good bit of information on CD based file systems.

Sreekiran started his talk by giving everyone a history lesson as to how the CD based file systems began their evolution. The High Sierra Group that followed the Yellow Book specifications formed it and the various modes followed by the CD based file systems. The first well known CD based file system was the ISO 9660 8.3 standard. This file system allowed only 8 characters for the file and directory names and 3 letters for its extension. The ISO 9660 Level 2 file system allowed upto 32 characters for the file and directory names, which were however not supported under DOS as it understood only 8.3 standard. Then there was the Romeo file system developed by Adaptec, HFS developed by Apple Computers for the MacOS, UDF file system which supports upto 127 characters for filenames.

El Torito file system was devised to produce bootable ISO 9660 CD ROMs. This process basically involves having a bootable image of a 1.44 MB floppy on the CD which when booted from a PC, fools the system BIOS into believing that the CD is a floppy boot image. Creating a bootable CD requires a bootable floppy and transferring the image to the CD using the dd command. The next file system discussed was the Rock Ridge file system which allows longer filenames, deeper directory hierarchy and attributes with a UNIX system feel. The next file system discussed was the Joliet file system developed by M$ which allows longer filenames upto 64 characters. It is supported by Linux from Kernel 2.0.6 itself. The next file systems for CDs that was discussed was the hybrid CDs. Various software vendors who ship products for various operating systems on a single CD basically use this method. So it was imperative that the file system on the CD would be compatible with the respective operating systems on which the application needed to be installed. This process involved formatting various portions of the CD with the different file systems so that it would be recognized by the various operating systems in which the CD needed to be used. By conclusion, any file system can be created on a CD. Ext2 file system on a CD performs slower than others as per Sreekiran. A certain bug in ISO 9660 was circumvented in Kernel 2.4.26. Sreekiran then concluded his talk with various references and useful URLs.

As the talks ended slightly ahead of schedule the food wasn't ready as yet. So before everyone could start splurging themselves on the food, the meet was thrown open for a general discussion just short of a BOF session on file systems. Everyone's favourite seemed to be using encryption in file systems. A good lively discussion followed among the audience with all sections of the audience contributing. After about 10 minutes, the food was ready and Kartik announced the availability of food and everyone queued up to indulge in eating and socializing with the various BLUG Members. There were a few first timers in this meet. Apart from the speaker Sreekiran, I came across 3 guys from Texas Instruments who came to attend this meet out of curiosity and to understand more about the BLUG. A brief explanation was given to them about what we do and introduced them to Mahendra as one of them was seeking to meet him. Overall, everyone had a good time and thus we came to the end of another good meet. Well, that is all for the April 2004 Meet people ! See you all at the next meet hopefully with larger numbers.

Regards,
Surjo.

FINAL ACCOUNTS