James Drew's picture


Just looking at some of the possibilities for improving tkl core on a per appliance basis. Some of the key changes would be in the file systems used for appliances. Its an area that doesn't get much attention but can greatly increase speed and/stability of an appliance. Think about XFS, ReiserFS/Reiser4, BTRFS as well as the usual EXT3/4.

Also another application which may help you would be SystemTap or LTTng . They are  immature versions of DTrace but they will help in finding bottlenecks in your appliances and your customers deployments. 

.....Also, I know this has been said quite a bit but.... Switching to 64bit does have advantages.... And to be honest even newer Atom processors are 64bit capable... smiley

The appliance lineup could be changed a bit to make it a bit more SME/mini web startup friendly. (e.g. The inclusion of noSQL databases and web servers. CouchDB and NGNIX come to mind, along with Sinatra for Ruby on Rails). Many modern web apps don't need the power of a web server. Also many web dev's cant use SQL DBs or even need to use their full functions.

Anyway these are just my few cents; hope they're of some help

let me know what you think and i could help with a list of new apps (aimed at speed and flexibility) if you would like.


Liraz Siri's picture

Thanks for the thoughtful feedback Drew.

Changing filesystems: technically easy but requires thought as to what is the best default. It's not a simple matter of deciding which performs best because that depends on the workload. For example, I use Reiserfs on my workstation and like that it performs better than ext3 for directories with many small files like my maildir. We almost went with reiserfs as the default last time around. We'll defintely be taking another look at this for the 10.04 based releases. We also factor in the strength of the development community, toolset support, etc. I'm pretty sure we'll standardize on btrfs at some point, but I don't think it's ready for production just yet.

LTTng / SystemTap: seem to have come a long way since I last checked. Beginning to look genuinely useful. I too wish we had a full port of DTrace for Linux. I'd like to take a closer look at these myself soon but this is one area where the community can help out with a lot. If you can find performance bottlenecks and suggest generally useful tweaks that don't increase the complexity of an appliance too much that is something we will be happy to work into subsequent releases.

64-bit support: We need to add support for 64bit on our build infrastructure. This shouldn't be too difficult to fix technically but our workload has been a bottleneck so far. Hopefully, 64-bit support should be available not too long after the 10.04 based releases comes out.

Appliance line-up: we started out with the low-end of technical needs and would like to gradually work up to include appliances which will be useful to an increasingly sophisticated audience. If you'd like shape the direction of things to come you can help by providing ideas on common usage scenarios we can help satisfy, and better yet - tested TKLPatches (they save us the most amount of work when doing an appliance)!

Liraz Siri's picture

We're actually doing development right now for a new batch of appliances. We'll consider adding NoSQL style databases to the list of supported apps.

Add new comment