What sort of performance gains can mailman expect from going from a single processor to a dual processor machine?
I'm considering going from a P4 2.0GHz machine to a dual 2.0 GHz Xeon system (also going from ide to scsi hdd).
My question is will mailman make use of the second processor with different processes? Say one processor could be used for serious bounce processing and the other for subscriber / unsubscribe requests and other functions? Or will all the locking on list files etc. negate the extra processor?
Second question/comments: My total subscriber base in as of today 219,829 (announce only) and it seems to be killing my current machine. I'd like to be able to scale up to over 3 or 400,000 if needed.
Am I perhaps being unrealistic about mailman's abilities to handle this? An idea occurred to me that perhaps the only way to stick with mailman and process large lists could be to split the volume up over multiple boxes.
Everything else seems to be able to handle around 200,000 subscribers except bounce processing (ie MTA can pump out the messages no problems).
It has been suggested Fil at Rezo that pre-processing the bounce messages without opening the lists and then only post-processing the bounces against the real lists could improve performance. Is this in keeping with plans for 3.0 or any other planned future release? If a change like this is likely to be accepted into CVS would any other large list users be interested in assisting with resources for someone to make the changes and submit them back into the mailman project?
From an economics per perspective the cost of ownership of additional box's to run mailman amounts to $1000's of dollars annually. If through a number of interested parties all making a small investment we can cut down of the need for extra machines AND help make mailman a better system then we have a big win-win.
Anyone interested in discussing this further on or off list with me would be appreciated (including anybody who might be interested in the contract to do the work or any ideas of the feasibility / time the change could make)
-- Robert Anderson
Managing Director Ecommerce Magic Ltd http://www.ecommerce-magic.com +64 21 808 525
participants (1)
-
Robert Anderson