[34229] in North American Network Operators' Group
Re: Scalable Mail solution with NAS
daemon@ATHENA.MIT.EDU (Lyndon Nerenberg)
Wed Jan 31 15:09:11 2001
Message-Id: <200101312005.f0VK59A01790@gollum.esys.ca>
To: Sebastien Berube <sberube@zeroknowledge.com>
Cc: Valdis.Kletnieks@vt.edu, chrisb@kippona.com, nanog@merit.edu
In-Reply-To: Message from Sebastien Berube <sberube@zeroknowledge.com>
of "Wed, 31 Jan 2001 11:06:09 EST." <Pine.LNX.4.21.0101311057250.26915-100000@localhost.localdomain>
Date: Wed, 31 Jan 2001 13:05:08 -0700
From: Lyndon Nerenberg <lyndon@gollum.esys.ca>
Errors-To: owner-nanog-outgoing@merit.edu
[ This belongs in comp.mail.misc. ]
>>>>> "Sebastien" == Sebastien Berube <sberube@zeroknowledge.com> writes:
Sebastien> 2. The more mailboxes you have, the slower the entire
Sebastien> popping process will be. The reason is very simple,
Sebastien> each pop process will spawn and read your mailbox
Sebastien> directory.
You have a highly deficient pop server implementation. The fork-on-connect
model is dead. A well written threaded server can handle tens of thousands
of concurrent connections.
Sebastien> /export/mailboxes/j/o/h/n/johndoe.mbox
That just moves the problem around. Traversing those intermediate
directories isn't exactly a light-weight set of operations.
Sebastien> Although this example is not good in the case where you
Sebastien> accept usernames with 3 or less characters.
A proper hashing algorithm would solve that.
--lyndon
