[opensuse] Re: on why maildir isn't standard...
Anton Aylward wrote:
On 08/20/2014 07:57 PM, Linda Walsh wrote:
That it takes almost 4x the space is bad enough, but that it 200-300 times slower is the real bear.
That makes no sense. If space were the issue then we would not use HTML mail, it would be banned from all major servers and archives. There would also be a ban on attachments: selfies, lolcats, ftp-by-email of documents and spreadsheets.
That's not space on an ISP... that's local space. Though the peformance hit is worse than the space hit.
So no I don't buy that. I'm not an ISP trying to sell you something.
As for the 'slower', no I don't but that either; its just a matter of juggling your metrics. You have made no provision for the spin-lops of mbox locking or the hassle that is needed to deliver mail while a user has the file open reading it; all of which involves copying the whole file back and forth.
---- Yes, I did. I don't need locks for grepping through my local mail. Mail is MOST often an append, then read-only resource. It's not generally used as a random access database. Dovecot keeps meta indexes for other info outside of the mbox with the data stored in 1 compact form.
You 234 Megabyte mailbox. Deleting or tagging a single message with maildir involves deleting a single file. With mbox it involves copying the whole file, stopping in the middle, skipping, the copying the rest.
NO IT DOESN'T... That's not the way Dovecot works. It keeps meta changes in it's index files. The data in the mbox is remarkably unchanged except when automated processes expire the email.
Your view of resources is a static one, driven by 1970s economics. The forces that apply today have more to do with the user experience.
?? Sorry, I prefer to wait .05 seconds to search through my 200MB mailbox rather than up to 60 seconds...
Personally I think your obsession is irrelevant and I'm more concerned that your absolutist attitude is spreading misinformation.
---- performance of 200-300x is relevant.
As far as I'm concerned all the MTAs and MDAs and mail readers I use can handle both mbox and maildir, so I pick what suits the context I work in. My INBOX is maildir; my archives, which I seldom access and never delete anything from, are mbox.
---- Nearly everything in my mail directory is an archive. I rarely if ever need random access updates to mailboxes. As for spreading misinformation -- I documented my claims in 2 cases, by using the sources you referred to. You are glossing over the fact that these things are not default and though your alligators won't stay off my crocodiles, our initial intent was to drain the swamp... trying to find out why something isn't working in it's initial setup -- that was the problem. In initial setup, things are in default state -- so that is what I try to use as a baseline in suggesting debugging and setup attempts. I certainly would not recommend how I have my system setup for most people. But I'm a computer person and I thihk you are two. Things that we do to computers would give most people nightmares... It's not about what we can do, but about going with the simplest setup from the defaults. Once that is working, THEN scale up complexity..
It might come down to usage, but putting each message in a separate file generally will waste space, time, resources... human time and cpu time...
Yes, that was the mentality of the 1970s when resources were scarce and cost more than human time.
--- No, it is today... from a cold cache, 0.2 seconds w/mbox, vs. 60+ seconds for maildir. You cannot claim 60 seconds is a reasonable time to wait for a spur of the moment search.
One of the reasons that Postfix and qmail are so popular is that sendmail has the ${DEITY}-${PERDITION} awful configuration setup. Postfix is not only more secure, and can run with lower privilege; qmail is almost trivial to set up and is also more secure; Smail3 setup is about the level that Marcus Ranum's proverbial cat can manage.
I will not argue that point. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On 08/20/2014 10:13 PM, Linda Walsh wrote:
Anton Aylward wrote:
On 08/20/2014 07:57 PM, Linda Walsh wrote:
That it takes almost 4x the space is bad enough, but that it 200-300 times slower is the real bear.
That makes no sense. If space were the issue then we would not use HTML mail, it would be banned from all major servers and archives. There would also be a ban on attachments: selfies, lolcats, ftp-by-email of documents and spreadsheets.
That's not space on an ISP... that's local space. Though the peformance hit is worse than the space hit.
So no I don't buy that. I'm not an ISP trying to sell you something.
I know you're not, Linda. I'm not accusing you of being so All I'm saying is geared the needs of a small system, a home system or a SMB. There's lots of documentation out there about all this, both sides of the argument, from oracle, HP, IBM and the like that describe multiple mail gateways, mail hubs mail servers, multitudinous local LANS each with local mail servers and mail readers and mail repositories on each workstation. The stuff that gives me nightmares, but probably makes equipment salesmen have wet dreams about their commissions. As with anything and everything ("Context is Everything") "It all depends". I have multitudes of old 800MHz single core machines with 1G of memory and a 20 or 30G drive courtesy of a desktop replacement scheme. I can run one of those as the 'mail server' under my desk. (Actually right now its acting as a support for my laserjet printer). Slow machine? yes. Single core so no parallelism such as Carlos talks of? Yes. But it doesn't matter. I'm not an ISP, as you point out. Years ago in DrDobb's a leter writer pointed out that the screaming fast compiler race was irrelevant unless you were a piss poor coder who was doing compile-fix-compile-fix cycles. What mattered was the speed of the compiled product. The letter writer said that if he fired up the compiler and went off to get coffee and chat with his colleagues and it was done when he got back, that was fast enough. Of course controversy ensued! But the principle holds. I'm not an ISP. My mail reader - Thunderbird - runs on a nice fast tower workstation (a bit too large and bit too hot for my liking, mind, but it will get upgraded in due course and perhaps will become the mail server: who knows) and talks to nearly half a dozen IMAP servers, one local, the other at various ISPs across Canada, USA and western Europe (nothing in the far east yet). I have a fetchmail for the couple more ISPs that don't support IMAP and have to be fetched using POP. All this on the single core 800MHz Closet of Anxieties Special. It used to run Madriva, then Mageia but now runs openSuse 13.1. There are a lot of things in life that we don't site there waiting for. "The watched kettle never boils", Well great, stick you cup of water in the microwave and go slice your bagel and op it in the toaster and get your creamed cheese out of the fridge and the milk and the teabag out of the jar in the cupboard and all of a sudden, just at the right time, the hot water is ready and the toaster pops up. Its not just parallelism it that you weren't there waiting for one thing. The speed of my mail *does* have to do with when the mail is ready on the remote ISP. That in turn depends on when *you* choose to write a reply. Al that dwarfs any time my CPU may spend. Programmers are advised to properly metricise (?sp? 'metricate' ?) their code before optimizing. There's no point in optimizing a non-critical part of the application. Just so here. The speed of processing we're talking about, issues of space, are not only context dependent (e.g. with ReiserFS doing space packing issue of waster block space are irrelevant, issues of i-node exhaustion as with ext3/4 are irrelevant; I'm sure you can make a similar case with XFS) but from the POV of the person reading the email are all irrelevant. We have many complaints about our mail readers and as Microsoft and the Joe Sixpack style purchasers and user, he kind of bickering we're engaging in is why we get labelled 'geeks' and often marginalized by the people who keep the lifeblood of the company - the money - flowing: sales and marketing. -- If I still had 8" floppy drives I'd be much more concerned about hardware failures, storage capacity and performance issues than BS viruses. -- Bill Campbell, CISM, CISSP, March 2007 -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On 2014-08-21 14:33, Anton Aylward wrote:
On 08/20/2014 10:13 PM, Linda Walsh wrote:
now its acting as a support for my laserjet printer). Slow machine? yes. Single core so no parallelism such as Carlos talks of? Yes.
You can parallelize spamassassin in a single core machine when the issue is waiting for the online tests to get an answer from somewhere in Internet, not cpu or disk load.
Years ago in DrDobb's a leter writer pointed out that the screaming fast compiler race was irrelevant unless you were a piss poor coder who was doing compile-fix-compile-fix cycles. What mattered was the speed of the compiled product. The letter writer said that if he fired up the compiler and went off to get coffee and chat with his colleagues and it was done when he got back, that was fast enough. Of course controversy ensued!
!!!! I would have been fired if I did that. So I just had wait looking attentively the screen...
The speed of my mail *does* have to do with when the mail is ready on the remote ISP. That in turn depends on when *you* choose to write a reply. Al that dwarfs any time my CPU may spend.
I need often to do text searches in my whole mail store, and they already take too long. On maildir, the search time increases by a factor between 10 and a 100 times. Of course, now I use lucene on dovecot, which speeds up the searches dramatically. But... the indexes are be rebuilt automatically too often for my liking, and they take long. This may improve after I fully switch to dovecot-lda, I still have to find out. I do see the advantages of Maildir, but they don't outweight the advantages I do need and use from mbox. It can, and is, different for other people or use cases. My preferred mail folder format would be "mix". But that will not happen... -- Cheers / Saludos, Carlos E. R. (from 13.1 x86_64 "Bottle" at Telcontar)
Anton Aylward wrote:
I have multitudes of old 800MHz single core machines with 1G of memory and a 20 or 30G drive courtesy of a desktop replacement scheme. I can run one of those as the 'mail server' under my desk. (Actually right now its acting as a support for my laserjet printer). Slow machine? yes. Single core so no parallelism such as Carlos talks of? Yes.
Single core != no parallelism but single core = no concurrency. (although with hyper-threading, you'd even get some concurrency. I.e. you could have spamassassin running in parallel, just not concurrently.
Programmers are advised to properly metricise (?sp? 'metricate' ?)
instrumentation. http://en.wikipedia.org/wiki/Instrumentation_%28computer_programming%29 -- Per Jessen, Zürich (15.1°C) http://www.dns24.ch/ - free dynamic DNS, made in Switzerland. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On 08/20/2014 10:13 PM, Linda Walsh wrote:
No, it is today... from a cold cache, 0.2 seconds w/mbox, vs. 60+ seconds for maildir. You cannot claim 60 seconds is a reasonable time to wait for a spur of the moment search.
All this line of argument, Linda, is pointless. What you're saying is "Its like that for me so I don't want to do that here". Fine. Don't. But "Context is Everything" and so much of what you assert is your context. I'm not disputing your context, I'm just saying that its not mine and not others. You've made it very clear in other threads that you run a very specific and very non-standard setup, disputing the use, for example, of initrd for booting. Fine, your context, your privilege. But don't tell us that not using initrd is the 'standard'. Yes, 'standards' of which there are so many and we can pick and choose from them, are useful. At times. And thy get revised and innovation and things like IBM, Microsoft and others create new 'standards'. Back when there was nothing else 'sendmail' was the standard, but once alternatives became available ... that was questionable. Conventions are not standards. Others are asserting that Postfix is the 'standard' with openSuse. Well either that or Exim have to be installed in anything except the most minimal. But a Choice is not a Standard. You can _choose_ whether to use mbox or maidir is used by anyy of the MTAs and MDAs we've discussed.
NO IT DOESN'T... That's not the way Dovecot works. It keeps meta changes in it's index files. The data in the mbox is remarkably unchanged except when automated processes expire the email.
Well good. but you've just done a switch form the general to the specific, and that's yet another of the trick in argument that Thules writes of. http://en.wikipedia.org/wiki/Straight_and_Crooked_Thinking We can bat these back and forth. Endlessly.
No, it is today... from a cold cache, 0.2 seconds w/mbox, vs. 60+ seconds for maildir. You cannot claim 60 seconds is a reasonable time to wait for a spur of the moment search.
Well excuse me! I bring up the menu click on the Thunderbird icon, have the KDE bouncing icon go at it, the the Thunderbird splash screen then T'Bird make the network connections to my local IMAP server (which is already running so the times you talk of are irrelevant) and the half dozen or so remote IMAP servers at various ISPs (see above). I then put my coffee cup down, add milk and fo ad sit at my desk once again. Sixty seconds? You jest? The reality is that a few weeks ago I booted the system with a new kernel, started up T'Bird and assigned it to a KDE desktop; firefox another, Konsle & tabs to yet another. All the times you speak of are a complete irrelevance as far s the UI are concerned. I have plucine and I got most of the ISPs to use it as well (Rogers seems to be the exception) as they need 'features' to stay competitive. (Rogers is pretty much a monopoly, they have divided up the country with Shaw.) I've also pointed out that I *chose* to use mbox for my archives, the stuff that never gets deleted or modified. That too has full indexing and plucine indexing so you are in fact agreeing with me over that. -- Just because something doesn't do what you planned it to do doesn't mean it's useless. Thomas A. Edison -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
participants (4)
-
Anton Aylward -
Carlos E. R. -
Linda Walsh -
Per Jessen