Dude. Check out http://openwebmail.org instead of Squirrelmail. It has a much more attractive interface. <<JAV>> ---------- Original Message ----------- From: Henry Standing To: suse-linux-e@suse.com Sent: Wed, 01 Sep 2004 09:47:34 +0100 Subject: Re: [SLE] Mail Server purchasing decision

Any P4 will do, even a fast P3. Don't over buy on the processor, save the

...

money for mirrored disks (raid). 512meg of memory should be plenty for the work load you would have.

Okay - how does this sound:

SuSE 9.1 Pro Postfix Squirrel light(ish) processor - P4 2GHz? Hardware-RAID0'ED SCSI disks 1GB RAM Gigabit ethernet (purely because I can, and my network is 90% Gigabit)

...

You do intend to run Amavisd or some virus scanner on it don't you?

I was thinking of running Sophos - anybody got experience with this?

http://www.sophos.co.uk/sophos/docs/eng/factshts/Sophos_PM_Unix_fsuk.pdf

On Wed, 2004-09-01 at 05:15, John Andersen wrote:

...

On Tuesday 31 August 2004 05:03 am, Henry Standing wrote:

...

Squirell Mail, SpamAssassin and Sendmail to serve c. 70 POP boxes

...

1. Has anybody got any advice on which model I should go for (budget of GBP 2000 ex VAT)?

2. Has anybody got any advice on which version of SUSE would be most appropriate?

That's not much of a load, if all it has to handle in inbound mail and pop.

Any P4 will do, even a fast P3. Don't over buy on the processor, save the money for mirrored disks (raid). 512meg of memory should be plenty for the work load you would have.

You do intend to run Amavisd or some virus scanner on it don't you?

-- Check the headers for your unsubscription address For additional commands send e-mail to suse-linux-e-help@suse.com Also check the archives at http://lists.suse.com Please read the FAQs: suse-linux-e-faq@suse.com ------- End of Original Message -------

Yes, yes RAID 1.... Mirrored, not striped. Sorry, I wasn't paying attention. All of my other servers are RAID5 as data is so important to the company. I learnt the hard way when a disk failed on a RAID5'ed Win2K server the true value of redundancy. Perhaps I should have explained the reason for h/w RAID over s/w RAID. I would like (although it is not an absolute requirement) to be able to Ghost off an image periodically. As the OS is providing RAID in s/w RAID, and to get Ghost32 up and running I have to boot off a custom PE disk, I would need RAID at the h/w level. That said, comments on flaky drivers are starting to worry me. Should I can the Ghost idea and rely on tarring/dumping to a Samba share on another server? That way I could use s/w RAID and avoid sketchy drivers. Finally - SCSI or IDE? I can't imagine that fast disk access is going to be a priority - comments?? Thanks for all your advice so far _ (\o/) H /_\ On Thu, 2004-09-02 at 08:18, John Andersen wrote:

On Wednesday 01 September 2004 12:47 am, Henry Standing wrote:

...

Hardware-RAID0'ED SCSI disks 1GB RAM

Raid0 - - as in ZERO? NO, that's just wrong on so many levels.....

Raid 0 just _increases_ your chance of data loss by the number of disks you have. (Loss of any 1 disk can cost you the whole array).

Go raid 1 or raid 5. Either will run right thru a single disk failure. Throw in a hot spare, and the system will probably recover from the failure before you even know it happened.

And at least evaluate software raid. Its not dependent on late arriving and often flaky controller drivers. (There are still reports of raid controller problems in SuSE 9.1).

I've been using software raid for about 5 years now and I really like it. Due to a power surge I had to replace one of the disks, and the software worked perfectly. Yes, it takes a tad more cpu power, but never exceeds 1% cpu on some pretty wimpy hardware. Software raid also allows you to mix scsi and IDE devices.

Henry wrote regarding 'Re: [SLE] Mail Server purchasing decision' on Thu, Sep 02 at 08:34: [...]

That said, comments on flaky drivers are starting to worry me. Should I can the Ghost idea and rely on tarring/dumping to a Samba share on another server? That way I could use s/w RAID and avoid sketchy drivers.

I don't know who's complaining about flaky hardware raid drivers, but I've been using various big name raid cards under Linux for 10 years and have *never* had a problem with them. A hardware RAID card is pretty much the same from a driver perspective as a SCSI card, essentially. Just go with someone who's well supported. I like 3ware's products, but the other big names are fine too. As long as you're not using some POS promise card - which is as likely to have problems in hardware as in the drivers - you should be fine. And hardware RAID *is* better than software raid in an enterprise situation (you're not likely to be mixing IDE and SCSI drives there). Then again, I'd probably tell you to use Reiser for a server mailstore (if you're doing it the smart way w/ Maildir), and I don't recall ghost working nicely with Reiser. So ghost may not be an option anyway. :) --Danny, who uses rsync + hardlinks to make daily snapshots...

That assumes they were slow to start with. And if you upgrade controllers, then you get the benefits that come with the new controller. You should always go HW controlled RAID if possible. The only reasons not too are, 1) you don't have a controller, or 2) there's no driver available (or reliable driver) for your controller. <<JAV>> ---------- Original Message ----------- From: John Andersen To: suse-linux-e@suse.com Sent: Thu, 2 Sep 2004 11:20:35 -0800 Subject: Re: [SLE] Mail Server purchasing decision

On Thursday 02 September 2004 05:30 am, Henry Standing wrote:

...

Perhaps I should have explained the reason for h/w RAID over s/w RAID. I would like (although it is not an absolute requirement) to be able to Ghost off an image periodically. As the OS is providing RAID in s/w RAID, and to get Ghost32 up and running I have to boot off a custom PE disk, I would need RAID at the h/w level.

Why could you not ghost off each drive independently? Especially with raid1?

Heck, I once took a software raid1 member, (just one of them) and mounted it read only on a totally different linux box and backed it up that way. I could not have done that with a hardware raid disk, I would have needed to get another identical controller.

When you upgrade the machine or the disks, software raid gets faster. Hardware raid is is just as slow after an upgrade as before, because the controllers are the limiting factor.

-- _____________________________________ John Andersen ------- End of Original Message -------

I would agree with many of your points save one. HW RAID (not including Promise IDE stuff because that's not a true HW controller in the same sense that a Compaq SMART Array Controller is) is always going to be faster. There is no way, for instance, SW RAID could compete with say a 6400 w/256MB Cache Module. There again, I'm taking the best possible controller. A quad 3Ghz Zeon system running SW RAID might outperform say a 3200 with little cache, but even there I'd want to see the test. I would agree, if you didn't have a controller then that would change the decision making process dramatically. If you have it, use it; esp. if its a Compaq controller because they are very good controllers. I don't the Promise IDE controllers would be serious contenders. <<JAV>> ---------- Original Message ----------- From: John Andersen To: suse-linux-e@suse.com Sent: Thu, 2 Sep 2004 12:17:24 -0800 Subject: Re: [SLE] Mail Server purchasing decision

On Thursday 02 September 2004 11:38 am, Joe Polk wrote:

...

That assumes they were slow to start with. And if you upgrade controllers, then you get the benefits that come with the new controller. You should always go HW controlled RAID if possible. The only reasons not too are, 1) you don't have a controller, or 2) there's no driver available (or reliable driver) for your controller.

The words SHOULD ALWAYS qualify as an opinion, and not fact.

I've used hardware controllers in the past, both SCSI and IDE and I prefer software raid.

Facts: Its just as easy to set up, more flexible (can you add ide members into your scsi raid?), just as fast or faster than hardware raid given an adequate platform, Its cheaper, easily replicated, every bit as robust, and the drivers are always there.

IMHO: The only reasons for hardware raid is You have a processor challenged platform (read whimpy or over worked) , OR you have a budget you have to spend, OR it was built into the machine you have anyway. Might as well use it.

Regarding the "given an adequate platform" condition (above) ... If your processor is too busy to handle the minimal additional IO related tasks associated with SWRaid, its probably too limited for your business needs.

-- _____________________________________ John Andersen ------- End of Original Message -------

On Fri, 2004-09-03 at 09:43, James Oakley wrote:

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On Friday 03 September 2004 9:57 am, Joe Polk wrote:

...

I would agree with many of your points save one. HW RAID (not including Promise IDE stuff because that's not a true HW controller in the same sense that a Compaq SMART Array Controller is) is always going to be faster. There is no way, for instance, SW RAID could compete with say a 6400 w/256MB Cache Module.

You really should look at http://spamaps.org/raidtests.php

This only applies to the promise IDE controller. You would have to make simular tests against a SCSI controller for comparison. SCSI disks are generally faster (and more reliable) all around than IDE, although IDE has made some strides lately. -- Ken Schneider unix user since 1989 linux user since 1994 SuSE user since 1998 (5.2) * PLEASE only reply to the list *

On Fri, 2004-09-03 at 14:05, James Oakley wrote:

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On Friday 03 September 2004 11:04 am, Ken Schneider wrote:

...

On Fri, 2004-09-03 at 09:43, James Oakley wrote:

...

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On Friday 03 September 2004 9:57 am, Joe Polk wrote:

...

I would agree with many of your points save one. HW RAID (not including Promise IDE stuff because that's not a true HW controller in the same sense that a Compaq SMART Array Controller is) is always going to be faster. There is no way, for instance, SW RAID could compete with say a 6400 w/256MB Cache Module.

You really should look at http://spamaps.org/raidtests.php

This only applies to the promise IDE controller. You would have to make simular tests against a SCSI controller for comparison. SCSI disks are generally faster (and more reliable) all around than IDE, although IDE has made some strides lately.

Maybe you should have actually followed the link instead of blindly assuming what it contained.

Allow me to quote the page:

Hardware: Tyan S2466-N4M AMD760MPX Chipset Motherboard 2x AMD Athlon MP1900+ CPU 512MB Crucial ECC Registered PC2100 DDR RAM running at 266Mhz. Single Maxtor 80GB HDD using onboard PATA for OS. 3ware Escalade 8506-8 SATA RAID Card 4x WDC250GB 7200rpm SATA HDDs

I did follow the link but the fingers overroad what the brain was thinking. Yes it is a 3ware controller, not a promise controller but the rest remains the same. SCSI is generally better and faster then SATA. Why do you think SCSI drives have a 5 year warranty and most IDE have one? because they are of better quality. And yes anything get break on day one. And you can always buy SCSI drives that are faster then even the fastest SATA drives. -- Ken Schneider unix user since 1989 linux user since 1994 SuSE user since 1998 (5.2) * PLEASE only reply to the list *

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Friday 03 September 2004 3:53 pm, Randall R Schulz wrote:

While I generally accept that longer warrantees are generally associated with better quality ... generally. But it's far from a sure thing.

Absolutely. A while back my brother bought a WD drive that had a 1 year warranty. In the box was an order form for a 3 year warranty for $10.

For what it's worth, I've always insisted on the fastest SCSI interface and drives I can get at the time I put a system together. Right now I'm running a dual channel Ultra 160 controller (Adaptec 39160) and 10,000 RPM drives. In the future, I may be willing to go the SATA route, though I would probably not be satisfied with only two SATA channels, as is still pretty common for on-board SATA support.

Check out these Epox boards which have 6 onboard SATA interfaces: http://www.epox.com/USA/product.asp?id=EP-8KDA3plus http://www.epox.com/USA/product.asp?id=EP-8HDA3plus http://www.epox.com/USA/product.asp?id=EP-8HDA5plus - -- James Oakley Engineering - SolutionInc Ltd. joakley@solutioninc.com http://www.solutioninc.com ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ This e-mail is CONFIDENTIAL and contains information intended only for the person(s) named. Any other distribution, copying or disclosure is strictly prohibited. If you have received this e-mail in error, please notify me immediately at 902 420 0077 or reply by e-mail to the sender and destroy the original communication. Thank You. ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) iD8DBQFBOS3Q4U2uQswGyDcRAjOZAJ0blpIsvAoYVEiZK8L/Q2sGlMe5MACdE7z3 sb4K6oPk5KZZV6BDmPt1iHg= =16UG -----END PGP SIGNATURE-----

On Friday 03 September 2004 02:57 pm, James Oakley wrote:

On Friday 03 September 2004 3:26 pm, Ken Schneider wrote:

...

I did follow the link but the fingers overroad what the brain was thinking. Yes it is a 3ware controller, not a promise controller but the rest remains the same. SCSI is generally better and faster then SATA. Why do you think SCSI drives have a 5 year warranty and most IDE have one? because they are of better quality. And yes anything get break on day one. And you can always buy SCSI drives that are faster then even the fastest SATA drives.

At what point did I claim otherwise?

I was addressing the commonly stated myth that hardware RAID is faster than software RAID.

-- James Oakley Engineering - SolutionInc Ltd. joakley@solutioninc.com http://www.solutioninc.com

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

I have but one question to all you guys participating in this thread. What does hard drives, SATA, SCSI, motherboards, etc. have to do with "Mail Server purchasing decision"? I'm not trying to start a "discussion", but everyone needs to quit stealing threads and start new ones. There are many reasons why, but mainly because it messes things up in the mail list and should anyone want to read some of the interesting information mentioned, they probably won't for skipping over the title of no interest to them. Also, should anyone want to do a search, for this info, it's not going to be found. Please stop stealing threads, start new ones. Thanks for your assistance. Lee -- --- KMail v1.7 --- SuSE Linux Pro v9.1 --- Registered Linux User #225206 Those Who Dance Are Considered Insane, by Those That Cannot Hear the Music!

On Friday 03 September 2004 10:57 am, James Oakley wrote:

...

And you can always buy SCSI drives that are faster then even the fastest SATA drives.

At what point did I claim otherwise?

I was addressing the commonly stated myth that hardware RAID is faster than software RAID.

And the article was spot on because it used the same hardware with and without the controller running in Raid mode. Often tests compare two different controllers which confuses the issue. This particular controller is often pushed forth as a shining example of hardware controllers, but in my tests the ones on Dell and Compaq (hp) are generally better performers. So now the only question remaining about this particular test has to do with whether the 3ware Escalade Sata model is dramatically worse than the 3ware SCSI models - but ONLY worse when running a raid. Unless I was being given the absolute top of the line controller, I think I'd still put _MY_ money into more ram. -- _____________________________________ John Andersen

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Friday 03 September 2004 12:26 pm, Danny Sauer wrote:

James wrote regarding 'Re: [SLE] Mail Server purchasing decision' on Fri, Sep 03 at 08:44:

...

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

On Friday 03 September 2004 9:57 am, Joe Polk wrote:

...

I would agree with many of your points save one. HW RAID (not including Promise IDE stuff because that's not a true HW controller in the same sense that a Compaq SMART Array Controller is) is always going to be faster. There is no way, for instance, SW RAID could compete with say a 6400 w/256MB Cache Module.

You really should look at http://spamaps.org/raidtests.php

They used the 3ware controller in both tests, using the 3ware controller's on-board cache in both cases. What kind of person is gonna buy a RAID controller and then run each disk on the controller individually, using software RAID instead of the hardware raid that they paid for?

Someone who wanted better performance?

If those disks had been on the motherboard IDE interface, like >99% of people who use software RAID, the software RAID would've lost, but it does show where a little ore performance could be gained *if* CPU load is irrelevent. What happens when the CPU starts getting loaded down, too, like on a busy database, mail, or web server? Hint, the hardware RAID card keeps up, while software RAID degrades.

Do you have numbers to back that statement up? Do you know how fast the PCI bus is?

With software raid, you also don't get the fancy management interface that most of the big names provide for their cards. Everyone loves a web interface! :)

http://evms.sourceforge.net/ A real GUI is far better than a web interface.

Anyway, I much prefer having the processor left as free as possible for dealing with processing, not wasting time handling I/O stuff that dedicated hardware can do just as well or better in real-world use. Feel free to disagree.

If it made a real difference (feel free to provide numbers) you'd just use a NAS anyway. - -- James Oakley Engineering - SolutionInc Ltd. joakley@solutioninc.com http://www.solutioninc.com ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ This e-mail is CONFIDENTIAL and contains information intended only for the person(s) named. Any other distribution, copying or disclosure is strictly prohibited. If you have received this e-mail in error, please notify me immediately at 902 420 0077 or reply by e-mail to the sender and destroy the original communication. Thank You. ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) iD8DBQFBOLQG4U2uQswGyDcRAn9EAJ927qiFNg0346pWpwVuxfyD0GECQwCguj8T sJHfVeGC55n+XtAybhuLDqk= =GqQc -----END PGP SIGNATURE-----

John wrote regarding 'Re: [SLE] Mail Server purchasing decision' on Sat, Sep 04 at 00:59:

On Friday 03 September 2004 07:26 am, Danny Sauer wrote:

...

They used the 3ware controller in both tests, using the 3ware controller's on-board cache in both cases.  What kind of person is gonna buy a RAID controller and then run each disk on the controller individually, using software RAID instead of the hardware raid that they paid for?  I

ME for one.

Look, the performance is BETTER!

I'll grant that - under a synthetic benchmark the performance gain is marginal. Whoopie. Software RAID finally catches up to be "as good as" hardware RAID, and maybe even a little better under circumstances unlikely to be encountered in regular use. Where's the benchmark under load, like anyone who can justify RAID will be experiencing? Nowhere? Hmm... Also note that they only used XFS on the hardware raid setup - the test was mostly to compare *filesystem* performance. Compare HWXFS to SWXFS, and note that the hardware solution is faster in some of the cases.

And CPU utilization for software raid is a MYTH, left over from the 486 days.

To quote the other reply: "got numbers?" There are some numbers on that test page linked to - they show the hardware RAID setup using about 1/3 of the CPU of the software RAID pretty consistently. Ok, so the CPU's only used a little now. It's still greater than with a hardware RAID controller. Windows XP uses more resources than '95 did, and programs are getting bigger all the time. What the heck? We've got memory and processor to spare now! The CPU is *still* used more, with no regard for whether or not the cycles consumed are important to the application at hand. I use dedicated hardware for sound, network interfacing, video, and disk controlling. All of that could be done in the processor, and the cycles consumed wouldn't really hurt anything. Are you running a winmodem now, since the processor can do some DSP on the phone signal "as well as" a hardware modem? Running sound and network over USB since it's "almost as good as" hardware on the PCI bus? Using integrated video without any 3D acceleration for those games, since the processor "can" do it? I dunno. Maybe it's the part of me that builds engines and prefers to eliminate things like an engine-driven fan, because it limits the amount of power I can use at full throttle. I don't drive at full throttle all the time, but I want as much engine as possible to make me go forward when I put the hammer down. Software RAID is that engine-driven fan that cools better under some situations, but is worse in other situations and is taking a possibly imperceptable amount of full-load speed away from me. Feel free to have the fastest benchmark computer around. I've got more cycles available when I *use* my computer.

Read the Article.

I did. Twice now. --Danny