RE: [Declude.JunkMail] Raid Controller

[Todd Holt]

I know that I’m coming into this late and I’m a little confused about much of this discussion.  I don’t see how it’s possible to get better performance by aggregating all of the disks and then partitioning them out.  This will still create contention when more than one partition is needed because they are on the same spindles.   I always thought the rule was to separate logical processes across separate spindles whenever possible and practical.  This would mean that multiple RAID sets would be faster when allocating a RAID set to each of the following separately: Windows system Swap file (or even multiple swap files on multiple RAID sets) Imail system Logging Webmail/calendaring   Now this would get expensive, but we are discussing performance here, right?   Going back to the SCSI vs. SATA price/performance issue: I have used SCSI for many moons and they are rock solid, but very expensive.  However, I usually end up replacing a drive every couple of years.  So my definition of rock solid is “no data loss” and “not more than one dead drive each year”.  That’s acceptable to me.  Besides, I can have RAID keep a hot spare to auto-rebuild the array and then I can hot replace the dead drive to become a new hot spare.  It seems that SATA controllers and drives can now give me this level of reliability.  And we have already agreed to the SATA performance being virtually equal to SCSI performance.   Now it just sounds like SCSI bigotry.  Kind of like the *nix people bashing the M$ people.     Todd Holt Xidix Technologies, Inc Las Vegas, NV USA 702.319.4349 www.xidix.com   -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of John Tolmachoff (Lists) Sent: Thursday, March 25, 2004 4:02 PM To: [EMAIL PROTECTED] Subject: RE: [Declude.JunkMail] Raid Controller   Matt, I agree with you. I am now confused, as I though it was better to separate physical Spans/Sets/groups by task, not logical partitions on one span/set/group by task.   John Tolmachoff Engineer/Consultant/Owner eServices For You   -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Matt Sent: Thursday, March 25, 2004 3:34 PM To: [EMAIL PROTECTED] Subject: Re: [Declude.JunkMail] Raid Controller   Ok, I'll bury this for the sake of everyone else on this list (though I though the full discussion wouldn't hurt since the topic comes up in brief often so I kept it here). Basically you are saying throw 4 disks into a span and mirror the span (8 drives total, one disk seen by the system, and partitioned into logical drives only for personal preference and not performance).  I was under the assumption that the logic was to separate spans for different tasks, in other words have multiple RAID 10 arrays instead of dedicating everything to just one.  I can see how redundancy isn't really an issue and performance is better than RAID 50 in this case with the only drawback being wasted space, but that is of no consequence here. Please feel free to correct me if I'm wrong, otherwise thanks for the discussion :) Matt Keith Anderson wrote: The harse ain dead yet.   Well, first thing is all RAID levels create one single volume that combines the total available drive space.  No matter what RAID level you use, all 10 drives become one big volume, just like the 24-drive RAID 10 that I've got here.  You can partition it through Windows only if you want to have more than one volume.   Raid 10 will always be the fastest redundant RAID.  Again, let's examine the process for a 4-disk system:   WRITE RAID 10:   Write to primary stripe (half of the drives, high-priority CPU cycles)   Copy to backup stripe (half of the drives, delayed, idle-time CPU cycles)      WRITE RAID 5:   Write to primary stripe (high-priority CPU cycles to all drives) READ RAID 10:   Read from primary stripe (half the drives)   READ RAID 5:   Read from the whole stripe (all of the drives)   There's also a calculative processor delay in RAID5 that RAID 10 doesn't have to worry about.  RAID 10 always knows where the data needs to go, RAID 5 has to figure it out, then create a parity block for every stripe.   You need to examine why you are asking this question-- what is your real storage need, performance vs. volume size vs. security?  Do you need the extra usable space with RAID 5 more than you need the 30-40% boost in performance that you get with RAID 10?  Do you need RAID 10's extra security of surviving a double-drive failure?   Keith       -----Original Message-----From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of MattSent: Thursday, March 25, 2004 3:06 PMTo: [EMAIL PROTECTED]Subject: Re: [Declude.JunkMail] Raid Controller Not to beat a dead horse, but... Am I mistaken about on RAID 5 array with 4 disks out performing one RAID 10 array with 4 disks?  RAID 10 will do double RAID 0 plus a slight hit for mirroring.  I though RAID 5 with 4 disks would out perform two striped drives despite the overhead. There is another issue though.  I can only get 10 drive in a packed 3U chassis, so I could only do two RAID 10 arrays, but with RAID 50, drive partitions wouldn't matter if I'm not mistaken, 1 would be the same as 5 partitions, or close enough at least.  With 8 disks in RAID 10, I could only separate the disk I/O for two logical drives. Matt    [AUTOMATED NOTE: Your mail server [63.147.33.8] is missing a reverse DNS entry. All Internet hosts are required to have a reverse DNS entry. The missing reverse DNS entry will cause your mail to be treated as spam on some servers, such as AOL.] ---[This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)] ---This E-mail came from the Declude.JunkMail mailing list.  Tounsubscribe, just send an E-mail to [EMAIL PROTECTED], andtype "unsubscribe Declude.JunkMail".  The archives can be foundat http://www.mail-archive.com.      -- =====================================================MailPure custom filters for Declude JunkMail Pro.http://www.mailpure.com/software/=====================================================