UNSOLVED

Sodhammer

updated

21 years ago

S

Sodhammer

8 Posts

0

10726

January 17th, 2006 17:00

Best practices for pe1800 CERC 6 SATA RAID

I have just purchased (has not arrived yet) a PE 1800 with the CERC - 6 channel SATA raid card.  It will be fully populated with 6*250 gig drives.  I am trying to determine the best way to utilize the drives for speed/redundancy/capacity.
 
I have read that it is better to have the OS located on a RAID 1 configuration (no need to calculate parity).  I have also read that the more drives you have in a RAID 5 configuration, the better it is (3 being not that great).  Since this machine has 6 drives, if I use 2 of them for the RAID 1 OS, that leaves me with 4 drives for the RAID 5.  But I also want a hot spare, which leaves my RAID 5 on only 3 drives.
 
The paragraph above assumes that I have dedicated an entire drive (well 2 drives) to the RAID 1.  I have not seen the RAID configuration utilities yet, but I assume that I could take a smaller junk of 2 of the drives to create the RAID 1 (which does not have to be that big), and use all 6 drives in the RAID 5 config.  What kind of a performance hit do I take for this since 2 of the 6 drives are now doing double duty?
 
Since this is a hardware RAID 5, is there a penalty for calculating parity?  If the card can calculate faster then the drives can write, I could see there being no penalty (while slower cards/solutions would have a penalty).
 
This particular server will be a file server only.  Data is ftp'ed to this machine from remote locations.  The data will then be processed and burned onto CD/DVD's (other machines).  The data is then erased from the server.  Some of these files will be very large (gigs).  Because of this, my primary concerns are that the Data volume be as large as possible and that the machine can stay up, even through a drive failure.  The data is also stored elsewhere; the copy on this server is simply a working local copy.
 
This is also a learning experience for me.  So any suggestions that cover other possible situations would be appreciated.
  • Sodhammer

    8 Posts

    377

    0

    Posted January 17th, 2006 17:00

    The point I was trying to make about the Hardware RAID 5 having no penalty for calculating parity (If this is true), is that it would then be better to have the OS drive on a RAID 5.  So many of these questions are based on the speed of this RAID solution, and I have not found any real documentation on the Dell websites that describe where the bottlenecks are in the current hardware.
  • Pelsia

    36 Posts

    377

    0

    Posted January 17th, 2006 18:00

    The major benefit of having the OS on its own disk is that if it fails you don't need to worry about your data as it's in its own container. When the OS disk dies, all you have to worry about it reloading the OS and then setting the shares and what not back up, as all your data will be safe in the other container. Performance wise this would be faster for the file transfer as when accessing the data disk there is no overhead in its buffer that might otherwise be partially taken up by the OS.

    If all you are doing is using the machine as a file server, you could leave the OS as a single disk and not bother with the raid 1 at all. With only the OS running there's not going to be much configuration on it, so reloading it should only take a hour or so. Even if you have a moderate amount of configuration, you could use a number of different ghosting/cloning software to image the disk. Since the disk is not part of an array you have a lot more options available that are a lot cheaper than trying to clone a raid array.

    Keep in mind that raid 1 really only excells when you have a hardware failure on disc 0, and then the mirror kicks on. If there is software failure or OS corruption, that gets mirrored onto the second disk and you effectively have 2 unusable disks, only option is to reload. Because of that I prefer to have the OS on its own drive, and then clone it so in the event it fails it can be back up in just the time it takes to put in a new drive and copy the image onto the new disk. Going this route you could have 4 disks in your raid 5 array with 1 hot swap.

    Though personally if you did go this way, I'd just pop for a 18 or 38 gb drive for the os, no reason to use a 250gb for something that's only gonna take up 10gb or so.
  • Sodhammer

    8 Posts

    377

    0

    Posted January 18th, 2006 14:00

    For this particular machine, my primary concern is 0 down time.  The system partition is also more important then the data partition, since the data is a temporary copy, I can easily have new copies sent.  While the system files (access rules and the like) are only stored on this machine (or to its backup device).

    So for this machine, I will want want both logical drives to have redundancy, since the loss of either will impact production until a tech can get to the machine.

    What I am hoping to find is someone that has taken this machine with a similar config (6 equal size large drives on the CERC 6 controller), and tested it to find the best config for performance and reliability.  I have read several other posts about these RAID's tanking when it comes to large file writes, so I am concerned about that.  There are also so many different posts out there, Its often difficult to evaluate how much they apply to me (different firmware, the definition of 'reliable').

    The machine should be arriving today, so I will post my results when done.

  • Pelsia

    36 Posts

    377

    0

    Posted January 21st, 2006 12:00

    With uptime being the crucial item, might consider doing RAID 10 then. Basically a mirror of your raid 5 arrary (raid 1 style) but done using another raid 5 array. From what I've read its the most reliable in terms of uptime (assuming clustering is out of the question). We just implemented a server using this configuration, was expensive as we had to buy larger disks as you effectively only get the storage capacity of 2 disks, but so far have had no problems with it (though its only been in production a week)
  • Sodhammer

    8 Posts

    377

    0

    Posted January 27th, 2006 16:00

    The machine is currently setup with 6 250 gig SATA drives.  I do have drive write cache on, but will probably turn off.  Since my data will only be coming in on a 100 connection, my tests have shown this config is more then equal to the task (massive overkill as a matter of fact).  The logical drives are:
     
    00 RAID 1, 30 gigs, using drives 0 and 1
    01 RAID 5, 90 gigs, using drives 2,3 and 4 (60 usable because of parity, using for local backup of the RAID 1 partition)
    02 RAID 5, 1000 gigs, using drives 0 - 4 (811 usable because of parity)
     
    Drive 5 is a global spare.  It appears that spare drives only apply towards RAID 5 containers.  It would be nice if it would also cover the RAID 1.  So I could lose 2 drives and the RAID 5 will still be operational (if not lost at the same time), while the RAID 1 can handle the loss of one drive.   This machine is in a remote location, and can not have remote management, so extended redundancy is my main concern.
     
    The machine is setup as an FTP server for receiving files that are then processed on other machines.  I have only been able to do limited testing so far.  Having another pc push files to the server over a 100 mb switch, it maxes out the network and maintains it (85-95%).  This was tested by pushing 10 copies of a 250 meg file concurrently. 
     
    I have not yet received my gig switch, so was only able to try sending over a direct cable connect.  This ran level at  11% of the gig network, but I believe this limit was due to the sending machines capabilities.  I tried copying files between the logical drives while recieving files over the gig port.  It appeared to have no effect on the network utilization level (so the network transfer speed was not being limited by how fast I could write to the RAID, since increasing the load on the RAID did not effect it). 
     
    I hope that others find this information useful.
     

    Message Edited by Sodhammer on 01-27-2006 12:53 PM