[email protected] wrote: > After you mentioned the maintenance of the array I started to wonder. > Does ‘nix handle this for me in the dmraid4-5 module?
I don't know, but it seems unlikely. (See below.) > Is there a way that I can see if I am really just using a Linux software > RAID setup like yours? The commands the other poster suggested and the results suggests that you are not. > Should I be setting something up in mdadm? I wouldn't expect so. mdadm is specifically design for managing the Linux software RAID driver, which you apparently are not using. > If things are still giving me grief, one option I have is to migrate my > array back to RAID 0 and give up on failure protection. I hate to do > that, and I’d want to be sure that my drives are really the problem before > doing that. I agree that I wouldn't do anything unless you determine that disk I/O is the problem. If you do, I'd recommend repurposing one of your existing drives as the boot drive, used in a non-RAID manner, and partitioned for Linux and Windows. Then separately build a RAID array for storing your recordings. (As the other poster mentioned, you might be able to pull this off by degrading your existing array, if you have enough SATA ports for the additional drives, and then do a disk-to-disk copy.) Avoid RAID as a boot drive makes things much simpler. Remember that RAID isn't a substitute for backups, it's a strategy for increasing uptime. Your boot partitions probably don't change that often (at least not in terms of valuable data), and you can use rsync or a traditional "bare metal" backup tool like Mondo Rescue to back them up. Even if you do use RAID for your boot drive, RAID 1 is a better choice. It's more likely to be supported by the kernel, and due to its simple nature is typically directly compatible among different software RAID implementations. So for example, you might be able to use both mdadm under Linux and NVRAID under Windows to maintain the same RAID 1 device. I'm currently using one RAID 5 array plus a newer RAID 1 array for recording storage. I plan to replace the older RAID 5 array with another RAID 1 array when I implement my next capacity upgrade. RAID 1 is easier to deal with when growing array capacity (the ability to grow a RAID 5 array has only recently been a working feature in mdadm). The theory is that with such a setup you can do a rolling capacity upgrade, where periodically - say every 6 months - you purchase whatever drive is in the "sweet spot" for cost per capacity and use it to replace your lowest capacity drive. Then you grow that RAID set to fit the smaller of the two drives and expand the file system on top. > MVPs that have issues in this regard. Why don’t the other PC based > frontends have similar issues? Do they buffer the data differently? As has been mentioned on the list, the MVP hardware has quite limited memory, and thus small buffers. Eric Sharkey wrote: > [email protected] wrote: >> What I don't understand is what does the linux dmraid4-5 module do? > > It looks like "dmraid" stands for device mapper raid. There's some info here: > > http://en.gentoo-wiki.com/wiki/RAID/NVRAID_with_dmraid > > It's a method for Linux to read data from a drive to allow it to > automatically configure a software raid. I know very little about dmraid, but my understanding (which is supported by the page referenced above) is that it acts as a driver between the kernel and a BIOS-based software RAID. More specifically, what this means is that the RAID device has some routines sitting in ROM somewhere, which provide a storage-device API to the operating system. To the operating system, the device most likely appears as a block device (or simple disk drive). >> Is any maintenance of the array handled by the dmraid4-5 module? > > It looks like it. > >> If maintenance of the array is required, i.e. synchronization for example, >> how often should this be done? > > The raid property must be maintained whenever a disk write occurs. > With software raid, this means that the operating system must actively > distribute/copy writes according to the desired raid configuration. I wouldn't classify the implementation of the RAID algorithm as "maintenance." I suspect the way it works with dmraid is that the kernel hands off a block of data to the API provided by the BIOS, and then the BIOS routines handle how that data gets encoded onto your disks to accomplish the RAID behaviors. This is why this type of BIOS-based software RAID can be used with multiple operating systems. Each OS relies on the low-level BIOS routines to implement the RAID structure. This is also why you can get locked-in with this kind of hardware. > The point is, if Linux can do a disk write and maintain the raid > property, then it can do all of the other things as well. I wouldn't assume that. The RAID device in this case most likely appears as a simple block device from the kernel's perspective, so it wouldn't know how to initiate a rebuild. Actually, even with mdadm things like manual rebuilds are accomplished through the mdadm administration tool being used to send commands to the RAID driver. Scott mentioned he used a Windows application to perform maintenance operations on the RAID array. Unless someone has developed an equivalent tool for Linux, he's likely stuck with that Windows app. > Raid verification is optional, it's just checking that everything is ok. Right. I don't see any mention of rebuilds on that page above. I'd be far less concerned with the ability to do preventative rebuilds, and far more concerned with your ability to be notified when the RAID encounters problems. If your RAID is going to serve its purpose, it is critical that you know when it is in a degraded state (and why). This could be accomplished with a physical light on the hardware, but with NVRAID it is likely accomplished with some system tray utility in Windows. I've heard that this can be a weak point in this kind of fake RAID under Linux. I don't see any mention of notifications on that page. I'd look into it further, and recommend testing it by disabling a drive and seeing whether that triggers any notification. Similarly, I'd want to understand better how or if dmraid can initiate a rebuild, after a hardware problem is corrected. Is it necessary to always run the Windows application and perform a manual rebuild? That would make notification all the more important. [email protected] wrote: > What is interesting is that after my full synchronization in the Nvidia > Mediashield software in Windows, all my grief seems to be over. I have > just finished watching about 3-4 hours of recordings while my main > frontend / backend machine was left on live tv for kicks. NO STUTTERING! Good news, though I don't have an explanation as to why an integrity check would improve performance. We can speculate that maybe your RAID was in a degraded state and the I/O performance suffered as a result, though in a traditional RAID the loss of performance in a degraded state occurs because the the RAID subsystem (hardware or software) is occupied with the rebuilding process. If your RAID had the ability to automatically rebuild, then the performance loss should be temporary, and go away after a day or so. Other things you should consider: Installing smartmontools (smartd) on both of your operating systems to monitor the health of your drives and perform periodic surface scans ("long tests"). smartd might not work with your RAID controller. Determine if your file system is fragmented, and set up a cron job to periodically defragment it. (I have one set up for XFS used on my recording drives. See the MythTV wiki for some info on this.) If your recordings drive is perpetually near full, it can be impossible to keep defragmented, so a capacity upgrade may have the side effect of improving I/O performance even if the drives stay the same speed. > Tom thank you for the tip in the right direction. You're welcome. -Tom ------------------------------------------------------------------------------ _______________________________________________ Mvpmc-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mvpmc-users mvpmc wiki: http://mvpmc.wikispaces.com/
