Re: [mythtv-users] INFO: RAID comparison for MythTV
MagicITX wrote: The problem is RAID 5 is hosed if two drives fail. The more drives you have the more statistically likely you are to suffer a two drive failure. If you are paranoid, then have you seen RAID6. It's now considered production ready. It will withstand 2 drive failures whereas RAID5 with a hot spare has a window of vulnerability during the resync period for the hot spare. It is slightly less performant that RAID5 but as you have said, for Myth that's usually not an issue. David ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
Bryce wrote: On Thu, April 21, 2005 4:15 pm, MagicITX said: As another hint - Start your RAID small and grow it. Buying your disks all at once means that they'll all tend to fail at once - Which is not a good thing. Unfortunately, there's no easy, reliable way I've found to extend a RAID without backing up (somewhere really huge :) and restoring. That's a myth about RAIDs and hard drives in general. I've built and maintained many many RAIDs over the years (with several of them running for years) and never have experienced a noticeable increase in drive failures or drives failing closer together, etc. Nor have I ever had a vendor even suggest such a procedure. Hard drives basically fail regularly and randomly over the entire life of the RAID. Even when EMC installs their big $500K frames, they don't add drives bit by bit or even from different manufacturing runs. This may be true for SCSI drives, or the "higher end" of IDE. However, I bought 9 drives and had 2 fail in the first month. On an array where each drive is used identically, you can't get away from the "bathtub" failure curve. Yes, the solution should be don't buy cheap drives. But I like my I in my RAID. Cheers, Allan. ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
Robert Johnston wrote: On 4/21/05, MagicITX <[EMAIL PROTECTED]> wrote: While the number of drives in a RAID 5 array is theoretically unlimited, some recommend no more than 14 drives. The problem is RAID 5 is hosed if two drives fail. The more drives you have the more statistically likely you are to suffer a two drive failure. Incorrect. As RAID5 uses drives in sets of 3 (2 data + 1 CRC), then you have to lose 2/3rds of the drives in the array for it to fail. With just 3 drives, that means if 2 of the 3 drives fail, the array is hosed. With 6 drives, that means 4 of those 6 have to fail, and so on. Generally with RAID5 arrays that aren't a multiple of 3 (14-drive, say), the array is configured with 12 drives in the array, and 2 drives as "Hot Spares" that are swapped in automatically if any one drive fails. Perhaps you are confusing RAID 3 with RAID 5. RAID 5 moves the parity bit around (rotating parity array), no one single drive contains the parity bits for all of the data. RAID 3 utilizes one drive to store parity, but I'm not sure if it is in sets of 3 drives. -Ack ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
Robert Johnston wrote: Incorrect. As RAID5 uses drives in sets of 3 (2 data + 1 CRC), then you have to lose 2/3rds of the drives in the array for it to fail. With just 3 drives, that means if 2 of the 3 drives fail, the array is hosed. With 6 drives, that means 4 of those 6 have to fail, and so on. Generally with RAID5 arrays that aren't a multiple of 3 (14-drive, say), the array is configured with 12 drives in the array, and 2 drives as "Hot Spares" that are swapped in automatically if any one drive fails. This is incorrect (see the wikipedia entry linked above). RAID 5 writes one set of parity across all drives. If 2 drives are lost, no matter the number in the set, all data is lost. Additional hot spares can mitigate the risk at the expense of spare disks. If you think about it, how could 2 disks (of your six disk set) possibly contain all the data from the failed 4? That would mean 200GB (for 100GB disks) storing all the data for a 600GB RAID set or at least all the parity data needed to rebuild that data. Kevin ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
RE: [mythtv-users] INFO: RAID comparison for MythTV
-Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Robert Johnston Sent: Thursday, April 21, 2005 7:14 PM To: MagicITX; Discussion about mythtv Subject: Re: [mythtv-users] INFO: RAID comparison for MythTV > Incorrect. As RAID5 uses drives in sets of 3 (2 data + 1 CRC), then > you have to lose 2/3rds of the drives in the array for it to fail. > With just 3 drives, that means if 2 of the 3 drives fail, the array is > hosed. With 6 drives, that means 4 of those 6 have to fail, and so on. > Generally with RAID5 arrays that aren't a multiple of 3 (14-drive, > say), the array is configured with 12 drives in the array, and 2 > drives as "Hot Spares" that are swapped in automatically if any one > drive fails. RAID5 means striped data with striped parity. It is not based on a 2 data, 1 CRC system. This is just the minimum amount of drive you need to create one. You can have any amount of drives greater than 2 in a RAID5. If you have a 14 disk RAID5, data is striped across all 14 disks. If you lose any drive, then the system will recover the missing information from the drive through parity. If you lose a 2nd drive, there is no parity information to recover the data from. The idea of 2 data + 1 CRC may work for a 3 disk RAID5, but does not work for any other. If a RAID5 loses a disk when it is reduced (1 drive already failed), then you've lost all of your data. In general you'd have to have 2 complete physical failures to lose all of your data. Because it's always possible to re-assemble the raid and recover data from the parts of the disk that are not destroyed (bad sectors, etc). ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
On 4/21/05, MagicITX <[EMAIL PROTECTED]> wrote: > While the number of drives in a RAID 5 array is theoretically > unlimited, some recommend no more than 14 drives. The problem is RAID > 5 is hosed if two drives fail. The more drives you have the more > statistically likely you are to suffer a two drive failure. Incorrect. As RAID5 uses drives in sets of 3 (2 data + 1 CRC), then you have to lose 2/3rds of the drives in the array for it to fail. With just 3 drives, that means if 2 of the 3 drives fail, the array is hosed. With 6 drives, that means 4 of those 6 have to fail, and so on. Generally with RAID5 arrays that aren't a multiple of 3 (14-drive, say), the array is configured with 12 drives in the array, and 2 drives as "Hot Spares" that are swapped in automatically if any one drive fails. Generally, the reason no more than 14 drives per RAID5 is recommended is that the maximum number of devices you can have on a Wide SCSI (UW, U2W, U160, U320) bus is 15, which taking one off for the controller card, leaves a maximum possible of 14 drives per channel. RAIDing across channel on SCSI is less than optimal. > I have seen the argument that you shouldn't get all of your drives at > the same time since they are more likely to fail at about the same > time. As mentioned above that is bad with RAID 5. This one is true, however. A company I was working for got bitten by this when a caseload of IBM U320 SCSI HDD's all failed within 90 hours of being installed into the new Domain Controllers, Mail Servers and live Web Servers, as the case had been dropped at the distributors. -- Robert "Anaerin" Johnston ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
On Thu, April 21, 2005 4:15 pm, MagicITX said: >> As another hint - Start your RAID small and grow it. Buying your disks >> all at once means that they'll all tend to fail at once - Which is not a >> good thing. Unfortunately, there's no easy, reliable way I've found to >> extend a RAID without backing up (somewhere really huge :) and >> restoring. That's a myth about RAIDs and hard drives in general. I've built and maintained many many RAIDs over the years (with several of them running for years) and never have experienced a noticeable increase in drive failures or drives failing closer together, etc. Nor have I ever had a vendor even suggest such a procedure. Hard drives basically fail regularly and randomly over the entire life of the RAID. Even when EMC installs their big $500K frames, they don't add drives bit by bit or even from different manufacturing runs. -- Bryce T. Pier ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
MagicITX wrote: While the number of drives in a RAID 5 array is theoretically unlimited, some recommend no more than 14 drives. The problem is RAID 5 is hosed if two drives fail. The more drives you have the more statistically likely you are to suffer a two drive failure. While still not perfect, this is why you would want a hot-standby drive in a RAID 5 setup. You will only be running in degraded mode for as long as it takes the system to rebuild the array on the hot-standby, not for as long as it takes for one to a) buy and install a new drive, or b) install a spare that was sitting on a shelf. Granted this still doesn't help if you lose two drives at the same time or lose a second drive while the system is rebuilding the array. Also RAID 10 can also lose all data in the entire array if two drives in the *same mirror* fail. Granted I am just picking nits here :-) :-) RAID 5 is more prone to the two drive failure problem as it can be *any* two drives in the array. Good article though. For me it is just TV. My video storage is on RAID 0. If a drive fails and I lose all my shows, I'll be sad, but life goes on ;-) In case anyone is interested: http://en.wikipedia.org/wiki/Redundant_array_of_independent_disks -- David HDTV frontend I'm working on (picture of back, mythmon source) http://mythhd.info ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
On 4/21/05, Allan Stirling <[EMAIL PROTECTED]> wrote: > MagicITX wrote: > > A week or so ago there was a thread discussing RAID for MythTV. We > > did a test comparing software RAID 10 and RAID 5 on a mythbackend > > server. You can find the article on our site if you're interested. > > Basically we found that the parity calculation for RAID 5 had a > > minimal impact on the performance of the system. We did see much > > better throughput under RAID 10 as one would expect. > > > Not bad. However, you say 75% of the storage is available for data. This > is incorrect. > > I have a 9 drive raid5, with 200Gb drives. One is hotspare, leaving 8 > drives in the array. I have 1.4Tb of storage available. > > As you increase the number of drives, the overhead of raid5 decreases - > This applies to the CPU overhead for the parity stripe as well as the > impact on storage available. > > Also, more spindles == more performance, almost always. Going for > smaller drives will get you more "bang for the buck" than larger, as > long as you have plenty of controller ports available. > > As another hint - Start your RAID small and grow it. Buying your disks > all at once means that they'll all tend to fail at once - Which is not a > good thing. Unfortunately, there's no easy, reliable way I've found to > extend a RAID without backing up (somewhere really huge :) and restoring. > > Cheers, > > Allan. > ___ > mythtv-users mailing list > mythtv-users@mythtv.org > http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users > Thanks for the correction, I'll fix the article. The storage efficency for me is 75% because I'm using 4 drives. The actual calculation is (n-1)/n * 100 = % efficiency. With 8 drives as you have it is 87.5%. While the number of drives in a RAID 5 array is theoretically unlimited, some recommend no more than 14 drives. The problem is RAID 5 is hosed if two drives fail. The more drives you have the more statistically likely you are to suffer a two drive failure. I have seen the argument that you shouldn't get all of your drives at the same time since they are more likely to fail at about the same time. As mentioned above that is bad with RAID 5. Thanks for the feedback. -- Tim www.magicitx.com ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
Allan Stirling wrote: As another hint - Start your RAID small and grow it. Buying your disks all at once means that they'll all tend to fail at once - Which is not a good thing. Unfortunately, there's no easy, reliable way I've found to extend a RAID without backing up (somewhere really huge :) and restoring. According to their pages, evms.sourceforge.net says it can do RAID 5 grow/shrink on Linux http://evms.sourceforge.net/ Kevin ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
Re: [mythtv-users] INFO: RAID comparison for MythTV
MagicITX wrote: A week or so ago there was a thread discussing RAID for MythTV. We did a test comparing software RAID 10 and RAID 5 on a mythbackend server. You can find the article on our site if you're interested. Basically we found that the parity calculation for RAID 5 had a minimal impact on the performance of the system. We did see much better throughput under RAID 10 as one would expect. Not bad. However, you say 75% of the storage is available for data. This is incorrect. I have a 9 drive raid5, with 200Gb drives. One is hotspare, leaving 8 drives in the array. I have 1.4Tb of storage available. As you increase the number of drives, the overhead of raid5 decreases - This applies to the CPU overhead for the parity stripe as well as the impact on storage available. Also, more spindles == more performance, almost always. Going for smaller drives will get you more "bang for the buck" than larger, as long as you have plenty of controller ports available. As another hint - Start your RAID small and grow it. Buying your disks all at once means that they'll all tend to fail at once - Which is not a good thing. Unfortunately, there's no easy, reliable way I've found to extend a RAID without backing up (somewhere really huge :) and restoring. Cheers, Allan. ___ mythtv-users mailing list mythtv-users@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users