zeusindc
asked on
Best way to configure RAID/ best practice
I am new to server management. I have few questions regarding RAID.
I have to built almost 2 servers every week, so I am trying to figure out best way/practice to configure RAID.
min. there are 5 drives on each of these machines, some have more.
for instance i have 5 x 72GB drives, and storage space is not an issue. redundancy is more imp.
Is it ok for me to create 2 RAID 1s and the 5th disk as hot spare for both?
what is the min. requirement to have RAID 10 with hot spare?
I have to built almost 2 servers every week, so I am trying to figure out best way/practice to configure RAID.
min. there are 5 drives on each of these machines, some have more.
for instance i have 5 x 72GB drives, and storage space is not an issue. redundancy is more imp.
Is it ok for me to create 2 RAID 1s and the 5th disk as hot spare for both?
what is the min. requirement to have RAID 10 with hot spare?
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
if I create RAID 1 for OS and RAID 5 for data then I wont have hot spare?
How is RAID 10 a better option, if cost is not an issue? if I have 7 disks, can i configure a RAID 10 of 6 disks and 7th as hotspare? this mean i will have one big logical drive. so if I install OS on this logical drive( obviously will create partitions) but OS crashes then i will lose the DATA too?
so the best way i see if to create a RAID for OS and that logical drive should only have OS. create another RAID 5 or 1 for DATA.
How is RAID 10 a better option, if cost is not an issue? if I have 7 disks, can i configure a RAID 10 of 6 disks and 7th as hotspare? this mean i will have one big logical drive. so if I install OS on this logical drive( obviously will create partitions) but OS crashes then i will lose the DATA too?
so the best way i see if to create a RAID for OS and that logical drive should only have OS. create another RAID 5 or 1 for DATA.
First I would do one raid for all, not seperate for os and data
We use RAID 10 on all our servers now as it gives you performance and reliabilty. You can configure RAID with 7 disks 6 in RAID 1 hot spare.
We use RAID 10 on all our servers now as it gives you performance and reliabilty. You can configure RAID with 7 disks 6 in RAID 1 hot spare.
You would only raid different depending on your requirements. Having data and OS in same raid group is risky, in my opinion. You'd want to split your I/O between the channels, if you can afford to.
Depending on what your reqs are (perf vs. redund) will dictate your RAID type.
I'd stick to 2xR1 (72GB)for your OS 3xR5 for your data. (173GB)
I wouldn't worry about hotspare, you're paying for the disk, might as well use it. You couldn't do this if you wanted a hotspare anyway. You'll still have redundancy.
If you have 7 disks then, of course, you'd have your hotspare.
Performance is only an issue when you're dealing with a lot of write activity (Exhange, SQL, etc). I'd stick to R5 data volumes if I were you.
Depending on what your reqs are (perf vs. redund) will dictate your RAID type.
I'd stick to 2xR1 (72GB)for your OS 3xR5 for your data. (173GB)
I wouldn't worry about hotspare, you're paying for the disk, might as well use it. You couldn't do this if you wanted a hotspare anyway. You'll still have redundancy.
If you have 7 disks then, of course, you'd have your hotspare.
Performance is only an issue when you're dealing with a lot of write activity (Exhange, SQL, etc). I'd stick to R5 data volumes if I were you.
it depends mostly on wht the servers are being used for, based on that you could configure your servers to get the most out of the boxes.
most of the apps work great with the below keeping in mind various factors, performance, cost, redundancy etc.,
2x72GB Raid 1 for OS
2x72GB Raid 5 for Data
1x72GB as Global Hotspare, either raid fails the hotspare is avaialbe for both the raid configs
Raid 10 is otherwise also known as Raid 0+1 or (Raid 0/1) it requires minimum 4 Drives, you would get best performance and highest data protection i.e, it can tolerate multiple drive failures on the other hand the disadvantages is high redudancy cost overhead, and since all the data is duplicated you would require twice the storage capacity.
most of the apps work great with the below keeping in mind various factors, performance, cost, redundancy etc.,
2x72GB Raid 1 for OS
2x72GB Raid 5 for Data
1x72GB as Global Hotspare, either raid fails the hotspare is avaialbe for both the raid configs
Raid 10 is otherwise also known as Raid 0+1 or (Raid 0/1) it requires minimum 4 Drives, you would get best performance and highest data protection i.e, it can tolerate multiple drive failures on the other hand the disadvantages is high redudancy cost overhead, and since all the data is duplicated you would require twice the storage capacity.
sorry typo error for the raid 5 hard disks
3x72GB Raid 5 for data (minimum 3 hard disks are required for RAID5 if you have 7 you could use 4x72GB)
3x72GB Raid 5 for data (minimum 3 hard disks are required for RAID5 if you have 7 you could use 4x72GB)
If you don't need more than 72GB of data then might as well have RAID1 for the OS, RAID1 for the data and a shared hot spare, will be faster than RAID5 and presuming you have a decent RAID controller you could always migrate from RAID1 to RAID5 later on.
It's really a waste of money to have a hot spare with so few drives though, unless you use cheap SATA disks there's very little chance of a second disk failing before the first failed disk gets replaced. (and your 72GB disks won't be SATA because that is one of the sizes of SCSI disks).
It's really a waste of money to have a hot spare with so few drives though, unless you use cheap SATA disks there's very little chance of a second disk failing before the first failed disk gets replaced. (and your 72GB disks won't be SATA because that is one of the sizes of SCSI disks).
Our 'standard' build at work has 6 drives - we set up a RAID1 set for the OS (And transaction logs is applicable, on a 2nd partition), and the remaining 4 disks configured with 3 as RAID 5 with the 4th as online spare. This raid 5 set is for the data.
Our rationale is that we can suffer 1 disk failure for the OS, and (any) 2 disk failures for the data (the first kicks in the on-line spare and once that's done rebuilding there's no performance hit. the second results in parity calculations coming into play and thus performance degradation). If we lose the OS, it's easily rebuilt from last nights back-up, but losing a day's worth of data of data is usually unacceptable.
The problem, imho, with RAID10/01 whatever, is that it restricts which drives can fail before data loss occurs. eg, RAID0 disks 2+3, then mirror them to 4+5. Losing 1 drive from each Raid0 set loses data - thus losing disk 2+4 for example means restoer time. losing 2+3 OR 4+5 is ok though.
Similarly, RAID 1 disks 2+3 and 4+5, then Raid 0 the 2 mirrors. Now if you lose 2+4 together, you're ok since 3+5 still maintain the RAID0 set. but lose 2+3 together and your're restoring.
If you have more disks for on-line spares then the limitation is mitigated somewhat. For us, having an on-line spare is important for 2 reasons - 1) if we lose a disk in RAID5, we don't want to suffer really bad performance until the disk is replaced - the on-line spare kicks in and rebuilds in a couple of hours and performance is back to normal. 2) our datacentre is remote, so changing a disk same-day is sometimes not possible. Whilst David_Fong is correct that SCSI is more reliable than SATA (by far) I have witnessed multiple disks failing within days of each other, presumeably becasue they're out of the same 'batch'. I've had a disk fail at midnight, and another at 5am in the same server and RIAD5 set - luckily the on-line spare kicked in and the rebuild completed before the second failure so we lost no data. As a result, I'd prefer to keep a spare any day.
Having said that, if your controller support RAID6, it too protects against 2 disks failing without losing information. But this too needs 4 disks.
With only 5 disks to play with, I'd be inclined to do 2 RAID1's, 1 for the OS and 1 for data, and allocate the 5th as spare to the data mirror. This way you can still lose upto 2 disks without losing your data (depending how close to each other they fail). Losing the OS is usually not such an issue compared to losing the user data - ie, going back to last nights OS is usually OK.
"How is RAID 10 a better option, if cost is not an issue? if I have 7 disks, can i configure a RAID 10 of 6 disks and 7th as hotspare? this mean i will have one big logical drive. so if I install OS on this logical drive( obviously will create partitions) but OS crashes then i will lose the DATA too?"
yes, you can configure it that way and no, crashing the OS shouldn't lose data. However, depending what you use the server for, placing the OS and data on the same drives but different partitions, may cause disk contention and thus poor performance.
Our rationale is that we can suffer 1 disk failure for the OS, and (any) 2 disk failures for the data (the first kicks in the on-line spare and once that's done rebuilding there's no performance hit. the second results in parity calculations coming into play and thus performance degradation). If we lose the OS, it's easily rebuilt from last nights back-up, but losing a day's worth of data of data is usually unacceptable.
The problem, imho, with RAID10/01 whatever, is that it restricts which drives can fail before data loss occurs. eg, RAID0 disks 2+3, then mirror them to 4+5. Losing 1 drive from each Raid0 set loses data - thus losing disk 2+4 for example means restoer time. losing 2+3 OR 4+5 is ok though.
Similarly, RAID 1 disks 2+3 and 4+5, then Raid 0 the 2 mirrors. Now if you lose 2+4 together, you're ok since 3+5 still maintain the RAID0 set. but lose 2+3 together and your're restoring.
If you have more disks for on-line spares then the limitation is mitigated somewhat. For us, having an on-line spare is important for 2 reasons - 1) if we lose a disk in RAID5, we don't want to suffer really bad performance until the disk is replaced - the on-line spare kicks in and rebuilds in a couple of hours and performance is back to normal. 2) our datacentre is remote, so changing a disk same-day is sometimes not possible. Whilst David_Fong is correct that SCSI is more reliable than SATA (by far) I have witnessed multiple disks failing within days of each other, presumeably becasue they're out of the same 'batch'. I've had a disk fail at midnight, and another at 5am in the same server and RIAD5 set - luckily the on-line spare kicked in and the rebuild completed before the second failure so we lost no data. As a result, I'd prefer to keep a spare any day.
Having said that, if your controller support RAID6, it too protects against 2 disks failing without losing information. But this too needs 4 disks.
With only 5 disks to play with, I'd be inclined to do 2 RAID1's, 1 for the OS and 1 for data, and allocate the 5th as spare to the data mirror. This way you can still lose upto 2 disks without losing your data (depending how close to each other they fail). Losing the OS is usually not such an issue compared to losing the user data - ie, going back to last nights OS is usually OK.
"How is RAID 10 a better option, if cost is not an issue? if I have 7 disks, can i configure a RAID 10 of 6 disks and 7th as hotspare? this mean i will have one big logical drive. so if I install OS on this logical drive( obviously will create partitions) but OS crashes then i will lose the DATA too?"
yes, you can configure it that way and no, crashing the OS shouldn't lose data. However, depending what you use the server for, placing the OS and data on the same drives but different partitions, may cause disk contention and thus poor performance.
http://www.acnc.com/04_01_05.html
what ever you do for raid I would have 1 hot spare