SAN kit purchased, no turning back!

[MrWu]

Cheers Synack,

 

Would you attempt RAID 50 for a bit more performance or would RAID 6 be a more fault tolerant setup ?

 

I won't be able to buy another SAN for replication but thinking about using Veeam to replicate VMs on seperate physical file servers running Hyper-v or another dedicated dl180 server with a Phyiscal DC(cobbled from older servers) on another site replicating via Veeam, so if server room or SAN goes, we can spool those up.:-)

[SYNACK]

Cheers Synack,

 

Would you attempt RAID 50 for a bit more performance or would RAID 6 be a more fault tolerant setup ?

 

I won't be able to buy another SAN for replication but thinking about using Veeam to replicate VMs on seperate physical file servers running Hyper-v or another dedicated dl180 server with a Phyiscal DC(cobbled from older servers) on another site replicating via Veeam, so if server room or SAN goes, we can spool those up.:-)

You loose a huge amount of space from 5+0 and the speed difference is not that large. I'd go with 6 personally. Depending on the SAN you may be able to snapshot whole LUNs in the SAN itself then backup from those.

[twin--turbo]

We have 2 twin Shelf P4500's , it's a dam good job they are twin shelf as they have a software bug that causes them to reboot occasionally. One went down last week and caused one of our Citrix XenServer hosts to fall over. Fortunatly HA had already moved the VM's to the other XS's... Not very impressive though from HP. The update will be applied when we have a suitable clear window.

 

Our old Dell/EMC Clarrion still chuggs on happily after nearly five years but it's hardly stressed now.

 

Rob

[MrWu]

Mind all over the place tonight.

 

Now working out the Power Consumption for UPS sizing etc.

 

The HP Power Advisor Tool HP Power Advisor great little utility to work out your VA rating for your servers with options, which will help you opt for the right PSU from HP as well, I was suprised how much power a 10Gbe card takes. (but the only snag with the Power Advisory Tool is that it doesn't give power rating for the MSA P2000 SANS) surprised how low actual VA ratings are required, I did calculate at 100 percent utilisation so over compensating somewhat..looks like 8kva will cover a stack load of servers and switches (15 servers, firewall, switches and routers - mmm might get back to that in the morning when my mind is clear) really liking what Borri offers but due to red tape getting an account opened with them through our finance department(long story) I might not be able to book an install for summer. :-( so could well have to go back to APC.

Edited July 5, 2012 by MrWu

[PotNoodleTech]

You loose a huge amount of space from 5+0 and the speed difference is not that large. I'd go with 6 personally. Depending on the SAN you may be able to snapshot whole LUNs in the SAN itself then backup from those.

 

Have you benchmarked this yourself? In my systems RAID50 batters raid 6 for read and especially write speed. It's also more reliable generally.

 

There is no place for Raid 5 based on it's awful scalability and poor write performance.

[SYNACK]

Have you benchmarked this yourself? In my systems RAID50 batters raid 6 for read and especially write speed. It's also more reliable generally.

 

There is no place for Raid 5 based on it's awful scalability and poor write performance.

It depends on the hardware, the cache and the usage profile. You are unlikely to need performance levels higher than 6 and you loose huge amounts of space. 5+0 can actually have less reliability as 6 will cope with any two drives failing, 5+0 can be killed by a two drive failure 50/50 odds. Raid5 is one of the most scalable sets but I agree with a large number of drives 6 should be the minimum. I'd trust 5 over 1 or 1+0 any day.

[MrWu]

I might be wrong but just stuck 12 x 600Gb Drive in the RAID calculator for RAID 6 and RAID 50 and they both spit out 6TB..is the calculator incorrect?

 

What I could do though is use RAID 6 for the 600Gb drives and RAID 10 or RAID 50 for the 300gb ones. for mixture of performance and resilience. :-)

[DrPerceptron]

Have the 12 HDDs divided into seperate RAIDS according to SQL / Exchange etc..but risk of over complication.

 

We've split off SQL Databases and Exchange Databases into dedicated RAID10 groups, then use another RAID 10 set for all the Host OSes.

 

I don't see that it complicates things etc, but it does mean that you need to have some extra disks over creating one giant 'super group'.

[SYNACK]

I might be wrong but just stuck 12 x 600Gb Drive in the RAID calculator for RAID 6 and RAID 50 and they both spit out 6TB..is the calculator incorrect?

 

What I could do though is use RAID 6 for the 600Gb drives and RAID 10 or RAID 50 for the 300gb ones. for mixture of performance and resilience. :-)

My bad was reading it as 5+1 instead, 6 is still more redundant.

[PotNoodleTech]

It depends on the hardware, the cache and the usage profile. You are unlikely to need performance levels higher than 6 and you loose huge amounts of space. 5+0 can actually have less reliability as 6 will cope with any two drives failing, 5+0 can be killed by a two drive failure 50/50 odds. Raid5 is one of the most scalable sets but I agree with a large number of drives 6 should be the minimum. I'd trust 5 over 1 or 1+0 any day.

 

Hmm its not as straight cut as that, as RAID5 and RAID6's overall reliability depend on the number of, speed of and size of disks used. In both R5 and R6 cases the more slower, larger disks used in one array the greater the chance of total loss of the array. As so many people seem to favour using large numbers (10+) of large SATA drives in Raid 5 or 6, they are seriously risking a total faliure (especially with RAID5) as if one drive goes down the time taken to rebuild the array onto the replacement disk is so high that the chance of another disk failing whilst the array is being rebuilt are alot higher.

 

With RAID50 you can keep expanding the number of disks in the array in groups of three and the overall chance of a total faliure stays the same as if you had just 6 disks. It's predictable.

 

Raid6 is good, don't get me wrong, but the rebuild speed penalty on large arrays combined with generally poor write performance for me is absoloutly not worth the extra free space gained by using RAID6. Your milage may vary.

[PotNoodleTech]

It's also worth pointing out that you can customise how much disk space you actually lose on RAID50 by increasing the number of disks in each raid 5 group.

 

E.g if I had 18 x 300GB drives set up as RAID50, I could either go with 6xsets of 3disks RAID5 all in RAID0 (and lose 6 disks worth of space) or I could go with 2 x SETS of 9 disk RAID5 all in RAID0 and only lose 2 x disks worth of space. The former will offer much better write performance at the cost of more disk space, the latter will offer less perfromance but you lose less disk space. So the choice is yours, RAID50 is very customisable.

[SYNACK]

Hmm its not as straight cut as that, as RAID5 and RAID6's overall reliability depend on the number of, speed of and size of disks used. In both R5 and R6 cases the more slower, larger disks used in one array the greater the chance of total loss of the array. As so many people seem to favour using large numbers (10+) of large SATA drives in Raid 5 or 6, they are seriously risking a total faliure (especially with RAID5) as if one drive goes down the time taken to rebuild the array onto the replacement disk is so high that the chance of another disk failing whilst the array is being rebuilt are alot higher.

 

With RAID50 you can keep expanding the number of disks in the array in groups of three and the overall chance of a total faliure stays the same as if you had just 6 disks. It's predictable.

 

Raid6 is good, don't get me wrong, but the rebuild speed penalty on large arrays combined with generally poor write performance for me is absoloutly not worth the extra free space gained by using RAID6. Your milage may vary.

But if you loose two drives across two stripe groups and your toast, you can loose any two under raid6. The set size is an issue with large slow drives for all types.

Edited July 6, 2012 by SYNACK