![]() | |
![]() |
| | Thread Tools | Display Modes |
#1
| |||
| |||
|
#2
| |||
| |||
|
|
Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor |
#3
| |||
| |||
|
|
Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor |
#4
| |||
| |||
|
|
This is typical SAN "engineering". More spindles = more I/O capacity. Except most designers use RAID5 and use up the extra I/O ops in RAID overhead. Most SAN layouts are designed around maximizing storage space, not maximizing I/O operations for SQL server. Unfortunately, most SAN companies believe in their "Magic SAN Dust" that somehow creates I/O cycles out of thin air, so you are stuck with this layout. You must create at least one LUN per instance. These LUNs get presented to all cluster Nodes. The Cluster service arbitrates ownership so that no data corruption occurs and only one Node "owns" a LUN at any given time. Partitioning a LUN gains nothing except additional management complexity. -- Geoff N. Hiten Principal SQL Infrastructure Consultant Microsoft SQL Server MVP "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news 6270607-4FC2-4BF1-BDA7-FE63C1AFC61E (AT) microsoft (DOT) com...Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor . |
#5
| |||
| |||
|
|
Geoff, I was under impressions I will need separate LUNs (mapped to corresponding drives) for tempdb, log, data files, etc. You are saying one LUN per instance is OK. In this case, will I create multple logical drives on top of single LUN? If so, can I define different RAID level (does RAID level even matter in the world of high end SAN)? Sorry, I am pretty novice in the SAN world. I am trying to figure out if single LUN with multple drives (corresponding RAID levels) is the way to go. Thank you, Igor "Geoff N. Hiten" wrote: This is typical SAN "engineering". More spindles = more I/O capacity. Except most designers use RAID5 and use up the extra I/O ops in RAID overhead. Most SAN layouts are designed around maximizing storage space, not maximizing I/O operations for SQL server. Unfortunately, most SAN companies believe in their "Magic SAN Dust" that somehow creates I/O cycles out of thin air, so you are stuck with this layout. You must create at least one LUN per instance. These LUNs get presented to all cluster Nodes. The Cluster service arbitrates ownership so that no data corruption occurs and only one Node "owns" a LUN at any given time. Partitioning a LUN gains nothing except additional management complexity. -- Geoff N. Hiten Principal SQL Infrastructure Consultant Microsoft SQL Server MVP "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news 6270607-4FC2-4BF1-BDA7-FE63C1AFC61E (AT) microsoft (DOT) com...Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor . |
#6
| |||
| |||
|
|
Geoff, I was under impressions I will need separate LUNs (mapped to corresponding drives) for tempdb, log, data files, etc. You are saying one LUN per instance is OK. In this case, will I create multple logical drives on top of single LUN? If so, can I define different RAID level (does RAID level even matter in the world of high end SAN)? Sorry, I am pretty novice in the SAN world. I am trying to figure out if single LUN with multple drives (corresponding RAID levels) is the way to go. Thank you, Igor "Geoff N. Hiten" wrote: This is typical SAN "engineering". More spindles = more I/O capacity. Except most designers use RAID5 and use up the extra I/O ops in RAID overhead. Most SAN layouts are designed around maximizing storage space, not maximizing I/O operations for SQL server. Unfortunately, most SAN companies believe in their "Magic SAN Dust" that somehow creates I/O cycles out of thin air, so you are stuck with this layout. You must create at least one LUN per instance. These LUNs get presented to all cluster Nodes. The Cluster service arbitrates ownership so that no data corruption occurs and only one Node "owns" a LUN at any given time. Partitioning a LUN gains nothing except additional management complexity. -- Geoff N. Hiten Principal SQL Infrastructure Consultant Microsoft SQL Server MVP "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news 6270607-4FC2-4BF1-BDA7-FE63C1AFC61E (AT) microsoft (DOT) com...Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor . |
#7
| |||
| |||
|
|
The only reason to create separate LUNS is if there are separate sets of disks behind those LUNS. Since your SAN engineer told you they all come from the same pool, further partitioning gains you no benefit. Again, all the disks get pooled together in a single large RAID set (according to your engineer). You get LUNS sliced off of this large RAID set. The RAID set, not the LUNS is the physical storage layer. -- Geoff N. Hiten Principal SQL Infrastructure Consultant Microsoft SQL Server MVP "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news:46DF762B-FEF8-49CF-B5FC-04BFD0BD9265 (AT) microsoft (DOT) com... Geoff, I was under impressions I will need separate LUNs (mapped to corresponding drives) for tempdb, log, data files, etc. You are saying one LUN per instance is OK. In this case, will I create multple logical drives on top of single LUN? If so, can I define different RAID level (does RAID level even matter in the world of high end SAN)? Sorry, I am pretty novice in the SAN world. I am trying to figure out if single LUN with multple drives (corresponding RAID levels) is the way to go. Thank you, Igor "Geoff N. Hiten" wrote: This is typical SAN "engineering". More spindles = more I/O capacity. Except most designers use RAID5 and use up the extra I/O ops in RAID overhead. Most SAN layouts are designed around maximizing storage space, not maximizing I/O operations for SQL server. Unfortunately, most SAN companies believe in their "Magic SAN Dust" that somehow creates I/O cycles out of thin air, so you are stuck with this layout. You must create at least one LUN per instance. These LUNs get presented to all cluster Nodes. The Cluster service arbitrates ownership so that no data corruption occurs and only one Node "owns" a LUN at any given time. Partitioning a LUN gains nothing except additional management complexity. -- Geoff N. Hiten Principal SQL Infrastructure Consultant Microsoft SQL Server MVP "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news 6270607-4FC2-4BF1-BDA7-FE63C1AFC61E (AT) microsoft (DOT) com...Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor . . |
#8
| |||
| |||
|
|
...in our setup we always go for the same number of luns for each sql (and sql cluster) build purely on "standard" build configuration, even if each lun does not map directly to a seperate physical array. Generally though I get to have a large say in the way disks get carved up so thankfully (usually) each lun maps to a seperate array. I dont like OS partitioning mostly because I prefer the disk architecture to be independant of OS (since this can introduce other issues). "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news:46DF762B-FEF8-49CF-B5FC-04BFD0BD9265 (AT) microsoft (DOT) com... Geoff, I was under impressions I will need separate LUNs (mapped to corresponding drives) for tempdb, log, data files, etc. You are saying one LUN per instance is OK. In this case, will I create multple logical drives on top of single LUN? If so, can I define different RAID level (does RAID level even matter in the world of high end SAN)? Sorry, I am pretty novice in the SAN world. I am trying to figure out if single LUN with multple drives (corresponding RAID levels) is the way to go. Thank you, Igor "Geoff N. Hiten" wrote: This is typical SAN "engineering". More spindles = more I/O capacity. Except most designers use RAID5 and use up the extra I/O ops in RAID overhead. Most SAN layouts are designed around maximizing storage space, not maximizing I/O operations for SQL server. Unfortunately, most SAN companies believe in their "Magic SAN Dust" that somehow creates I/O cycles out of thin air, so you are stuck with this layout. You must create at least one LUN per instance. These LUNs get presented to all cluster Nodes. The Cluster service arbitrates ownership so that no data corruption occurs and only one Node "owns" a LUN at any given time. Partitioning a LUN gains nothing except additional management complexity. -- Geoff N. Hiten Principal SQL Infrastructure Consultant Microsoft SQL Server MVP "Igor Marchenko" <IgorMarchenko (AT) discussions (DOT) microsoft.com> wrote in message news 6270607-4FC2-4BF1-BDA7-FE63C1AFC61E (AT) microsoft (DOT) com...Hello! We will be installing SQL Server 2008 Active/Active Cluster on Windows 2008. SAN engineer told us that all space will be allocated accross the same set of drives. I know this is not recommended because of potential IO issues, but I guess this is something we would have to live with. The question we would have to answer right now is whether we should create one LUN per Node (my understanding is that the same LUN can not be shared by two nodes) and then create logical drives from single LUN or one LUN per logical drive. I suppose there will be no difference performance wise since LUNs will be connected to the same set of drives, correct? Also, SAN engineer told us that RAID level is irrelevant which I have hard time believing to. Thank you in advance, Igor . . |
![]() |
| Thread Tools | |
| Display Modes | |
| |