HP LeftHand P4X00 : Network Raid Explained

Source : HP StorageWorks P4000 SAN Solution user guide

Data protection results from creating data redundancy for volumes on the SAN. Configure data protection levels, called Network RAID, when you create a volume. You can store two, three, or four mirrored copies of your data using Network RAID-10, Network RAID-10+1, or Network RAID-10+2. Network RAID-5 and Network RAID-6 store parity on multiple storage systems in the cluster. Under some workloads that write the data infrequently, Network RAID-5 and Network RAID-6 provide better 190 Provisioning storage capacity utilization and similar high availability to Network RAID-10 and Network RAID-10+1. Because data is stored redundantly on different storage systems, all data protection levels are tied to the number of available storage systems in a cluster

 

Basic Terminology

  • Storage System : This is an individual storage node.
  • Network RAID : The redundancy system which can be configured across multiple boxes.
  • (Basic) Raid Level : The raid level you configure on a disk group within one storage node.

So you can configure a local “raid level” on the “storage system”, and can provide redundancy across multiple nodes with “Network Raid”

 

Supported Local Raid Levels

  • RAID 0 – Striping (No Fault Tolerance) : Offers the greatest capacity and performance without data protection. If you select this option, you will experience data loss if a hard drive that holds the data fails. However, because no logical drive capacity is used for redundant data, this method offers the best capacity. This method offers the best processing speed by reading two stripes on different hard drives at the same time and by not having a parity drive.
  • RAID 1 – Mirroring : Offers a good combination of data protection and performance. RAID 1 or drive mirroring creates fault tolerance by storing duplicate sets of data on a minimum of two hard drives. There must be an even number of drives for RAID 1. RAID 1 and RAID 1+0(10) are the most costly fault tolerance methods because they require 50 percent of the drive capacity to store the redundant data. RAID 1 mirrors the contents of one hard drive in the array onto another. If either hard drive fails, the other hard drive provides a backup copy of the files and normal system operations are not interrupted.
  • RAID 1+0 – Mirroring and Striping : Offers the best combination of data protection and performance. RAID 1+0 or drive mirroring creates fault tolerance by storing duplicate sets of data on a minimum off our hard drives. There must be an even number of drives for RAID 1+0. RAID 1+0(10) and RAID 1 are the most costly fault tolerance methods because they require 50 percent of the drive capacity to store the redundant data. RAID 1+0(10) first mirrors each drive in the array to another, and then stripes the data across the mirrored pair. If a physical drive fails, the mirror drive provides a backup copy of the files and normal system operations are not interrupted. RAID 1+0(10) can withstand multiple simultaneous drive failures, as long as the failed drives are not mirrored to each other.
  • RAID 5 : Offers the best combination of data protection and usable capacity while also improving performance over RAID 6. RAID 5 stores parity data across all the physical drives in the array and allows more simultaneous read operations and higher performance. If a drive fails, the controller uses the parity data and the data on the remaining drives to reconstruct data from the failed drive. The system continues operating with a slightly reduced performance until you replace the failed drive. RAID 5 can only withstand the loss of one drive without total array failure. It requires an array with a minimum of three physical drives. Usable capacity is N-1 where N is the number of physical drives in the logical array.
  • Raid 6 : Offers the best data protection and is an extension of RAID 5. RAID 6 uses multiple parity sets to store data and can therefore tolerate up to 2 drive failures simultaneously. RAID 6 requires a minimum of 4 drives. Performance is lower than RAID 5 due to parity data updating on multiple drives. RAID 6 uses two disk for parity; its fault tolerance allows twodisks to fail simultaneously. Usable capacity is N-2 where N is the number of physical drives in the logical array.

 

Network Raid : RAID-10

Network RAID-10 data is striped and mirrored across two storage systems. Network RAID-10 is the default data protection level assigned when creating a volume, as long as there are two or more storage systems in the cluster. Data in a volume configured with Network RAID-10 is available and preserved in the event that one storage system becomes unavailable.
Network RAID-10 is generally the best choice for applications that write to the volume frequently and don’t need to tolerate multiple storage system failures. Such applications include databases, email, and server virtualization. Network RAID-10 is also good for Multi-Site SANs. Using Network RAID-10 in a Multi-Site SAN ensures that data remains available in the event that one site becomes unavailable. However, if one site does go down, the Network RAID-10 volumes are then not protected from complete system failure or reboot.

Network Raid : RAID-10+1

Network RAID-10+1 data is striped and mirrored across three or more storage systems. Data in a volume configured with Network RAID-10+1 is available and preserved in the event that any two storage systems become unavailable. Best applications for Network RAID-10+1 are those that require data availability even if two storage systems in a cluster become unavailable.

Network Raid : RAID10+2

Network RAID-10+2 data is striped and mirrored across four or more storage systems. Data in a volume configured with Network RAID-10+2 is preserved in the event that any three storage systems become unavailable. Network RAID-10+2 is designed for Multi-Site SANs to preserve data in the event of an entire site becoming unavailable. Best use for Network RAID-10+2 volumes is for data that must be synchronously replicated between two locations and that must remain fully redundant in the case of an entire site failure. Using Network
P4000 SAN Solution user guide 193 RAID-10+2 ensures that data remains available after half of the SAN is unavailable, and continues to remain available even with the loss of a single storage system in the remaining site.

Network Raid : RAID-5

Network RAID-5 divides the data into stripes. Each stripe is stored on three of the storage systems in the cluster, and parity is stored on a fourth storage system. The data stripes and parity are distributed evenly over all the systems. Data in a volume configured with Network RAID-5 is available and preserved in the event that any single storage system becomes unavailable.
Network RAID-5 volumes are configured as thin provisioned by default. Network RAID-5 also requires a snapshot schedule. When a new Network RAID-5 volume is created, a basic snapshot schedule is also created. If an existing volume without a snapshot schedule is converted to Network RAID-5, a basic schedule is created when the volume is converted. This basic snapshot schedule can be edited to fit the needs of the volume.
Best applications for using Network RAID-5 volumes include applications with mostly read, sequential workloads, such as file shares and archiving.

Network Raid : RAID-6

Network RAID-6 divides the data into stripes. Each stripe is stored on four of the storage systems in the cluster, and parity is stored on a fifth and sixth system. Data in a volume configured with Network RAID-6 is available and preserved in the event that any two storage systems become unavailable.
Network RAID-6 volumes are configured as thin provisioned by default. Network RAID-6 also requires a snapshot schedule. When a new Network RAID-6 volume is created, a basic snapshot schedule is also created. If an existing volume is converted to Network RAID-6, if that volume does not have a snapshot schedule, a basic schedule is created when the volume is converted. This basic snapshotschedule can be edited to fit the needs of the volume.
Best applications for using Network RAID-6 volumes include applications with mostly read, sequential workloads on larger clusters, such as file shares and archiving.

Netto Usage

Basic Set-up : A Multi-Site SAN, with 4 HP P4500 nodes of 12x600GB

Raw Capacity :  4x12x600GB = 28,8TB

Local RAID5 : 21,2TB Usable

+ NRAID0 : 21,2TB

+ NRAID10 : 10,6TB

+ NRAID10+1 : 7TB

+ NRAID10+2 : 5,3TB

+ NRAID5 : 15,9TB

+ NRAID6 : X (not enough nodes)

 

Advertisements

Virtual Connect : “Patch Panel” or “Active Component”?

A while ago we had an internal discussion whether or not a Virtual Connect (HP Blade Technology) is to be considered as an active component or as a patch panel.  The answer is that it is kinda like a switch. It cannot be seen as an actual switch, where it comes close to a virtual switch (like defined by VMware). More details can be found op page 25 in the document (linked below), where a nice comparison table is listed.

Source :HP Virtual Connect: Common Myths, Misperceptions, and Objections, Second Edition (Google cache / quickviewer)

#1: VC Ethernet is just another switch
Incorrect: While VC uses tried-and-true, IEEE standard, Layer 2 bridging functionality, its primary purpose is to provide many server virtualization and management features that are non-existent in traditional switches. VC may perform some functions like a traditional switch; however, VC has many additional features which clearly distinguish it from a traditional switch. …

#18: VC Ethernet doesn’t provide Layer 3 routing capabilities
Correct: Virtual Connect is not a router, therefore, Virtual Connect does not provide Layer 3 capabilities (routing).

#14: HP server blade NICs stay active even after VC Ethernet uplink failure
Incorrect: Virtual Connect provides many features for ensuring highly available network connectivity for HP server blades. One feature, SmartLink, is used to disable a server blade NIC port anytime the NIC is connected to an external network where all VC uplink(s) have failed. In other words, VC can be configured to proactively disable a server NIC port whenever the server NIC is isolated from the external network. VC’s SmartLink feature, combined with NIC Teaming on the server, allows for highly available network configuration with no single point of failure.

#3: VC Ethernet doesn’t support Spanning Tree (STP)
Correct: Much to the delight of VC users, Spanning Tree support on VC is not needed. VC provides HP server blade network connectivity just like a hypervisor provides virtual server network connectivity and neither of these technologies require Spanning Tree support. VC doesn’t have to support Spanning Tree just like hypervisor hosts don’t have to support it, yet both provide network redundancy and load balancing. Just like a hypervisor host, VC provides network redundancy and load balancing features that are modeled after NIC Teaming/bonding technology instead of switch technologies like Spanning Tree. A Spanning Tree configuration error on any single switch in the data center can negatively affect any other connected switch in the network, in addition to all servers connected to the same network. With Virtual Connect, any redundancy and load balancing configuration problems only affect a single blade enclosure1
Fundamentally, VC doesn’t require support for protocols like STP because VC presents itself to the network as a “termination endpoint”, as does a typical server or a hypervisor host. VC is not and does not present itself as a “transit device”, as does a traditional switch.

#24: VC Ethernet doesn’t support the Cisco Discovery Protocol (CDP)
Correct: VC supports Link Layer Discovery Protocol (LLDP) – the industry standard (IEEE) version of the Cisco proprietary protocol CDP. Many Cisco devices support both CDP and LLDP (for example, 14 ). The use of the IEEE standard version, LLDP, is recommended by HP to ensure customers are not locked into a proprietary protocol.

#17: VC only supports a limited number of VLANs
Correct: The supported limit is 320 VLANs per Virtual Connect Ethernet module when using Shared Uplink Sets. The VC architecture supports 1000+ VLANs per c-Class enclosure.

Welcome to WordPress.com

After having hosted my own blog myself since the beginning, I’ve decided to move to WordPress.com. It might look like an odd choice to make, yet this leaves me without the worries of maintaining (upgrades/upgrades/security/…) of my own.