Contents:
As companies become ever more reliant on server-based applications and data, any event that interrupts access to that data -even for a few hours -threatens productivity and profitability. To ensure uninterrupted business activity, IT managers need flexible storage solutions that enable them to create high-availability environments that help prevent outages to ensure business continuance under all conditions.
Storage Area Networks (SANs) are ideal for facilitating such a high-availability environment. A SAN interconnects multiple shared storage environments and servers into a storage network, with all storage resources available to any server, regardless of physical location. SANs offer the advantages of built-in redundancy, dynamic failover protection and automatic traffic rerouting. A high-availability SAN can also be tuned to enable load sharing to significantly improve overall performance.
Creating a high-availability SAN with Adaptec SANbloc 2502 storage array, Emulex LP10000 HBA and Vixel InSpeed 9050FF switch
The Adaptec SANbloc 2Gb RAID is a Fibre Channel RAID that delivers the price/performance, availability and flexibility to meet today's real-world enterprise requirements. Based on a modular design, it can be scaled in multiple dimensions, enabling flexible configuration of capacity, performance and functionality.
The SANbloc RAID is designed for maximum availability, with redundant, hot-swappable RAID controllers, dual host ports, power supplies, power connections, advanced cooling modules, port bypass circuitry and environmental monitoring, and battery-protected mirrored write cache.
The configurations described and tested in this Solution Report combine the Adaptec FS4500 Storage Array with Emulex LP10000 Fibre Channel Host Bus Adapter and Vixel InSpeed 9050FF loop switch in a Fibre Channel SAN environment. Three SAN configurations were tested:
- Fabric SAN Configuration
- Windows 2003 2-node Cluster Configuration
- Microsoft Exchange 2003 LoadSim Performance Configuration
Note: Various Fibre Channel peripheral devices were tested randomly in different configurations from those listed below.
1. Fibre Channel Disk Array information
Brand and Model: Adaptec SANbloc 2502 (Dual Controller)
Disk Array Firmware Level: 9.40.00
Enclosures: 2 cascaded
Number of Disk Drives: 28
2. Fibre Channel Switch information
Brand and Model: Vixel InSpeed 9050FF
Switch Type: Fabric Switch
Firmware Level: 1.02.00290
Switch used for I/O Testing:
Brand and Model: Vixel InSpeed 355
Switch Type: Loop Switch
Firmware Level: 2.04
3. SFP information
Brand and Model: Finisar FTRJ-8519-3-2-5
Brand and Model: IBM SFF-PGG-200-M5-SN-1
Brand and Model: Stratos MPLC-20-8-1-8L
Brand and Model: Agilent HFBR-5720L
4. Fibre Channel HBA information
Brand and Model: Emulex LP10000
Firmware Level: 1.81A1 and 1.90a4
BIOS Version: Fcode Firmware 3.21a2
Driver Version: 5-5.10a10
PCI/PCI-X: PCI-X
5. Application information
Backup application: Computer Associates ARCserve
v11.1 for Windows
Server Application:Microsoft Exchange 2003 Server
6. Operating system information
Windows 2003 (Enterprise Server)
Windows 2000 SP4 (Advance Server)
7. Management Software
Spheras RAID Watcher v1.53
HBAnywhere v.2.0a13
8. Test Tools
NTstress v4.3 (block-level test tool)
NTFileCopy v1.0 (file-level test tool)
Microsoft LoadSim 2003
Topology 1 Information
Transparent Failover Configuration (Active/Passive)
Server 1
Make and Model: Supermicro SS8-GM
CPU: P4 /3.06GHz Xeon (x2)
BIOS: AMI
Chipset: ServerWorks GC-SL
Operating System: Windows 2003
HBA: LP10000
SFP: Stratos
Storage: LV0 -RAID10; 150GB(128K stripe size);
Primary Controller Port 0
Server 2
Make and Model: ASUS TR-DLS
CPU: P3 /1.0GHz
BIOS: AMI 7.00
Chipset: ServerWorks GC-LE
Operating System: Windows 2000 SP4
HBA: LP10000
SFP: IBM
Storage: LV0 -RAID10;150GB(128K stripe size);
Primary Controller Port 0
|
|
Transparent Failover Configuration (Active/Active)
Server 1
Make and Model: Supermicro SS8-GM
CPU: P4 /3.06GHz Xeon (x2)
BIOS: AMI
Chipset: ServerWorks GC-SL
Operating System: Windows 2003
HBA: LP10000
SFP: Stratos
Storage: LV0 -RAID10; 150GB(128K stripe size);
Primary Controller Port 0
Server 2
Make and Model: ASUS TR-DLS
CPU: P3 /1.0GHz
BIOS: AMI 7.00
Chipset: ServerWorks GC-LE
Operating System: Windows 2000 SP4
HBA: LP10000
SFP: IBM
Storage: LV0 -RAID10;150GB(128K stripe size);
Secondary Controller Port 0
|

Topology 1:Fabric SAN Configuration
Test results
| Test Case Number |
Test Title |
Storage Path |
Results |
| 1001 |
LUN zoning or disk assignment with management software to ensure each server has its own storage (Use multiple loop for FC4100 if supported) |
S1=>(HBA1,FCDA1)->LV0
S2=>(HBA2,FCDA1)->LD0 |
PASS |
| 1002 |
Disk Array LUN Boot with associated OS installation |
S1=>(HBA1,FCDA1)->LV0
S2=>(HBA2,FCDA1)->LD0 |
PASS |
| 1003 |
NTStress |
S1=>(HBA1,FCDA1)->LV0
S2=>(HBA2,FCDA1)->LD0 |
PASS |
| 1004 |
NTFileTest |
S1=>(HBA1,FCDA1)->LV0
S2=>(HBA2,FCDA1)->LD0 |
PASS |
| 1005 |
Dynamic configuration test (cable pull, target power recycle, switch port enable-disable/power recycle, etc.) |
S1=>(HBA1,FCDA1)->LV0
S2=>(HBA2,FCDA1)->LD0 |
PASS |
| 1006 |
Transparent RAID controller failover (non multi-path failover) |
S1=>(HBA1,FCDA1)->LV0
S2=>(HBA2,FCDA1)->LD0 |
Active/Passive-PASS
Active/Active- PASS |
| 1007 |
ARCserve v11.1 for Windows installation and configuration |
S1=>(HBA1,FCDA1)->LV0 |
PASS |
| 1008 |
Perform disk-to-disk backup test |
S1=>(HBA1,FCDA1)->LV0 |
PASS |
Topology 2 Information
Server 1
Make and Model: Supermicro X5DA8
CPU: P4 /3.2GHz Xeon (x2)
BIOS: Phoenix
Chipset: Intel E7505
Operating System: Windows 2003
HBA: LP10000
SFP: Finisar
Server 2
Make and Model: Supermicro X5DPL
CPU: P4 /3.0Ghz Xeon
BIOS: Phoenix
Chipset: Intel E7501
Operating System: Windows 2003
HBA: LP10000
SFP: Agilent
|
PDC
Make and Model: Micro Star 649D Pro
CPU: P3 /700MHz
BIOS: Award
Chipset: VIA 694XDP
Operating System: Windows 2003
|

Topology 2: Windows 2003 2-node Cluster Configuration
Test results
| Test Case Number |
Test Title |
Storage Path |
Results |
| 1101 |
2-Node Cluster Install Test |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
PASS |
| 1102 |
Cluster Node add/remove Test |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
PASS |
| 1103 |
Cluster Node Failure Test |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
PASS |
| 1104 |
Cluster SAN Connection Failure Test |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
PASS |
| 1105 |
Cluster Node Network Failure Test |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
PASS |
| 1106 |
MS Exchange 2003 Cluster with LoadSim Test (50 simulated users) |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
PASS |
| 1107 |
MS Cluster HCT - 2 tests |
S1=>(HBA1, FCDA1)
S2=>(HBA2, FCDA1) |
Move N Node -PASS
Crash N Node -PASS |
Topology 3 Information
Server 1
Make and Model: Supermicro X5DA8
CPU: P4 /3.2GHz Xeon (x2)
BIOS: Phoenix
Chipset: Intel E7505
Operating System: Windows 2003
HBA: Emulex LP10000
SFP: Agilent
Storage:
LV0-RAID10
150GB (128K stripe size)
Primary Controller Port 0
LV1-RAID10
150GB (128K stripe size)
Primary Controller Port 0
LV2-RAID10
150GB (128K stripe size)
Primary Controller Port 1
|
PDC
Make and Model:Micro Star 649D Pro
CPU: P3 /700MHz
BIOS: Award
Chipset: VIA 694XDP
Operating System: Windows 2003
Client Load Generators
Make and Model: Supermicro Super Server 5011E
CPU: P3 /1GHz
BIOS: AMI
Chipset: Intel 815E
|

Topology 3: Microsoft Exchange 2003 LoadSim Performance Configuration
Test results
| Test Case Number |
Test Title |
Storage Path |
Results |
| 1201 |
LoadSim performance test |
S1=>(HBA1, FCDA1) |
PASS |
For Exchange Server 2003, performance is measured using the MAPI Messaging Benchmark 3 (MMB3) included in the Microsoft LoadSim 2003 utility. The benchmark measures the performance of an Exchange Server and provides the results as a function of response time. In particular, the 95th percentile response means that the vast majority of the clients attached to the Exchange Server under test will experience this response time when accessing the Exchange Server. Microsoft guidelines suggest that the weighted average of Load-Sim-measured 95th-percentile responses measure at less than 1 second (expressed as 1000ms).
Below are the 95th Percentile Response Time results with the Emulex LP10000 HBA, compared to those with the Emulex LP1050 HBA.
|
95th Percentile Score (in milliseconds) |
|
Emulex LP10000 |
Emulex LP1050 |
| Client 1 |
1062 |
1372 |
| Client 2 |
865 |
1141 |
| Client 3 |
832 |
1111 |
| Client 4 |
991 |
1312 |
| Client 5 |
631 |
791 |
| Average |
876 |
1145 |
Additional performance information
Slow I/O performance was encountered when using PCI-X 100MHz slots on the Supermicro X5DA8 motherboard. The LoadSim 95th percentile results were substantially (more than 6 seconds) slower when the LP10000 was installed in a PCI-X 100MHz slot.
Module/Test Case No. |
Test Suite / Test Title |
Test Case Description |
| 1000 |
General Tests |
| 1001 |
LUN zoning or disk assignment with management software to ensure each server has its own storage |
Use associated management software to create LUNs and assign them to different servers. |
| 1002 |
Disk Array LUN Boot with associated OS installation |
Use the assigned FC LUNs to install operating systems.
Boot from these LUNs after OS installation. |
| 1003 |
NTStress Test |
Run NTStress test tools with the FC LUNs. |
| 1004 |
NTFileTest Test |
Run NTFileTest (file copy/compare) test tool with the FC LUNs. |
| 1005 |
Dynamic configuration test (cable pull, target power recycle, switch port enable-disable/power recycle, etc.) |
Single host join with I/O.
Configure one to many servers in the SAN, start I/O from all but one of the hosts.
While I/O is running join the SAN with the other host, ensure that the I/O recovers and there is no data miscompare.
Single host restart with I/O
Configure one to many hosts in a SAN, start I/O from all but one of the servers. While I/O is running restart the server with no I/O, ensure that the I/O on other hosts recovers and there is no data miscompare
Cable pull with I/O
Configure two to many hosts in a SAN, start I/O from all of the servers.
While I/O is running pull the FC cable from a storage device that has I/O running to it, wait 5 to 20 seconds.
Re-insert the FC cable into the storage device, ensure that all I/O to all hosts recovers and there are no data miscompares.
Device pull from storage enclosure with I/O
Configure two to many hosts in a SAN, start I/O from all of the servers.
While I/O is running pull a device (hard disk, etc) from any storage enclosure that has I/O running to it, wait 5 to 20 seconds.
Re-insert the device into the storage enclosure, ensure that all I/O to all hosts recovers and there are no data miscompares.
Interconnect restart.
Configure two to many hosts in a SAN, start I/O from all of the servers.
Unplug the power cord from the hub/switch while I/O is running.
Plug the cord backs in and wait for the hub/switch to boot. Ensure that when the hub/switch comes back up all hosts resume I/O without data miscompare. |
| 1006 |
Transparent RAID controller failover (non multi-path failover) |
Configure the disk array to allow transparent failover of controller (if the disk array supports
Connect the hosts to the switch and run IOs with their associate LUNs/disks.
Fail one controller by pull it out of the enclosure.
Verify the I/Os continue on hosts. |
| 1007 |
ARCserve v11.1 for Windows installation and configuration |
Set up and configure the ARCserve 11.1 for Windows server on a Windows 2003/2000 host.
Create file system devices with the FC LUNs or disks. |
| 1010 |
Disk-to-Disk backup |
Perform disk-to-disk backup with scheduled backup and source >= 30GB |
| 1100 |
Cluster Tests |
| 1101 |
2-Node Cluster Install Test |
Verify that you can successfully install and setup a 2-Node Cluster and that both Nodes report active in the Cluster Management Utility for the OS you are using. |
| 1102 |
Cluster Node add/remove Test |
Using the Cluster utility for the OS you are testing REMOVE one of the nodes from the Cluster.
Ensure that the resources, which were running on that Cluster node, migrate to a second node in the Cluster.
After the resource migrates, re-join the Cluster with the node you removed.
Migrate the resource back to the joined Cluster and verify its operation. |
| 1103 |
Cluster Node Failure Test |
While all nodes are in the Cluster, turn off one server and insure its resources migrate to another node.
Turn the server back on and verify it rejoins the Cluster.
Verify that the node resources can be migrated back to it. |
| 1104 |
Cluster SAN Connection Failure Test |
While all nodes are in the Cluster, disconnect the SAN connection from one of the nodes in the Cluster
Ensure that the Cluster node fails and that all its resources migrate to another node in the Cluster.
Plug the SAN connection back into the failed node and VERIFY you can re-join the Cluster with that node.
Ensure you can migrate the node resources back to it. |
| 1105 |
Cluster Node Network Failure Test |
While all nodes are in the cluster, disconnect the public LAN connection from the active node to force it to fail over to the other node.
Verify the cluster resources migrate over to the other cluster node.
Reconnect the downed node's private LAN connection and restart the server.
Make sure it successfully re-joins the cluster and can the cluster resources fail back to it. |
| 1106 |
MS Exchange 2003 Cluster with LoadSim test |
Configure an Exchange cluster server to support running LoadSim.
On the client system run LoadSim with 50 users for 4 hours. |
| 1200 |
MS Exchange 2003 LoadSim Performance Test |
| 1201 |
LoadSim performance test |
Setup 3 additional mailbox stores on the Exchange 2003 server. These mailboxes will use the RAID10 Logical Volumes from the FS4500.
Create a topology of 1250 simulated users using LoadSim.
Configure the test parameters so that the test will run for 4 daytime hours.
Initialize the topology from each of the Client Load Generator systems.
Verify that 0 errors occur during initialization prior to running the test.
After performance run has completed, execute the lslog command to obtain performance results from each Client Load Generator system. |
The combination of the Adaptec SANbloc Fibre Channel RAID and Emulex LP10000 Fibre Channel HBAs and Vixel InSpeed 9050FF Fabric Switch provide a flexible, reliable and easily managed solution for deploying high-availability SANs using a variety of configurations. Topologies can be designed to meet specific availability or architecture requirements without compromising ease of management or reliability. | |