Solved: SAN Multi-pathing best practices?

steven.pw.lau · ‎09-15-2010

Hi,

I have a few questions on SAN Multi-pathing using UCS.

I will itemized each questions.

Since there's no failover option for vHBA, we would need to use the underlying MPIO features of the O/S (e.g. ESX/ESXi)? I believe MPIO for ESX/ESXi is enabled by default but for Win2008, we'll need to enable it?
My Clariion array has 2 Storage Processors (SP-A and SP-B). Each SP has 2 ports. I these 2 ports meshed to two MDS with one MDS connecting to one Fabric Interconnect respectively. Should I be creating two SAN fabrics for these 2 paths (2 different VSANs)?
What is the convergence time for the host to use another path when failure occurs? (few seconds or a minute or so?) This is related to question 2 if for example, the convergence time were to take around 1 min+, then it makes no differences having 2 different SAN fabrics as compared to 1 because we'll be experiencing downtime anyway.
I'd have one boot LUN (LUN 0) and LUN 1 as the datastore LUN for each host. Since, my boot policy in UCS (for blade/host 1) will have vHBA1 (Primary -> SP-A/port-0, Secondary -> SP-B/port-0) and vHBA2 (Primary -> SP-A/port-1, Secondary -> SP-B/port-1), should I give the ownership for all the LUNs for host1 to SP-A?
- Will I have problems if I only put the boot LUN (LUN 0) in SP-A and the datastore LUN (LUN 1) in SP-B? For this case, I'm not sure how the path selection will work as the writeable path is always on SP-A and only if path failure occurs or SP failures, it will only use the secondary path. Would appreciate some advice on this

Thanks.

Robert Burns · ‎09-15-2010

Answers inline. Others can chime in if they have addtional comments or suggestions.

Cheers,

Robert

steven.pw.lau wrote:

Hi,

I have a few questions on SAN Multi-pathing using UCS.

I will itemized each questions.

Since there's no failover option for vHBA, we would need to use the underlying MPIO features of the O/S (e.g. ESX/ESXi)? I believe MPIO for ESX/ESXi is enabled by default but for Win2008, we'll need to enable it?

[Robert] - Correct. There's a native multipathing pluging (NMP) in VMware, but for Windows you'll need additional config/software.

2. My Clariion array has 2 Storage Processors (SP-A and SP-B). Each SP has 2 ports. I have these 2 ports meshed to two MDS with one MDS connecting to one Fabric Interconnect respectively. Should I be creating two SAN fabrics for these 2 paths (2 different VSANs)?
[Robert] - Yes. Connect one VSAN on each FI. Your FC uplinks will each connect to an upstream MDS (assuming each MDS is not connected to each FI). You will still have 4 storage paths available to each host using two vHBAs.

3. What is the convergence time for the host to use another path when failure occurs? (few seconds or a minute or so?) This is related

to question 2 if for example, the convergence time were to take around 1 min+, then it makes no differences having 2 different SAN fabrics as compared to 1 because we'll be experiencing downtime anyway.
[Robert] - Failover for FC traffic is handled by the host, not UCS. It's up to the host OS to decide how & when to redirect traffic to the alternate path. Normally FC failover is pretty seamless. FC traffic can't tolerate any lengthy of delays. As soon as the SCSI command times out, it will jump to the other path. Again, this will vary from OS to OS, but generally its seamless to the application/OS.

4. I'd like have one boot LUN (LUN 0) and LUN 1 as the datastore LUN for each host. Since, my boot policy in UCS (for blade/host 1) will have vHBA1 (Primary -> SP-A/port-0, Secondary -> SP-B/port-0) and vHBA2 (Primary -> SP-A/port-1, Secondary -> SP-B/port-1), should I give the ownership for all the LUNs for host1 to SP-A?
[Robert] Yes. Your primary boot target must be the "owner" of the LUN (and should be the "default" owner for that matter). Only when SPA fails, will the LUN tresspass to SP-B and the secondary boot path will succeed during server boot up. Keep in mind, this is only for booting. Once its up & running, the OS handles failover activities.

5. Will I have problems if I only put the boot LUN (LUN 0) in SP-A and the datastore LUN (LUN 1) in SP-B? For this case, I'm not sure how the path selection will work as the writeable path is always on SP-A and only if path failure occurs or SP failures, it will only use the secondary path. Would appreciate some advice on this

[Robert] Boot policy & data path selection are two different things. If your boot LUN is targeting SPA, it will likely maintain that path as the OS loads. Once the OS loads you can manipulate the path for any LUN. If desired you could set the "Active" path of your Datastore LUN within vCenter to be the other SP (SPB in your case) by modifying the path selection policy and setting it to "Fixed" PSP. The default Path Selection Policy is MRU (Most recently Used). This is normally adequate, but if you want to manually load balance your traffic the Fixed PSP policy is the way to go.

Thanks.

View solution in original post

Robert Burns · ‎09-15-2010

Answers inline. Others can chime in if they have addtional comments or suggestions.

Cheers,

Robert

steven.pw.lau wrote:

Hi,

I have a few questions on SAN Multi-pathing using UCS.

I will itemized each questions.

Since there's no failover option for vHBA, we would need to use the underlying MPIO features of the O/S (e.g. ESX/ESXi)? I believe MPIO for ESX/ESXi is enabled by default but for Win2008, we'll need to enable it?

[Robert] - Correct. There's a native multipathing pluging (NMP) in VMware, but for Windows you'll need additional config/software.

2. My Clariion array has 2 Storage Processors (SP-A and SP-B). Each SP has 2 ports. I have these 2 ports meshed to two MDS with one MDS connecting to one Fabric Interconnect respectively. Should I be creating two SAN fabrics for these 2 paths (2 different VSANs)?
[Robert] - Yes. Connect one VSAN on each FI. Your FC uplinks will each connect to an upstream MDS (assuming each MDS is not connected to each FI). You will still have 4 storage paths available to each host using two vHBAs.

3. What is the convergence time for the host to use another path when failure occurs? (few seconds or a minute or so?) This is related

to question 2 if for example, the convergence time were to take around 1 min+, then it makes no differences having 2 different SAN fabrics as compared to 1 because we'll be experiencing downtime anyway.
[Robert] - Failover for FC traffic is handled by the host, not UCS. It's up to the host OS to decide how & when to redirect traffic to the alternate path. Normally FC failover is pretty seamless. FC traffic can't tolerate any lengthy of delays. As soon as the SCSI command times out, it will jump to the other path. Again, this will vary from OS to OS, but generally its seamless to the application/OS.

4. I'd like have one boot LUN (LUN 0) and LUN 1 as the datastore LUN for each host. Since, my boot policy in UCS (for blade/host 1) will have vHBA1 (Primary -> SP-A/port-0, Secondary -> SP-B/port-0) and vHBA2 (Primary -> SP-A/port-1, Secondary -> SP-B/port-1), should I give the ownership for all the LUNs for host1 to SP-A?
[Robert] Yes. Your primary boot target must be the "owner" of the LUN (and should be the "default" owner for that matter). Only when SPA fails, will the LUN tresspass to SP-B and the secondary boot path will succeed during server boot up. Keep in mind, this is only for booting. Once its up & running, the OS handles failover activities.

5. Will I have problems if I only put the boot LUN (LUN 0) in SP-A and the datastore LUN (LUN 1) in SP-B? For this case, I'm not sure how the path selection will work as the writeable path is always on SP-A and only if path failure occurs or SP failures, it will only use the secondary path. Would appreciate some advice on this

[Robert] Boot policy & data path selection are two different things. If your boot LUN is targeting SPA, it will likely maintain that path as the OS loads. Once the OS loads you can manipulate the path for any LUN. If desired you could set the "Active" path of your Datastore LUN within vCenter to be the other SP (SPB in your case) by modifying the path selection policy and setting it to "Fixed" PSP. The default Path Selection Policy is MRU (Most recently Used). This is normally adequate, but if you want to manually load balance your traffic the Fixed PSP policy is the way to go.

Thanks.

JEFFREY SESSLER · ‎09-27-2010

If you're EMC is running a current FLARE with ALUA support, set your fail-over mode to 4 (ALUA). This will allow boot from SAN no matter which SP is owner including when in fail condition without the need to trespass luns. Path selection will prefer the owner SP, so access to LUN 1 will use SP-B paths with LUN 0 using SP-A paths. If you want optimized path failover and load balancing, you'll want to purchase and install PowerPath (Powerpath/VE for VMware).