Windows 10 or Windows Server for high speed file share?

[jstroud17]

I run a production company, we currently have an 80TB RAID array (10x 8TB drives with a MegaRAID controller) shared across a 10GB ethernet network... it's been working great until the last month (was installed in April 2018).

 

The 'server' has Windows 10 pro installed with a max of 4 network users editing or accessing content over the network (Netgear 10Gbe switch) via 3 macs / 1 PC.

 

I didn't build the system and unfortunately the people who did apparently didn't install enough cooling as we had an issue last month where under heavy usage the HDD temps were rising north of 60*C. This caused severe performance issues, files weren't copying / verifying correctly and it was all a bit of a mess. I installed more fans, backed everything up, all has been fine since then.

 

However, today the server had a major meltdown where during a fairly basic copy of about 60Gb of data, from one of the networked Macs, the copy failed and the network interface on the server crashed (it just disconnected and all the lights on the card went out) taking a restart to get back up and running. This problem re-occurred every time the data was copied, yet copying to the local drive on the mac was fine.

 

The question is will I see better, more stable results by installing Windows Server 2019 rather than running Windows 10? And perhaps more error reporting too?

 

I have a feeling the latest networking error has come from using a teamed connection through the Intel 10GB NIC I have installed so I'm going to reverse this, however I know this is a feature I could use from Windows Server if we had it rather than using the Intel utility in Windows 10.

 

Thoughts appreciated

[jake9000]

From a business perspective, Server offers many concrete benefits: 

 

Spoiler

 

DFS Namespaces - Lets you group shared folders on different servers under a single logical structure. So users can type in something like \\fileshare\department\mycoolfolder and namespaces will automatically take them to the correct server, even if you move that share to other servers.

 

DFS Replication: synchronize folders on multiple servers across a local or wide network. 

 

FSRM: Set up file management tasks, folder quotas, storage reports, etc. Also some powerful anti-ransomware automation to be had

 

Data deduplication: Single file stored multiple times by different users? With this it is only stored on-disk once. Helps conserve disk usage. 

 

VSS Agent services: volume shadow copies. Makes recovering lost files /  undeleting stuff a snap. Provides a nice interface to various other backup tools like Veeam. 

 

 

 

From a technical perspective, you need to address the root cause of the issue, which is probably some kind of firmware or driver issue with your NIC. 

[leadeater]

43 minutes ago, jstroud17 said:

The question is will I see better, more stable results by installing Windows Server 2019 rather than running Windows 10? And perhaps more error reporting too?

 

I have a feeling the latest networking error has come from using a teamed connection through the Intel 10GB NIC I have installed so I'm going to reverse this, however I know this is a feature I could use from Windows Server if we had it rather than using the Intel utility in Windows 10.

Performance wise there will be no difference, having more official teaming support may help but unless you actually need more than 10Gb and your network is capable to balance the traffic correctly and actually have the bandwidth at all points in the chain then you won't see any difference there either.

 

Check the temperature of the NIC, most 10Gb cards are designed for servers with high airflow and get really hot, a sustained file copy can potentially overheat the controller chip causing problems like you are getting.

 

Agree with taking away the teaming, test at the most basic configuration as possible to reduce variables that could be a problem.

[tech.guru]

you will if you implement the technologies built into windows server,

         there is solutions using scale out file servers. they use Microsoft cluster with cheap local storage and you can cluster and replicate using block based. that would be perfect for file servers for your video editing where you are working on large files.

 

the newer client operating system (vista and above) support transparent failover. if one went down the other file server could take over clients would just notice a brief pause but otherwise be uninterrupted. you can also use the built in storage replica to keep both local disks in sync

 

its very compelling and would provide better performance than a single server and eliminate failures you mention causing an outage .large data sets such as videos are a perfect for an active-active file cluster that can provide excellent performance and reliability without buying an expensive SAN.

On 8/8/2019 at 7:05 PM, leadeater said:

Performance wise there will be no difference, having more official teaming support may help but unless you actually need more than 10Gb and your network is capable to balance the traffic correctly and actually have the bandwidth at all points in the chain then you won't see any difference there either.

 

Check the temperature of the NIC, most 10Gb cards are designed for servers with high airflow and get really hot, a sustained file copy can potentially overheat the controller chip causing problems like you are getting.

 

Agree with taking away the teaming, test at the most basic configuration as possible to reduce variables that could be a problem.

 

[leadeater]

1 hour ago, tech.guru said:

you will if you implement the technologies built into windows server,

         there is solutions using scale out file servers. they use Microsoft cluster with cheap local storage and you can cluster and replicate using block based. that would be perfect for file servers for your video editing where you are working on large files.

 

the newer client operating system (vista and above) support transparent failover. if one went down the other file server could take over clients would just notice a brief pause but otherwise be uninterrupted. you can also use the built in storage replica to keep both local disks in sync

 

its very compelling and would provide better performance than a single server and eliminate failures you mention causing an outage .large data sets such as videos are a perfect for an active-active file cluster that can provide excellent performance and reliability without buying an expensive SAN.

Cluster Shared Volumes in Windows Server relies on shared block storage between the cluster nodes so will need a SAN for this. To use local storage in a scale out fashion like you describe you need to use Storage Spaces Direct which is a Windows Server Datacenter only feature which makes it very expensive, out of reach of most if not all smaller businesses.

 

While you can use SMB Multichannel to load balance and give some amount of HA it only works for SMB and not other protocols, however in regards to performance unless the backend network actually has multiple paths having an aggregate performance higher than a single port on the server you won't gain any extra bandwidth.

 

It also makes no difference if the client bandwidth demand total is below a single port on the server as well. That said I, every should, implement teaming to two different switches for servers in either active-passive or active-active (switch stacking or virtual stacking required).

 

Edit:

Storage Replica is also really a Datacenter only feature, explicitly was in Server 2016 but the restrictions in Server 2019 make it rather un-useful.

Quote

Standard Edition, has the following limitations:

• You must use Windows Server 2019 or later
• Storage Replica replicates a single volume instead of an unlimited number of volumes.
• Volumes can have a size of up to 2 TB instead of an unlimited size.

 

[tech.guru]

32 minutes ago, leadeater said:

Cluster Shared Volumes in Windows Server relies on shared block storage between the cluster nodes so will need a SAN for this. To use local storage in a scale out fashion like you describe you need to use Storage Spaces Direct which is a Windows Server Datacenter only feature which makes it very expensive, out of reach of most if not all smaller businesses.

 

While you can use SMB Multichannel to load balance and give some amount of HA it only works for SMB and not other protocols, however in regards to performance unless the backend network actually has multiple paths having an aggregate performance higher than a single port on the server you won't gain any extra bandwidth.

 

It also makes no difference if the client bandwidth demand total is below a single port on the server as well. That said I, every should, implement teaming to two different switches for servers in either active-passive or active-active (switch stacking or virtual stacking required).

 

Edit:

Storage Replica is also really a Datacenter only feature, explicitly was in Server 2016 but the restrictions in Server 2019 make it rather un-useful.

 

expensive is relative compare to other competing solutions. 

look at vmware virtual san and other solutions.

 

the user mentioned 80 tb worth of data, thats a large amount of data not to have a redundant solution especially when he mentions the concerns around failure that has already happened.

 

i stand by what i said either go with a vsan or physical san solution based on the tolerance of risk. Microsoft was suggested based on his comments about going to windows server.

[leadeater]

1 hour ago, tech.guru said:

expensive is relative compare to other competing solutions.

For the price of Datacenter licenses you can purchase a complete NAS with 100TB of storage, or a Windows Server with Standard license and about 80TB of storage. Increasing $/GB should only be done when there is a clear and tangible benefit to doing so and for a storage server for 4 computers isn't really that. It gets even worse if you want to introduce scale out storage on such a small deployment, spending way too much on server systems as opposed to spending it on storage capacity.

 

1 hour ago, tech.guru said:

the user mentioned 80 tb worth of data, thats a large amount of data not to have a redundant solution especially when he mentions the concerns around failure that has already happened.

80TB isn't actually much, not for video editing or larger businesses. There are more cost effective ways to get redundancy in a storage platform however redundancy is not a replacement for backups. I maintain on behalf of our internal departments about 20 Windows servers with between 120TB to 240TB of storage each, while I'd prefer them not to buy these servers they have purchase authority to do so and our enterprise storage solution that we can offer them is too expensive, Netapp with just over 2PB. We're currently rolling out a 'small' 800TB Ceph cluster to address that lower cost storage demand and do it in a more resilient manor than just Windows Servers with DAS, that said we've never had any data loss or any servers fail resulting in an outage of service in the last 10+ years, doesn't mean it's a good way of doing things though but at least they are all backed up.

 

As to the issue that is being asked about that is a stability problem, solve that and there is no reason to go and implement a complex resiliency solution when backups is more appropriate. Like RAID the resiliency isn't a replacement for backups and you shouldn't solve a system stability issue by adding more things in to the equation, solve the root cause of the issue first.

 

1 hour ago, tech.guru said:

i stand by what i said either go with a vsan or physical san solution based on the tolerance of risk. Microsoft was suggested based on his comments about going to windows server.

Yes a SAS attached storage array with dual controllers would be a much better solution but that's a long term solution that won't resolve the immediate problem.