New Physical SIMS Server - thoughts on specifications?

[enbiggen]

Hi,

I'm trying to unify three separate quotes with different specifications into one specification to "rule them all" so that we can compare like-with-like.

Our current server (HP ProLiant DL380 G7) has been fine specification-wise, but it's old, and we need to migrate to a new operating system and version of SQL:

• Smart Array P410i with 8x 137GB SAS 15k 6Gbps drives in a RAID 10 configuration.
  
• 2xIntel X5650 processors at 2.67GHz
  
• 24.0GB of RAM
  
• 4x1GB NIC (only one port used)
  
• 2x PSU
  

Our SIMS database is about 12GB in size, and FMS is 0.5GB in size. Storage is fine, using less than half of the total space available.

From the specifications we've got back, the main split seems to be in the drives, and to a lesser extent, the processors.

• Drives (all including a P408i storage controller with 2MB cache and battery):
  

 

• One has two 2x 960GB hot-swap SAS solid state drives (read intensive) - which means that we'd operate in Raid 1 mode, as there's only two drives?
  
• One has 8x 300GB SAS 12G enterprise mechanical disks.
  
• One has 8x 240GB SATA 6G Read Intensive SSD drives.
  

 

• Processors:
  

 

• 2x Intel Xeon 4208
  
• 2x Intel Xeon 4210
  
• 2x Intel Xeon 4215
  

 

• Memory = 2x 32GB
  
• NIC
  

 

• 4x 1GB Nic
  
• 2x 10GB Nic
  

 

• 2x PSUs
  

Should we go for SSD over mechanical? I know solid state is faster, but my mind thinks back to the horror stories of them all failing at the same time (I'm sure they've improved). Is striping still necessary for the best performance, in which case we'd 4x multiple smaller disks, or are they so fast/reliable that you could go with a larger SSD mirrored?

Should we go with 2x 10GB, even though our infrastructure is still 1GB, or use 4x 1GB and upgrade later if our infrastructure is upgraded?

Processor-wise, are they really all over kill for what we need, and we could go with the 2x Intel Xeon 4208? (I know the clock speed is less than our current processors, but they have a lot more cache and memory throughput.)

 

We're using SQL standard server with separate client CALs model.

 

What do you think, which bits should I standardize on when I go back to ask for requotes? I won't forget to specify a 5 Year NBD warranty this time round - one company read my mind and included it thought, the others went with the default 3 Years warranty.

 

Hopefully I've thought of everything - making big (for me) purchases like this always gives me angst.

 

Thanks,

Enbiggen.

[3s-gtech]

I can't help but think your approach is a bit dated - of course with SIMS onsite it always will be, but I strongly recommend you don't run it on bare metal.

 

Fast disks are always going to be a benefit, but SQL will load as much of the DB into memory as possible and leave it there. As long as the OS + services + SIMS DB fit in RAM it'll be pretty quick. If it has to swap on mechanical disks, it'll crawl.

 

RAID and your SSDs will about reliability rather than speed, to me. RAID 10 is a good choice as you have some fault tolerance - can you hot swap the drives though? I'd want the OS on a separate volume for wear reduction.

 

NICs - you can always add a card. A lot of the newer ones have an SFP+ slot anyway. 4 x 1Gb should still be sufficient.

[ITGURU]

I can't help but think your approach is a bit dated - of course with SIMS onsite it always will be, but I strongly recommend you don't run it on bare metal.

 

Fast disks are always going to be a benefit, but SQL will load as much of the DB into memory as possible and leave it there. As long as the OS + services + SIMS DB fit in RAM it'll be pretty quick. If it has to swap on mechanical disks, it'll crawl.

 

RAID and your SSDs will about reliability rather than speed, to me. RAID 10 is a good choice as you have some fault tolerance - can you hot swap the drives though? I'd want the OS on a separate volume for wear reduction.

 

NICs - you can always add a card. A lot of the newer ones have an SFP+ slot anyway. 4 x 1Gb should still be sufficient.

 

SIMS will always be slower on virtual. I have 3 sites, 2 run virtual (hyper-V and ESXi), and 1 physical. The physical does everything quicker and companies always tell us that the physical server is quick! Physical server OS RAID 1, Database RAID 10, and the virtual servers all on RAID 10 volume.

[enbiggen]

I can't help but think your approach is a bit dated - of course with SIMS onsite it always will be, but I strongly recommend you don't run it on bare metal.

We can't afford to go with a proper virtualized setup at the moment, so we're going with the physical server. Ideally we would be virtualised, and I love the thought of being able to migrate virtual machines between different physical hosts.

 

Fast disks are always going to be a benefit, but SQL will load as much of the DB into memory as possible and leave it there. As long as the OS + services + SIMS DB fit in RAM it'll be pretty quick. If it has to swap on mechanical disks, it'll crawl.

 

Even with the existing server the entire SIMS database can be held in memory. With 2x32GB there's going to be no issues.

 

RAID and your SSDs will about reliability rather than speed, to me. RAID 10 is a good choice as you have some fault tolerance - can you hot swap the drives though? I'd want the OS on a separate volume for wear reduction.

 

All of the drives are hot-swappable (within the limits of the RAID configuration obviously). This would tend to favour the 8x smaller drives approach, as then I could have 4x drives for the OS, and 4x drives for data? (Sorry if that's not what you meant.) If they were SSDs though, could I get away with mirroring, rather than mirroring and spanning, or does spanning (by nature of spreading the data across disks) mean that there's reduced "wear" on the individual disks making it more reliable?

 

Thanks,

Enbiggen.

 

 

[enbiggen]

SIMS will always be slower on virtual. I have 3 sites, 2 run virtual (hyper-V and ESXi), and 1 physical. The physical does everything quicker and companies always tell us that the physical server is quick! Physical server OS RAID 1, Database RAID 10, and the virtual servers all on RAID 10 volume.

True, but I took it as more of a suggestion to think about virtualisation as a whole, given that we're buying new servers - that said, we can't afford to go down that route at the moment.

[3s-gtech]

It will be slower, due to overheads. However, the counter to that will be in terms of redundancy and failover, easy movement, and backups against possible ransomware. Having goosed my own SIMS server host recently, and SIMS being virtual on it, I had it back up and running within an hour or so. Had it been physical, I'd have had a really bad day.

 

In reply to your newer post - you can virtualise it at no extra cost, as Windows Server can host a Hyper-V VM within the license. Something to look at.

 

I'm not enough of an expert on RAID to be sure that the combination of mirroring and striping would reduce wear - I'd guess not, as the striped changes are then mirrored anyway. It's academic on two drives.

Edited November 29, 2021 by 3s-gtech

[synaesthesia]

Whatever you do, make sure you include the cost of a new server in any renewal costings - suprised anyone would be sticking with SIMS long enough to warrant replacing a physical server, rather than going with something modern and suitable for the task at hand.

[Jawloms]

We can't afford to go with a proper virtualized setup at the moment, so we're going with the physical server. Ideally we would be virtualised, and I love the thought of being able to migrate virtual machines between different physical hosts.

 

If you're installing Windows Server to run SIMS, install Hyper-V on it, and then just run one VM on it which is your SIMS server. You then are running SIMS in a VM and it cost you no more. Much easier to backup and move about if necessary.

[enbiggen]

Whatever you do, make sure you include the cost of a new server in any renewal costings - surprised anyone would be sticking with SIMS long enough to warrant replacing a physical server, rather than going with something modern and suitable for the task at hand.

Weirdly we seem to quite like SIMS here - even if we don't like some of their tactics about contract length. We do have some additional software that integrates with it, such as Class Charts, so that might take some of the rough edges off it? It's also possible that we don't know what we're missing :-)

[jmak]

If you're installing Windows Server to run SIMS, install Hyper-V on it, and then just run one VM on it which is your SIMS server. You then are running SIMS in a VM and it cost you no more. Much easier to backup and move about if necessary.

^^This

You don't need to be into cluster aware failover and similar complexities to benefit from virtualization. If you backup complete machines you can just connect the backup to a different host (another machine you've got that runs something else, even if it compromises performance or an old machine or a bare metal warranty replacement) and you can be up and running in minutes.

[dhicks]

If you're installing Windows Server to run SIMS, install Hyper-V on it, and then just run one VM on it which is your SIMS server.

 

Just to add my agreement to this strategy, also look at Windows Storage Spaces - rather than use a RAID array, let Windows handle how it deals with using multiple disks for redundancy / performance. This might actually be the cheaper option as you don't need a RAID controller (although maybe a small one just for the Windows boot volume), just an HBA card. If you want to have SSD performance but the capacity of larger disks then you can let Storage Spaces handle that for you, too, via tiered storage.

[enbiggen]

^^This

You don't need to be into cluster aware failover and similar complexities to benefit from virtualization. If you backup complete machines you can just connect the backup to a different host (another machine you've got that runs something else, even if it compromises performance or an old machine or a bare metal warranty replacement) and you can be up and running in minutes.

Just back from lunch duty, where I was thinking about the advantages of this approach: Am I right in thinking that this is done as your virtual host is a "generic", so you can move it to a completely different physically different server without worrying about drivers etc, as they all conform to the same "generic" virtual machine? I'm definitely veering towards it in my head now.

 

"decoupling" - that's the word I was looking for - we decouple the sims server from the physical hardware that it's running on to make it easier to move somewhere else in the event of hardware failure, regardless of it being the only virtual machine on the host.

 

Thanks,

Enbiggen

Edited November 29, 2021 by enbiggen

[3s-gtech]

That’s certainly an aspect of it - ‘generic’ VMs that will run on different hardware. There can be some fun when swapping between Intel and AMD hosts but it still works.

[enbiggen]

^^This

You don't need to be into cluster aware failover and similar complexities to benefit from virtualization. If you backup complete machines you can just connect the backup to a different host (another machine you've got that runs something else, even if it compromises performance or an old machine or a bare metal warranty replacement) and you can be up and running in minutes.

Just reading backwards from the end - yes, I see this now - thanks. Very useful.

[Rob_D]

Just to add my agreement to this strategy, also look at Windows Storage Spaces - rather than use a RAID array, let Windows handle how it deals with using multiple disks for redundancy / performance. This might actually be the cheaper option as you don't need a RAID controller (although maybe a small one just for the Windows boot volume), just an HBA card. If you want to have SSD performance but the capacity of larger disks then you can let Storage Spaces handle that for you, too, via tiered storage.

Just be careful about using tiered storage on anything other than a virtual host. It doesn't (or didn't a few years back when we tried it) handle lots of smaller files very well.

[psydii]

It’s speed of the cores and the performance of the storage where tempdb is stored that seems to make the biggest difference once you have more than enough ram and optimised indices

 

At least that what the conclusion was from a MS PSS Engagement a few years ago.

[Guest ZFarnworth]

I can’t lie but I do like sims better than bromcom. I like the layout of sims much better weirdly [emoji28]

[dhicks]

It doesn't (or didn't a few years back when we tried it) handle lots of smaller files very well.

 

Does it make any difference what kind of file system you have on top of your Storage Spaces volume? I think our main shared folders are served from a ReFS file system on top of the Storage Spaces volume, and performance is okay as far as I know.

[Rob_D]

Does it make any difference what kind of file system you have on top of your Storage Spaces volume? I think our main shared folders are served from a ReFS file system on top of the Storage Spaces volume, and performance is okay as far as I know.

Quite possibly. As I understand it, our issues were due to hitting the Storage Spaces file-operations-per-second limits.

[dhicks]

As I understand it, our issues were due to hitting the Storage Spaces file-operations-per-second limits.

 

Does having Storage Spaces certified hardware make any difference, would you know? We went for a server with NVMe storage for the fast tiered storage layer, all certified to work with Windows Server / Storage Spaces. I'm not sure if that makes any difference - you'd think hardware would simply work at the full speed it was capable of, but Windows Server might be looking out for specific features it knows it can use for better performance?

[Rob_D]

Does having Storage Spaces certified hardware make any difference, would you know? We went for a server with NVMe storage for the fast tiered storage layer, all certified to work with Windows Server / Storage Spaces. I'm not sure if that makes any difference - you'd think hardware would simply work at the full speed it was capable of, but Windows Server might be looking out for specific features it knows it can use for better performance?

I wouldn't have expected the hardware to effect our problem as it was due to us hitting a documented limitation in the system. Although I can't find any mention of it now, so maybe MS has patched out that particular problem, as I said it's been a couple of years since we virtualized the file server to fix the issue.