Question Design Server feeding multiple thin clients

[Christian @ 2CT Media]

No problems, just cost involved.

 

[WildWestDesigns]

No problems, just cost involved.

After all this starting to get squeamish over the licensing?

Should be looking at bulk versus individual if starting to go this route.

 

[Christian @ 2CT Media]

Not squeamish at all, what gave you that impression? I just stated that those prices didn't include licensing which is a variable based on each Profile needed.

 

[WildWestDesigns]

Not squeamish at all, what gave you that impression? I just stated that those prices didn't include licensing which is a variable based on each Profile needed.

Just the phrasing. "And then there is licensing".

In my experience saying "and then there is [whatever]", is usually followed by something negative/problem/what have you. Now granted that is just in my experience. I would imagine that shoresigns was probably thinking along the same lines since it was asked "what problems are you expecting with licensing?"

In my warped mind, it's different then saying "..but that doesn't include licensing cost." But I am very much a semantics person (cue eyerolling here) and I'm sure I was adding inflection that wasn't there.

 

[ColorCrest]

So meeting went well, they are going to spec out a a few options but we are sub $65,000 for everything we discussed as of now.

They recommend Hyper-V to start since we are under 25 concurrent users. But a 64 core 128 thread server with 1tb memory and 10tb of nvme is around $58,000, 10g thin clients are $880/ea and then there is licensing

I'm wondering about any human administration costs of time for such a setup including the admin of apps. Minimal so much as it's not even worth the trouble to calculate, or minutes a day, week, month, etc.?

 

[Christian @ 2CT Media]

No worries, I just didn't add in that cost since they didn't have all of it.

 

[Christian @ 2CT Media]

I'm wondering about any human administration costs of time for such a setup including the admin of apps. Minimal so much as it's not even worth the trouble to calculate, or minutes a day, week, month, etc.?

$300/mo for remote maintenance, monitoring, updating and tech support. We do all hardware / on site physical networking.

 

[ikarasu]

Does it still seem worth it to you?

$65,000 for under 25 users... you could buy every a top of the line PC / video card for cheaper.

And the $300 a month maintenance... thats $3600 a year... Thats buying 2-3 brand new PC's per year, just in maintenance costs alone. granted you're probably already paying someone for support...so maybe itll end up being cheaper.

But, imo... I'd be worried about a crash or something that knocks everyone down for even a day.

 

[Christian @ 2CT Media]

Its still very much worth it in my mind. Especially considering auto-allocation of resources based on demand. So if this case if a designer is consuming their allocated resources based on a job demand, the system will automatically provision more resources to help them power through projects.

 

[WildWestDesigns]

Its still very much worth it in my mind. Especially considering auto-allocation of resources based on demand. So if this case if a designer is consuming their allocated resources based on a job demand, the system will automatically provision more resources to help them power through projects.

I always just get more computer then I can envision needing and just run with it. Better to have it and not need it, then to need it and not have it.

 

[Christian @ 2CT Media]

Hard to do now unless I switch to getting the next class up which dramatically increases costs meaning a the server is even better of value.

 

[ColorCrest]

Its still very much worth it in my mind. Especially considering auto-allocation of resources based on demand. So if this case if a designer is consuming their allocated resources based on a job demand, the system will automatically provision more resources to help them power through projects.

Just as a perspective...

Simply adding more workstations and / or disks (local RAID for I/O or networked for resources) per operator is a very typical solution.

 

[Christian @ 2CT Media]

Add more workstations, why? That is a harder solution to manage yet alone a person can only physically work at one station at a time.

Disks are a thing of the past, we already use NVME raid arrays. But compute scale and ram scale is limited on non server based architecture. Moving to HLE Workstation class hardware is so close to server grade it's lost value.

 

[ikarasu]

There is a point where you start to see diminishing returns though.

If everyone is allocated 10 2ghz cores... Is they're maxing them out working on a file, just because their cores get doubled doesn't mean the performance is doubled.

For instance... I believe photoshop was limited to under 8 cores. And it only uses those 8 cores if you're doing a specific 8 things that can utilize them all at once. (Going off of memory, it may have changed recently also). So cpu wise a higher speed 8 core would run faster than if you allocated 100 cores to it at a slower speed.

Illustrator is even more limited.

Where as gpu plays a pretty big roll in photoshop... Not too much in illustrator. Are you going to have a bunch of high end gpus to utilize?

Also... YouRe an onyx user right? Onyx limits you to cores via license.... You can buy more cores for faster ripping, but it's expensive.


I think that's my main issue. You can build a pc that'll max out the performance you can get on all the software our companies use for decently cheap.... Where as you can buy your central server.... And throw as much cores and ram at a system as possible, but you gain negligible performance over a system you built specifically for the task.

I know you're doing it to simplify things, and in that regard I think it will in certain esys. Heck.. Id like to do it for the coolness factor alone. I just don't think it's practical in the performance sense.. I'm fairly sure I could build a computer for under your per user rate, that will outperform A user on the system no matter what resources you dedicate to it.

I do hope If you guys decide to go that way it works out well for you! I just think it'll be a "this sounded like a good idea at the time" purchase.

Plus, you know as soon as you drop 65k on a system they're going to come out with quantum computing and you'll have blown all your money

 

[ColorCrest]

That is a harder solution to manage

How so, exactly, in your experience?

yet alone a person can only physically work at one station at a time.

An operator may use one workstation for design, say and another for RIP and even another for work admin / correspondence, etc., all simultaneously. It has been common for many years in my experience.

Disks are a thing of the past,

I used the term "disk" generically.

But compute scale and ram scale is limited on non server based architecture.

Do you know for certain that apps such as Photoshop will work only in RAM without hitting a disk in normal operation? A RIP? Any app?

Moving to HLE Workstation class hardware is so close to server grade it's lost value.

Since I don't know what this is, it's a good time for me to wish you good luck on your quest. Truly.

 

[Texas_Signmaker]

I'm still on the fence. I agree with ikarasu 100%

Everyone running off one server? Problem with the server means everyone is down.

$58,000 is a good chunk of $.

That's a big change and a daring leap, let us know if its beneficial

 

[Christian @ 2CT Media]

So everyone running off 1 server is a risk, but we are looking at trying to mitigate that by using NUCs for critical terminals (Design, RIP) instead of pure thin clients.

Yes there is the point of diminished returns, if you can only scale 1 performance aspect. We would have Core, RAM, and File IO scaleable. Which would help Photoshop (No cap in 2019+), Illustrator, and Onyx 19 (uses load balancing beyond your singular rip cores).

ColorCrest HLE is High Level Enthusiast ie Workstation Class hardware without going full server grade (Intel 10980xe, AMD Threadripper 3, Intel Xeon Ex series, etc)

Btw I like this challenged thinking, you guys are bringing up things I didn't first think about. This is an exercise for us to think outside the box to solve a few of our problems. My biggest problem is I can't be tech support any more... To much growth to help steer.

 

[WildWestDesigns]

Yes there is the point of diminished returns, if you can only scale 1 performance aspect. We would have Core, RAM, and File IO scaleable. Which would help Photoshop (No cap in 2019+), Illustrator, and Onyx 19 (uses load balancing beyond your singular rip cores).

I cannot imagine that there still isn't going to be software bottlenecks in there. This is definitely not the typical use case for their customers, so for them to have their software code based optimized for such deployment, I can't see it. Typically those that have sophisticated systems like this typically have custom pipelines that also help with the main system.

ColorCrest HLE is High Level Enthusiast ie Workstation Class hardware without going full server grade (Intel 10980xe, AMD Threadripper 3, Intel Xeon Ex series, etc)

This is where I am at for my main office/designing rig. Couldn't be happier with it.

 

[Christian @ 2CT Media]

I cannot imagine that there still isn't going to be software bottlenecks in there. This is definitely not the typical use case for their customers, so for them to have their software code based optimized for such deployment, I can't see it. Typically those that have sophisticated systems like this typically have custom pipelines that also help with the main system.



This is where I am at for my main office/designing rig. Couldn't be happier with it.

Software is always the problem, and those problems will rear it's ugly head on a individual system as well.

We have talked about building all of our future machines as a server class workstation, but cost goes up dramatically which lead my thoughts down this path.

 

[WildWestDesigns]

Software is always the problem, and those problems will rear it's ugly head on a individual system as well.

On a system like your are proposing, those odds are significantly higher and it will be all up to y'all to handle support on this when it comes to a system like this. I don't even know if service tickets/bug reports will be accepted as easily with a setup like this going thru Adobe.

We have talked about building all of our future machines as a server class workstation, but cost goes up dramatically which lead my thoughts down this path.

Where you can get a little bit of an advantage is that when it's time to upgrade. Businesses are on a rotation for their hardware, regardless if it's played out or not (and on this class of hardware there is always a good bit of life left in those components when they upgrade) so the price of scaling up is significantly cheaper compared to traditional desktop components.