Server Advice


Recommended Posts

patseguin

Thanks to everyone who devoted so much time to helping me. I was getting stressed out so I just made a decision. I went with a build with 2.4TB drives in RAID6 and 2 480GB ssd drives in RAID1 for the os. I also decided on Server 2019 Standard instead of essentials just to make sure FileMaker server works without issues. I'll set up an external backup and also find a good cloud backup solution.

 

As far as deploying, I may have tons of questions. Server 2008 was extremely difficult for me to configure since I have no training or certification. I had to do a ton of googling to get it working right as a domain controller. I'm hoping with 2019 they made it a bit more user friendly. Is there maybe some way to export my configuration from my current server and import it to the new one? Another thing I thought of was keeping my older 2008 Dell server up and running for any reason. I'm thinking probably a bad idea.

Link to post
Share on other sites
Mindovermaster
4 hours ago, patseguin said:

Another thing I thought of was keeping my older 2008 Dell server up and running for any reason. I'm thinking probably a bad idea.

Could always use it as a backup. :)

Link to post
Share on other sites
sc302

Get it on the internet.  I will take it from them for you. 

Link to post
Share on other sites
DevTech
9 hours ago, patseguin said:

I went with a build with 2.4TB drives in RAID6

RAID-6 is even worse than RAID-5 for a Database due to even one more parity creation on every write! Might help with making your new server feel much like the old server!

 

But it may not matter for your specific data base as the "Teddy Bear Fans" here keep pointing out.

 

If you ever get a chance to compare running your FileMaker on your RAID-1 SSD vs your slow spinners, I would be very appreciative of the test results. Without a test there is a case to be made for 5 users being a "sleepy hollow" scenario and it won't matter what hardware it runs on. But if the Embroidery software actually does any real processing on the data, then the opposite could be true. After all the guessing, you are in a perfect spot to test it with both SSD and the "Ancient Ones" and yeah with figuring out Server 2019, you probably won't have the time to consider this, but given the mysteries involved, had to ask :)

 

 

 

Link to post
Share on other sites
Mindovermaster
40 minutes ago, DevTech said:

But it may not matter for your specific data base as the "Teddy Bear Fans" here keep pointing out.

*ding ding ding*

Link to post
Share on other sites
sc302

I ran an entire erp system on 15000 rpm disks on a fiber channel array. Went to flash...0 increase.  Oracle jd Edwards Erp system with 150+ users is going to be a bit heavier than a 5 user file maker pro db. 

  • Like 2
Link to post
Share on other sites
DevTech
2 hours ago, Mindovermaster said:

*ding ding ding*

"Ring Ring Ring"

 

I can sound like a Children's TV show too!

 

I used the word "may" but it is SOLIDLY in the EDGE CASE area! The VAST MAJORITY of Database software will have a VERY noticeable performance improvement on SSD.

 

Logically, "no difference" is a possible outcome, which is why I logically agreed it was possible. Possible is no where near the same concept as PROBABLE.

 

And find a single reference paper in a Database Journal that says RAID-5 (or 6) is good for a Database. Simply NOT professional to provision that way even if you can run around like a Bull in a China Shop and make it happen.

 

Bad science is bad science no matter what words you put on it. I can't help it if computers have been around long enough fro traditions to have built up and for VooDoo recipes to spread in a social manner. That doesn't make it science. No matter how many people start believing the moon landing never happened, it did. Period.

 

Same thing here. No matter how many people repeat some incorrect but trusted traditional I.T. "everybody knows" won't make it correct. Science is testing.

 

Which is why I politely asked when time was available that a simple test be made to see if this Embroidery Database is an edge case that does not benefit from performance or if it is more squarely in the norm of what Humanity Actually Knows About Databases

 

Link to post
Share on other sites
Mindovermaster
5 minutes ago, DevTech said:

And find a single reference paper in a Database Journal that says RAID-5 (or 6) is good for a Database. Simply NOT professional to provision that way even if you can run around like a Bull in a China Shop and make it happen.

Guess what, SC302 has done exactly that, so... What does that show?

Link to post
Share on other sites
DevTech
2 hours ago, sc302 said:

I ran an entire erp system on 15000 rpm disks on a fiber channel array. Went to flash...0 increase.  Oracle jd Edwards Erp system with 150+ users is going to be a bit heavier than a 5 user file maker pro db. 

Maybe yes. Maybe no. Depends on the code. Depends on the DB structure. Depends on what exactly the Database is being used for. Does it contain a boatload of Embroidery patterns that need to be accessed and processed? Who knows...

 

Anecdotal experience is not science. Your measurement shows for that database, there was no difference. You see that I hope? if the 5 user Embroidery DB gets tested and has the same result as yours it would still not be insightful, just a second anecdote. Standard science: Correlation is NOT causation. To prove (or have reasonable confidence in) causation, you need a controlled experiment or at least a good understanding of the breakdown (like in this case the DB layout, the sprocs etc)

 

But no matter what  this is an anecdote I'm curious about.

 

 

5 minutes ago, Mindovermaster said:

Guess what, SC302 has done exactly that, so... What does that show?

HUH? what on earth are you talking about? NO database people would say that!

 

 

EDIT: Please stop the Teddy bear stuff. Next you will say the Earth is flat. Please, enough is enough. He can make a server however he wants and I have said nothing about that, no suggestions whatsoever since his decision. So no need to pervert reality or how on earth will people learn anything?

 

Link to post
Share on other sites
Mindovermaster
5 minutes ago, DevTech said:

HUH? what on earth are you talking about? NO database people would say that!

He ran platter drives and moved to flash drives. He saw 0 difference, that should teach you that, in that scenario, it didn't matter. In OP, however, does it REALLY matter?

Link to post
Share on other sites
DevTech
1 minute ago, Mindovermaster said:

He ran platter drives and moved to flash drives. He saw 0 difference, that should teach you that, in that scenario, it didn't matter. In OP, however, does it REALLY matter?

Read my explanation again and try to understand.

 

The OP has made a decision and I have made no comments on his decision. And I will continue to not make any further comments on his configuration.

 

I have asked him if he can find the time sometime to perform the same type of test that @sc302 performed and I full expect the PROBABLE outcome to be that th SSD will show a large performance improvement. Because that is the logical result to expect.

 

If it does not improve I would just add it to a mental check list of ANOMALIES for which both computer science and reality are plentiful.

 

Would I take an ANOMALY and use it to NOT suggest SSD to a client. No possible professional way! If some pattern matched the anomalies described in this thread, would I point out that the possible outcome might be different in that special case? Sure thing. That is responsible DUE DILIGENCE.

 

There is a thing called Computer Science and with that there is Database research. Nothing I am saying is a secret. You can look it up actually!

 

 

 

 

Link to post
Share on other sites
Mindovermaster
7 minutes ago, DevTech said:

Read my explanation again and try to understand.

 

The OP has made a decision and I have made no comments on his decision. And I will continue to not make any further comments on his configuration.

 

I have asked him if he can find the time sometime to perform the same type of test that @sc302 performed and I full expect the PROBABLE outcome to be that th SSD will show a large performance improvement. Because that is the logical result to expect.

 

If it does not improve I would just add it to a mental check list of ANOMALIES for which both computer science and reality are plentiful.

 

Would I take an ANOMALY and use it to NOT suggest SSD to a client. No possible professional way! If some pattern matched the anomalies described in this thread, would I point out that the possible outcome might be different in that special case? Sure thing. That is responsible DUE DILIGENCE.

 

There is a thing called Computer Science and with that there is Database research. Nothing I am saying is a secret. You can look it up actually!

 

 

 

 

No one else talks tech like you... I've been here awhile. NO on else takes up 4 posts and 6 paragraphs of stuff (except for BudMan...)

 

Around here, we do not make scenarios, we get right to the center of the problem. What you are talking about is fantasy, it does not equal to real life.

 

That's just my opinion. Take it or burn it. Or... something...

  • Like 1
Link to post
Share on other sites
DevTech
3 hours ago, sc302 said:

I ran an entire erp system on 15000 rpm disks on a fiber channel array. Went to flash...0 increase.  Oracle jd Edwards Erp system with 150+ users is going to be a bit heavier than a 5 user file maker pro db. 

If you are interested, we could explore this.

 

The default result to expect is a fairly large performance INCREASE. 

 

To me, that means there is another more overriding performance bottleneck that is preventing the difference from being seen. CPU/RAM/Network/Oracle/Security software/Type-of-RAID/etc could all erase a perf diff.

 

Link to post
Share on other sites
DevTech
3 minutes ago, Mindovermaster said:

No one else talks tech like you... I've been here awhile. NO on else takes up 4 posts and 6 paragraphs of stuff (except for BudMan...)

 

Around here, we do not make scenarios, we get right to the center of the problem. What you are talking about is fantasy, it does not equal to real life.

 

That's just my opinion. Take it or burn it. Or... something...

You've gone over the edge with a personal attack.

 

It would appear that you enjoy social bullying and because nobody else is precise, detailed and has a depth of knowledge I am providing you have some right to make a personal attack instead of taking the time to read up on databases. 

 

I'm not impressed here.

 

  • Facepalm 1
Link to post
Share on other sites
sc302
15 minutes ago, DevTech said:

If you are interested, we could explore this.

 

The default result to expect is a fairly large performance INCREASE. 

 

To me, that means there is another more overriding performance bottleneck that is preventing the difference from being seen. CPU/RAM/Network/Oracle/Security software/Type-of-RAID/etc could all erase a perf diff.

 

You keep talking like you know for sure it will increase his database speed by switching to flash/ssd.  I wouldn’t bet a single cent on it.   You say it is faster, and it is true that it is capable of being faster, however if he isn’t taxing the system as it sits, he will more than likely not see any gains.  Unfortunately I can no longer perform extensive tests to verify db speed as the old storage has been disconnected and disassembled, I can tell you that the user experience has not changed from before to after and I do not see any speed increase with vms (booting, launching apps, running queries, etc).  From what you say is that I should see a huge performance gain.   I wouldn’t see performance gains even if I went to 40Gb/s on the back bone.  The links aren’t taxed enough to necessitate even the 8Gb/s backbone. Upping to flash/ssd was ultimately unnecessary for speed.  

 

But what do I know, I just manage and architected the data center, I have absolutely no clue what I am talking about.  I am not seeing real life, it is all fictitious make believe..it is all in my head. 

 

You keep beating the horse that says he will see a performance gain and I will defend the horse saying that it isn’t necessarily true. Put whatever facts and whatever you believe to be true and I will continue to doubt your theory. I have seen where it isn’t true, just because it is a ssd doesn’t mean the db/client interface will be any faster. 

Link to post
Share on other sites
Mindovermaster
4 minutes ago, DevTech said:

You've gone over the edge with a personal attack.

 

It would appear that you enjoy social bullying and because nobody else is precise, detailed and has a depth of knowledge I am providing you have some right to make a personal attack instead of taking the time to read up on databases. 

 

I'm not impressed here.

 

Umm, it seems like SC302 agrees with me.

 

I'm not attacking anyone. I am just saying what I see.

 

Social bullying? I'm not disregarding your information. I'm not saying you are wrong.

 

  • Facepalm 1
Link to post
Share on other sites
DevTech
3 minutes ago, sc302 said:

You keep talking like you know for sure it will increase his database speed by switching to flash/ssd.  I wouldn’t bet a single cent on it.   You say it is faster, and it is true that it is capable of being faster, however if he isn’t taxing the system as it sits, he will more than likely not see any gains.  Unfortunately I can no longer perform extensive tests to verify db speed as the old storage has been disconnected and disassembled, I can tell you that the user experience has not changed from before to after and I do not see any speed increase with vms (booting, launching apps, running queries, etc).  From what you say is that I should see a huge performance gain.   I wouldn’t see performance gains even if I went to 40Gb/s on the back bone.  The links aren’t taxed enough to necessitate even the 8Gb/s backbone. Upping to flash/ssd was ultimately unnecessary for speed.  

 

But what do I know, I just manage and architected the data center, I have absolutely no clue what I am talking about.  I am not seeing real life, it is all fictitious make believe..it is all in my head. 

I was not attacking your experience.

 

I was taking your previous description of your approach to things and offering to work with you on it in some sort of collaborative manner. It sure looks like somewhere you have a mysterious and interesting bottleneck.

 

Were you making that up about wanting to learn stuff? I feel very let down if so.

 

But it seems like everything here has to be adversarial. Mindmaster sees it as a "fight" but that is boring and tedious.

 

I am loosing my faith in this community to be frank...

 

Link to post
Share on other sites
sc302
31 minutes ago, DevTech said:

I was not attacking your experience.

 

I was taking your previous description of your approach to things and offering to work with you on it in some sort of collaborative manner. It sure looks like somewhere you have a mysterious and interesting bottleneck.

 

Were you making that up about wanting to learn stuff? I feel very let down if so.

 

But it seems like everything here has to be adversarial. Mindmaster sees it as a "fight" but that is boring and tedious.

 

I am loosing my faith in this community to be frank...

 

No.  I am interested in learning new things however in this case I already have people all over this. While a second or third set of eyes is helpful in this case I am uncertain of your abilities beyond emc/dell or understand the storage collects that I have sent over to verify storage configs. They have stated I have more compute capability than storage and that my sp utilization is low. Essentially the sans are underutilized even the 15000k storage was also underutilized. Far more capable than what I am putting them through which is why it is fast and why I am not seeing performance gains. It was over spec’d for the environment in both cases. 

 

Far more  goes into it than just flash vs spinning disk.  

 

For him, being that the db is also under utilized and the server is way under utilized (I wouldn’t consider 5 users in a db would kill the db throughout or really tax a decent server even if it was a few years old), he won’t really see a performance increase. 

 

Dont be so bull headed.  It isn’t always that simple where you can say do this one thing and you will absolutely see a performance increase....I haven’t dealt in absolutes ever.  I don’t make an assumption that others don’t have their stuff in order unless they prove it or ask for help. I am not asking for help, I have my environment pretty good.  Where were you during my sales calls or during the design phase. You have no idea of how over spec’d my environment is (remember I stated I used to recommend what I would buy, understand that I over plan/over spec for my environment).  Not seeing an increase in performance wasn’t unexpected, but note that there is no noticeable performance increase. 

Link to post
Share on other sites
DevTech
24 minutes ago, sc302 said:

No.  I am interested in learning new things however in this case I already have people all over this. While a second or third set of eyes is helpful in this case I am uncertain of your abilities beyond emc/dell or understand the storage collects that I have sent over to verify storage configs. They have stated I have more compute capability than storage and that my sp utilization is low. Essentially the sans are underutilized even the 15000k storage was also underutilized. Far more capable than what I am putting them through which is why it is fast and why I am not seeing performance gains. It was over spec’d for the environment in both cases. 

Well if it is possible to see anything I write down in a non-adversarial way, I will say:

 

1. The SSD WILL write faster, just no way that is not possible.

 

2. The main perf limit on most databases is the insane difference between RAM and disk which is why all sorts of cache and other algorithms are applied to avoid a disk write and why "In Memory" database software like Redis has become so popular.

 

3. Oracle could be caching the writes so well that within the measurement window for your DB perf instrumentation there is equally enough time for the 15K write or the SSD write. But still no matter how much spare compute you have the DB should be faster on some sort of measurement.

 

4. It could also be the case that you have an exceptionally high read to write ratio.

 

5. I am sure you will enjoy the search for the mysterious weirdness - I always suspect security software in any enterprise for performance issues. 

 

 

My "area of expertise" is software programming/design/architecture and I have been CTO of the last few companies (start-ups) that I been associated with FWIW which IMO isn't much meaning since data is data always. I am for sure NOT CEO material and I recognize that.

 

Still feeling rather disturbed and I might take a hiatus from the forums for a while, but that's no comment on your personal wonderful attempt to be understanding.

 

Link to post
Share on other sites
patseguin

Someone told me that raid6 is better than raid5 because 2 drives can fail. As for speed, I only build my home systems with ssd's and am well aware of the performance increase. For my needs, platter drives were much more cost effective and have been plenty fast enough for FileMaker server. I do have the freedom to replace my drives with ssd when and if the need arises.

 

Let's watch the personal attacks too. I had no idea I'd be creating such an emotional thread. 😀

Link to post
Share on other sites
sc302
5 hours ago, DevTech said:

Well if it is possible to see anything I write down in a non-adversarial way, I will say:

 

1. The SSD WILL write faster, just no way that is not possible.

 

2. The main perf limit on most databases is the insane difference between RAM and disk which is why all sorts of cache and other algorithms are applied to avoid a disk write and why "In Memory" database software like Redis has become so popular.

 

3. Oracle could be caching the writes so well that within the measurement window for your DB perf instrumentation there is equally enough time for the 15K write or the SSD write. But still no matter how much spare compute you have the DB should be faster on some sort of measurement.

 

4. It could also be the case that you have an exceptionally high read to write ratio.

 

5. I am sure you will enjoy the search for the mysterious weirdness - I always suspect security software in any enterprise for performance issues. 

 

 

My "area of expertise" is software programming/design/architecture and I have been CTO of the last few companies (start-ups) that I been associated with FWIW which IMO isn't much meaning since data is data always. I am for sure NOT CEO material and I recognize that.

 

Still feeling rather disturbed and I might take a hiatus from the forums for a while, but that's no comment on your personal wonderful attempt to be understanding.

 

 

 

Dev you are simply going overboard and I am simply trying to reel you in a bit.  My messages aren’t to attack but to explain it isn’t always the case....I will give you a piece of information that you didn’t know before and the storage is a little red herring that I threw out there, the middleware is java based...single treaded java processes.   The only way I can ever see speed is if I upgrade compute, less cores/more Hz.  I am on the latest release :(  No storage upgrade will fix that, esp since I am no where near even 10% utilization on the storage processors and bandwidth to the hosts. I peak at 8, average is about 5.  The unity arrays are way overkill. 

Link to post
Share on other sites
+BudMan

yeah this system seems a bit over the top if you ask me.. You have 5 users.. Really raid 6 for what amount of data?  How much were those drives?  They are only 1.2TB?  You prob have saved money just getting SDD and doing Raid 1 if you were worried about something failing..

 

The box is prob way over the top cost wise for 5 people... You could prob of gotten by with some $1000 off the shelf nas to be honest... You sure don't need to be paying the MS tax for windows server licenses with 5 users.. Any NAS os would allow you to be a "domain" and central user management, without paying the ms tax..

  • Like 1
Link to post
Share on other sites
DevTech
18 hours ago, sc302 said:

 

 

Dev you are simply going overboard and I am simply trying to reel you in a bit.  My messages aren’t to attack but to explain it isn’t always the case....I will give you a piece of information that you didn’t know before and the storage is a little red herring that I threw out there, the middleware is java based...single treaded java processes.   The only way I can ever see speed is if I upgrade compute, less cores/more Hz.  I am on the latest release :(  No storage upgrade will fix that, esp since I am no where near even 10% utilization on the storage processors and bandwidth to the hosts. I peak at 8, average is about 5.  The unity arrays are way overkill. 

The point, is I have no point other than curiosity.

 

I get that overall, you can't see a storage effect. And I won't point out that using it as an example for seeing no perf improvement between storage types meant that it was a data point of zero value to anyone trying to have an understanding of that.

 

But it is still interesting to just measure the DB component. DBs are usually very sensitive to data writes so if a RAID-5 spinning plater write takes .5 seconds (for example) and the SSD write takes .005 seconds, there should be a cumulative performance boost that should bubble up in the DB perf instrumentation. And if you see that, then it is just normal expectation, even if it has no "overall system" effect. But if you see something else then it becomes an interesting anomaly. 

 

I am sad that nothing ever calmed down enough to jointly discuss interesting anomalies. But that's the way the cookie crumbled and so thanks for the info, looks like you are doing good stuff...

 

Link to post
Share on other sites
DevTech
13 hours ago, BudMan said:

yeah this system seems a bit over the top if you ask me.. You have 5 users.. Really raid 6 for what amount of data?  How much were those drives?  They are only 1.2TB?  You prob have saved money just getting SDD and doing Raid 1 if you were worried about something failing..

 

The box is prob way over the top cost wise for 5 people... You could prob of gotten by with some $1000 off the shelf nas to be honest... You sure don't need to be paying the MS tax for windows server licenses with 5 users.. Any NAS os would allow you to be a "domain" and central user management, without paying the ms tax..

There have been so many subjective viewpoints on this config that it has been an eye-opener for me as everyone clings to their favorite thing like life and death. Or as I liked to portray it - clinging to their warm and comfy Teddy Bears...

 

Personally, I was attracted to the SuperMicro mobo suggestion by @Bryandue to the "Homebrew at Super High Performance for Far Less Money" vibe, but I can see where everyone on this thread is coming from.

 

And warning, your suggestion to use SSD will have the "Teddy Bear Police" on you like greased lightening on a water drenched groundhog...

 

 

Link to post
Share on other sites
patseguin

Here's a question for you guys. Can I keep my current server on the network since it already has my embroidery design database running and just use it solely as an embroidery design database server? I'm wondering if I'll have to change its IP address and demote it from a domain controller. I'm guess I will have to do that. Right now it is 192.168.10.2 which AFAIK is the standard setting for a server. I'll make my new server 192.168.10.2 and just change the old server to to a dynamic IP and connect it to the domain that will be set up on new server. Does this sound like a reasonable plan?

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    No registered users viewing this page.