How to organize the server 1C and SQL?

Hello

Is server:
Intel Xeon E5-2690v4 Processor * 2
128 GB RAM
SSD 100 GB * 2
SAS 2 TB * 4

There is a task:
To organize the work of the server 1C. users are now up to 30, but plan to increase to 100.

What do you think is more profitable:
1) on the host machine Hyperv-server. on the virtual 1s and sql ?
2) on the host machine MS Server 2012R2 and it is the server 1C and sql ?

I would have carried on virtualnye _)) But now many say it can be put on 1 car and 1C and sql and all will be well and faster at 30 and above protsentov))
June 8th 19 at 16:35
4 answers
June 8th 19 at 16:37
Put on the physical, do shared memory, limit the ms sql memory and all will be well)
Physical uncomfortable to administer, for a quick recovery from possible failures.
Modern virtual machine Hyper-V, ESX, etc. - eat a measly interest of performance. - efrain.Rolfso commented on June 8th 19 at 16:40
June 8th 19 at 16:39
Only virtual machines. On each machine - one role. And a second server for failover. Performance will not.
You can of course, as Neanderthals, all on one set, but then do not write here questions what to do if after the next update of any system you have lost EVERYTHING. And you for the night you will need to raise and customize them all.
unfortunately there is no second server,_))
if he was, we could separate hardware to deploy, although the fault tolerance of 0 then _)) - efrain.Rolfso commented on June 8th 19 at 16:42
In any case, supply the bare hyperv on any computer and deploy backups of virtuallock much simpler system recovery on another platform. - Rhianna commented on June 8th 19 at 16:45
you, apparently with the enterprise faced. Deploy to dev with MSSQL, which requires 64GB operatives and 100-200GB SSD for every computer is not always possible. - Andres_Crooks commented on June 8th 19 at 16:48
but I was faced with moving a system to another SAS controller. Don't want to. - tomasa_OConn commented on June 8th 19 at 16:51
why 64 and 100-200? when there is fire and everything is falling apart, to raise the working database at a hastily assembled gland, very valuable.. the day it will work and not with a bunch of RAM and SSD - Andres_Crooks commented on June 8th 19 at 16:54
June 8th 19 at 16:41
On the path - easier to maintain.
The difference with the iron in performance of 5% percent, that is, you do not even notice.
But in the case of a failure to raise the V-world from a backup (if you have one) - a matter of minutes, all sorts of different not comparable with many hours of tuning on the hardware.
June 8th 19 at 16:43
1) on the host machine Hyperv-server. on the virtual 1s and sql ?
Beneficial in terms of ease of administration.
Extremely disadvantageous in terms of performance. If your convenience is more important than performance, then it is a normal variant.

2) on the host machine MS Server 2012R2 and it is the server 1C and sql ?
Maximum performance, minimum standard.
Rheine is disadvantageous in terms of performance.

Well, if a half percent for you "Extremely lowers productivity".... - efrain.Rolfso commented on June 8th 19 at 16:46
, What's a half percent? In two or three times at least.
Or get the data directly from memory, or to drive data through two network interfaces, queues and delays inherent in networks. - Rhianna commented on June 8th 19 at 16:49
the point is that the server 1C and mssql is undesirable to breed for different machines, whether virtual or physical - Andres_Crooks commented on June 8th 19 at 16:52
why undesirable? - tomasa_OConn commented on June 8th 19 at 16:55
,

What a half percent? In two or three times at least.
Or get the data directly from memory, or to drive data through two network interfaces, queues and delays inherent in networks.


1) You're being cheated.

2) Remember, the student, in any task where intense DBMS is used, when the competent organization of this task - the main load on the digestion of data just to lie on the DBMS. And already digested in a DBMS for a small amount of the result goes over the network to the application.

3) Locally, you're going to refer to the DBMS via network interfaces too. And Yes, they will of course virtual, not tied to real hardware. But not much more simple than the virtual interfaces of the virtual machines. - Andres_Crooks commented on June 8th 19 at 16:58
,
answer for "student"
1) no one was deceived, the man is clearly working with 1C ;)
2) you know how the 1S works with database? there are used only the most basic f-tsii the base without digestion forces the DBMS and the reduction of costs in this place positively affects the performance
3) locally to the DBMS can (and in this context we need!) contact via SharedMemory (this is the way the answer to why it is undesirable to carry the server) - aliya.Miller82 commented on June 8th 19 at 17:01
,

you know how the 1S works with database? there are used only the most basic f-tsii the base without digestion forces the DBMS and the reduction of costs in this place positively affects the performance


You describe seven. She has passed away for 10 years. Though some where used.
;) - efrain.Rolfso commented on June 8th 19 at 17:04
you extremely overestimate the eight ;)
the situation is not much better - earl.Weissnat commented on June 8th 19 at 17:07
,

1) no one was deceived, the man is clearly working with 1C ;)


Yes, it's obvious that he was working with 1C:

The Effect Of Dunning — Kruger - aliya.Miller82 commented on June 8th 19 at 17:10
,

3) locally to the DBMS can (and in this context we need!) contact via SharedMemory (this is the way the answer to why it is undesirable to carry the server)


And, you can do from the operating system to refuse. She takes part of the resources.
Oracle at the time, did your DBMS to work with disks directly, bypassing the file system. What? I think once a fantastic growth of productivity? Not at all. Silver bullets do not exist.

What exactly 1S to PHYSICALLY distribute servers is not worth it.
But on different virtual machines within one physical machine - there has been no terrible performance.
The network uses 1S not so much. - efrain.Rolfso commented on June 8th 19 at 17:13
,

you extremely overestimate the eight ;)
the situation is not much better


Based on what an eight should show the user a bunch of information when users scroll through the log screen documents for example - of course it pulls a lot of data.

However, guess what else in coffee grounds. Take at least the built in Windows Profiler and measure the load on the network.

Will be very surprised.

Seems to be Yes, 1C uses the network intensively. But the network card is not so even and loaded.

Well, the dev is quite fit.
It slows down the work on these funny

What a half percent? In two or three times at least.

as written, Artem.
What are the two times, Tom?

Even in the worst scenario, and 10 percent reaches. - efrain.Rolfso commented on June 8th 19 at 17:16
"The Effects Of Dunning — Kruger"

I'm not a particularly you to prove it, but when confronted by a different large enterprises with integration of 1C, ALL the integrators definitely recommend to put on the server together... or are they not professionals? don't understand the poor the network does not slow down? what sharedmemory stupidity?

fantastic growth of course not, but the fight performance in large databases, 1C is one of the most pressing problems stretching from seven so far, and still there is no definite way to victory in this battle are all the possible ways - aliya.Miller82 commented on June 8th 19 at 17:19
Oh what a discussion divorced.
What's a virtual machine anyway? Was given the specific situation.
Post on different machines the server 1C and whining is not recommended due to the drawdown performance.
It is a fact. Proven experience in operation of standard solutions 1C.
Yes, in the eight if to take and to optimize the code, it is possible to minimize the exchange of information between the server 1C and Skolem.
But who the hell will do it? To hire programmers and optimize a bunch of standard code that is constantly changing. On the support splurging, programmers 1C very greedy people, money is love.

What are the two times,
The figure is not taken from the ceiling. Base model 1C operate in the SharedMemory two to three times faster than over the network.
So spread 1C and skul ' on different servers in two situations -
1)Iron is banal does not cope to pull both.
2)When the performance is not very important, resources are guaranteed more than necessary, and in the first place is convenience.

Seems to be Yes, 1C uses the network intensively. But the network card is not so even and loaded.
How do you assess the response of the network card? In the task Manager? Or similar instruments?
Network card can be loaded by 1% and thus can be wild brakes due to the fact that network card can not cope. - efrain.Rolfso commented on June 8th 19 at 17:22

Find more questions by tags Hyper-VWindows ServerSQL Server