Today we look at the choice of a server "hall" for a small organization for 25-30 employees, with a separate infrastructure (retail outlets, a warehouse), which kind of terminal server is needed and the program "1C: Enterprise". These services serve all practitioners.
Most of the small companies, in order to cheapen the cost of ownership, will allow to minimize the amount of equipment that they buy and ask the administrators to "pump" all the services powered by them into one physical server. Bazhannya is more sensible and provocative, but here there are nuances.
You can organize a terminal server and win the file version of 1C there, but for such a large number of coristuvachiv, the retail company recommends switching to the client-server version. That's why we need another server called "1C: Enterprise" and a database server. Let me clarify that it is possible to organize a terminal server, SQL server and 1C server on one operating system, but from the point of view of security and stability of robotic services, it is not recommended in the region. And if you still want to win one physical server for all three roles, then it is recommended to win virtualization, for example, VMWare ESXi or Hyper-V.
In this order, there are three options:
To complete these tasks, you can request the following server configuration:
Wu vipadu z one physical server We reuploaded the choice on a Dell R710, with two six-core Xeon X5650 processors, 64 GB of RAM and six drives: two SSDs for RAID 1 and two SAS drives for RAID 10.
Wu vipadu z two physical servers We've zipped vibir on the following configurations:
From a glance of the processor hour, terminal sessions take up a small part. For the provision of terminal solutions in various organizations, for support of comfortable work 30 cores, 4-6 physical processor cores will be enough, one core for 6-8 sessions. For a small database, SQL Server needs one core. But let's focus on the expansion of the database in the future (or increase the number of databases) and take two cores in SQL.
For the "1C: Enterprise" server, the number of cores is not so important, as their clock frequency and bus frequency. Therefore, we put two more cores on the 1C server.
And let's not forget that we need one or two cores in order to secure a robotic host operating system.
Together we have to enter:
We wonder a little, how much operational memory is needed for services: 
Ce traditional plyashkov neck bagatioh systems. Correct choice hard disks are important for securing swedcode servers. When working with 1C with a SQL base, there are impersonal read / write operations per second (IOPS). How to work hard on a terminal server with thin clients (that is, it is completely hard to work on a terminal server like a working medium), but it strongly interferes with the disk system of the server. For example, 30 terminal server boxes on RAID 1, SATA 3 Gb/s, with WD Velociraptor disks will feel uncomfortable for an hour of work due to active surfing on the Internet. For terminal servers, we recommend that you use SSD drives. For database servers - SAS-disks, selected in stack to drive array. Krim accumulating, slid to give respect to the disk controller. Modern servers can do good controllers on board, for example HP SmartArray and DELL PERC. However, it is not correct to choose the "onboard" solution in case of serious ambition, if the maximum productivity is required. Having spared the troubles, you can easily take away a hard server, which is not a burden. Therefore, the controller can be hardware, not software, because of its non-volatile memory.
Let's take a look at the options for rozvyazannya of this task.
Another array is better created from two SAS disks in RAID 10 (mirror + stripe), or it is possible to use two SSD-storage disks in RAID 1. Choose to store only a few disks of that server model.
As it is said above, small organizations should have a great opportunity to spread all the services on one server. Until the victory of one server and virtualization, you can use lower energy and flexible resources between virtual machines. That porting of virtual machines, in a different way, is richer than the porting of physical OS.
However, two servers may have more room to upgrade. For example, in our version, an inexpensive IBM x3550 M3 with the addition of one more processor and RAM is transformed into an elegant short terminal server for 50 and more cores.
One more "high school" in our mind, as it is necessary to be safe when choosing two physical servers, the exchange of data between them is as good as possible. For virtual servers, data exchange goes through a virtual switch. Right there, to increase the throughput capacity of the network, you can insert into the skin server on the basis of the card with two gigabit interfaces, so you can aggregate between yourself and without intermediary connection between servers and aggregated 2 gigabit links. Otherwise, you can try different cards with SPF + 10GBASE, but on the way you will be satisfied.
So, if you need to expand, or the number of services will increase, then there are great prospects, and for the time being, the servers will effectively manage their tasks. Possibly, through the river, we will not be able to increase the number of coristuvachiv vdvіchі, up to 60 osіb? Please.
Once you have won one DELL R710 server, you can buy an inexpensive IBM x3550 M3, put it on a new hypervisor, transfer a virtual machine with a database and a 1C server to it, and all resources on DELL and use a virtual machine with a terminal. Tse will be quick, and it will not be necessary to "throw everything away and buy new."
Just as you have won two IBM servers, then the x3550 M3 with the addition of another processor, that small amount of RAM is transformed from an average machine to an exhaustive machine. And in the x3250 M5, you can upgrade the processor from E3-1220v3 to E3-1285v3.
How to organize a comfortable robot 7 and more people for 1c? How to secure the robot without interruption with the 1c system? How to guarantee the safety of all data 1s? Yaky buy server for 1c ta yak yogo vibrati correctly? It's too early to stand before IT-fahivtsy and kerіvniki organіzatsіy.
Select configuration servers for robots 1C to lay aside the basis for the activity of the work with it. The following recommendations are developed on the basis of the 1C company's best practices and practice. Покупая 1С сeрвeр необходимо убедиться что он отвечает сoврeмeнным трeбoвaниям oткaзoустoйчивoсти и прeдусмaтривaет пoвышeниe нaгрузки при нeoбхoдимoсти рaсширeния прoизвoдствeнных прoцeссoв нa прeдприятии, гaрaнтирoвaть высoкую рaбoтoспoсoбнoсть сeрвeрoв 1с при любых нaгрузкaх, их дoлгoвeчнoсть и высoкиe эксплуaтaциoнныe хaрaктeристики
| up to 20 | up to 30 | up to 50 | up to 100 | |
|---|---|---|---|---|
| Processor | 4-core processor Intel Xeon E3-12xx | 2 processors Intel Xeon E5-26xx | ||
| Memory | 16 GB RAM | 16-32 GB RAM | 32 GB RAM | 64 GB RAM |
| Number of units | type 1U | 1U or 2U | 1U or 2U | type 3U |
| Disk subsystem | 2 x SAS | 4 x SAS | 8 SAS hard disks (RAID 10), can be configured with SSD disks | 16 SAS hard drives (RAID 10), can be configured with SSD drives |
| Hardware RAID controller | cache is recommended | iz zahistoy kesh | iz zahistoy kesh | iz zahistoy kesh |
| Possibility scaling that platform from the installation at the rack |
Є | Є | Є | Є |
| Orientation variant of 1 server for the database | $2 000 | $4,000 | $5 600 | $9 990 |
| Contact our consultant for clarification of the configuration and variance | ||||
| Recommended number of servers for architecture pid 1C | 1 | 1 | 2 servers in a cluster | |
| Zovnishnya disk police | Hi | Hi | So | So |
| Commentary | For the database (May change the functions of the 1C server), Internet gateway, File server. It is possible to upgrade single-processor machines in cases with 8 disks. | For database, Internet gateway, File server | It is recommended to have 1 or 2 database servers, united by a cluster, called SRS. Physical subserver is recommended: database server, addendum server, terminal server | It is recommended to have 1 or 2 servers for the data base, united by the cluster, called the SRS. Physical subserver is recommended: database server, addendum server, terminal server |
On the basis of the above table, you can add the 1c server configuration for a small office
* Dodaєtsya, as it is necessary to overcome the distant working tables.
Organizational diagram of such a solution looks like this.
Yak database server We recommend following machines:Dell PowerEdge T320, Dell PowerEdge R420, Dell PowerEdge T620.
servers Dell T320і R420 mostly constructive (subjective and stichkovy v_dpovidno), and Dell T620 accommodating a larger number of disks and operational memory, which may be needed with a high level of ambition, or "with an eye" on the future, as a business company. Even if there is not enough space in the country, you can transfer your respect to a compact 1U server Dell R320.
The main areas of the data base server are the disk subsystem and connections to the operative memory. The amount of data base expansion for such companies is usually small (typically no more than 5-10 GB), then it is possible to increase the amount of cash. В oбщeм-тo этo нe oбязaтeльнo, oсoбeннo eсли aктуaльнa нe вся БД (нaпримeр в нeй присутствуют дaнныe пo прoшлым гoдaм, нужныe лишь врeмя oт врeмeни), нo кaк минимум нужнo зaлoжить oбъeм OЗУ нe мeнee 30-50% oт рaзмeрa БД для цeлeй кэширoвaния . Plus, zrozumіlo, like a minimum of 1 GB for the OS needs. How to work on which physical server addendum server 1C, Then you need to see the memory and you - from 1 GB to 2-4 GB (it's better to consult with the franchisee - it's better to deposit in the form of їх configurations).
As a data base server with software 1C 8.2, we will rent for 50-100 vouchers, it is recommended to voucher servers Dell PowerEdge T620, Dell PowerEdge R720і Dell PowerEdge R720XD. The stench may be hard disk subsystems for 16 and 24 disks. Eсли нaгрузкa нa сeрвeр БД пoстoяннo рaстeт, мы рeкoмeндуeм нe экoнoмить и выбирaть сeрвeры с бoльшим кoличeствoм дискoв, пусть дaжe нe в пoлнoй нaбивкe - лучшe пoтoм дoбaвить дискoв и пaмяти, чeм чeрeз гoд пoкупaть бoлee мoщную мaшину.
Yak addendum servers Be the best choice Dell PowerEdge T420 4-8 GB of memory. In principle, you can certainly marvel at a single-processor machine, or rather, two more weak processors, lower one hard one (do not cheat with 1000000000000000). The adventage on the server of the addendum must be very stale due to the 1C configuration you have chosen, it is recommended that you consult with your provadzhuvachami.
Z terminal server easier than anything - terminal servers scaled horizontally. So you can just put two or three Dell PE R420 or R620- Zalezhno vіd navantazhennya. Moreover, the data is automatically secured - in the event of a breakdown of one server, client sessions can be transferred to another. Golovnya - far ahead of time to supply the ZZU from stock
Selecting which server is needed for 1C, the next memory, what is the hour of robotic work with it, there will be an impersonal operation, reading and recording data in a second.
For everything, I realized why it is so important to competently design a server for 1C - it’s like it’s “slowed down” on the back of the head, it’s chosen incorrectly and it doesn’t show the pressure on the system, it’s a risk, it’s a risk, it’s because of interruptions, it’s an important data to get involved. From the other side, create a 1C server, buy all the equipment for the new one software security You can save the most important amount for the company, so it’s better to pick up the possessions so that you can save your money.
If it is necessary for our fahіvtsy robiti vibіr config for server 1C, first, about scho stink inquire - skіlki koristuvachіv bude pratsyuvati z 1С in the company and some set of services are planned vikoristovuvat, what will be, hto and yaknіmіm_vіkonu Vіd tsikh vіdomosti mi vіdshtovkhuєmosya pіd hour of creation of the server 1C.
In the hardware structure of a 1C server, the characteristics of the processor, RAM, disk subsystem and interface will be important for us.
It is necessary that the stench ensure stable and productive work of the attacking components:
The optimal number of processor cores should be allocated, depending on the fact that it is necessary to reserve 1-2 cores for the OS robot, 1-2 cores for the SQL database robot, and 1 additional core for the server robot, and approximately 1 core per skin 8-10 one-hour core sessions (Shob koristuvachі potіm not skarzhilis, scho server 1C galmuє).
To pay attention to the fact that the speed of the power supply is not so much due to the number of cores, as it is to the clock frequency of the processor, and the number of cores is more influencing the stability of the work with a large number of corystations and one-hour tasks in them.
Skіlki mem'yatі it's necessary for server 1С
In addition, if you need a 1C server for 100 or more cores, we recommend building a cluster with at least two 1C physical servers.
Rozmіr nebhіdnoї operative ї mem'yatі mi proponuєmo rahuvati, based on such indications:
Let's set up our orientation of the distribution of the parameters of the server 1C 8.3:
RAM is better bought with a margin - it is one of the most important factors in the high productivity of a 1C server and at the same time one of the found components at the same time. If there is not enough memory on the 1C Enterprise server, it will be even more important to work for an hour, if it costs food, which 1C server to choose, start paying respect to those who have enough RAM.
Selecting which server is needed for 1C, the next memory, what is the hour of robotic work with it, there will be an impersonal operation, reading and recording data in a second. This parameter - with some kind of security, a hard disk allows data processing - it is also one of the key ones for the 1C server's security code.
When designing a 1C server, it is possible to achieve the following before having a disk subsystem:
At these stats, we have brought some deacons to give rise to approximately rozrahunki, how to choose a server for 1C, I guess, the stench will be red for you.
At the end of the day, one more thing - don’t varto magatisya to spare, vikoristovuyuchi computer for the 1C server (as it often works in small companies) - koristuvalnitska "zalіzo" kudi less nadіyno and vіdmovostіyke, nіzh similar to the productivity of the server. Do not varto rizikuvat with the oblіkovoj system of your undertaking. Even if you don’t invest in your budget, it’s possible, next, you can look at the possibility of opening 1C in the dark
It’s easy for you to sort out, which server to choose for 1C Enterprise 8.3, how to build a 1C server, so you didn’t stick to these tasks earlier, you can always go back to the company-system integrator, so that you will be able to complete the project nalashtuvati vidpovidny you server for 1C.
1С:Pіdpriєmstvo 8 can appear as a resource-saving supplement to build with a small number of coristuvachivs. Vibirayuchi server pіd 1C, be it a sorcerer who wants to be undone "birth trauma" - laying the foundations for a new potentially narrow area. From the other side, today, few people buy servers and supra-mundane tension, on the virist. Dobreyakscho profile navantazhennya vdaetsya znyat zazdalegіd - then design the server for a specific configuration addendum in the company is simpler.
For reference, let's look at the platform "1C: Enterprise 8.2" in її popular basic configurations "Accounting appearance", "Trading and warehouse", "Salary and Personnel Management", "Management of Trade Enterprise" and, in particular, "Management of Vacuum Production". It seems that for undertakings from 10 and more spivrobitniks, which work at 1C, “1C: Undertaking 8.2. Addendum Server. You can change the option of working in the remote desktop mode (Remote Desktop), with the number of one-hour data bases up to 100-150. Recommendations will be fixed for "important" DB 1C, but "important fluctuations" will depend on an individual approach.
Since the company is small (2-7 files per system), the base is small (up to 1GB), and "1C: Enterprise 8.2" works in the file mode on the computer, then we need a classic implementation of the file server. For such tasks, the CPU is likely to be called Intel Core i3, more so Intel Xeon E3-12xx. The amount of RAM required is even easier: 2GB for the operating system and 2GB for the system file cache.
If in the company of 5-25 1C cores, the data base size is up to 4GB, then the "1C: Enterprise 8.2" add-on can drive 4-core Intel Xeon E3-12xx or AMD Opteron 4xxx. Crim 2GB RAM under OS, need to see 1-4GB under 1C:Enterprise 8.2. Addendum server” and more styles and more MS SQL Server as a cache - total 8-12GB RAM. For small databases, the amount of cash in the operative memory is at least 30% of the database, and more often, all 100%.
Vіdomy (though not particularly advertised) fact: “1C: Pіdpriєmstvo 8.2. Don't like the add-on server, if the operating system takes it to a swap file on a hard drive, and it's easy to use it in any case. That's why on the server, where the "Additional Server" is launched, you can always have a reserve of free space in the operational memory - it's more inexpensive today.
The companies have more coristuvach 1C sound through remote access to the program (Remote Desktop) - that's in the terminal mode. As a rule, with 10-100 1C data bases with a data base of 1GB and more, “1C: Enterprise 8.2. Add-on server" and add-on "1C: Enterprise 8.2" run on the same server.
To determine the necessary processor resources, it should be taken into account that one physical core can efficiently process no more than 8 core streams - this is due to the internal architecture of the processors. As practice shows, under the 1C + Remote Desktop task, it is not possible to take server processors of young lines with low frequencies of rozrachunk cores and a rudimentary architecture. As a rule, it is rather small (up to 15-20), to extract one processor from the high-frequency Intel Xeon E3-12xx. At least one physical core (2 threads) will consume SQL Server, another one (2 threads) will consume “1C: Enterprise 8.2. Add-on server, and 2 physical cores (4 threads) - for OS and terminal cores. If the number of 1C cores is more than 20, or if the database is larger than 4GB, it's time to switch to 2 processor systems on Intel Xeon E5-26xx or AMD Opteron 62xx.
Razrahunok required to obsyagu operational memory is quite simple: 2GB required OS, 2GB and more - MS SQL Server as a cache (not less than 30% of the database), 1-4GB - pid "1C: Enterprise 8.2. Auxiliary server”, the server memory can be saved for terminal sessions. One terminal box, deposited according to the configuration, living with the appendices “Accounting Form”, “Trading and Warehouse” - 100-120MB, “Salary and Personnel Management”, “Trading Enterprise Management” - 120-160MB, . In order to run an add-on on MS Word, MS Excel, MS Outlook servers, the skin add-on needs to see about 100MB. As a rule, the minimum for a terminal server is 12GB RAM.
For example, for a 1C server with the usual software package, 50 terminal boxes in the “Commercial Business Administration” configuration, that data base for 8GB will be optimally calculated by the intensity of two Intel Xeon E5-2650 processors (8 cores, 16 threads, 2.0 GHz). RAM you need at least 2 (OS) + 4 (SQL) + 4 (1C-server) + 8 (160 "UTP" * 50 cores) = 18GB, and more shortly 24-32GB (6-8 DIMM channels of 4GB each) .
More scarg for the full work of the servers 1C:Pіdpriєmstvo 8 pov'yazan z unreasonable, like they are vivouyutsya type of operation introduction-visualization, over some kind of tribute that z іntensivnіstyu. The most important disk subsystem is the key to ensuring sufficient productivity of the server as a whole - even for database navigation, the biggest problem is blocking tables during one-hour work with them without a faceless coristuvachiv or with mass acquisitions / vivantages / postings. Monitoring and optimization of the disk subsystem of the server.
1С May 5 data streams for the disk subsystem, with which they won't work:
The structure of the data in 1C is object-oriented, with no objects and links between them. p align="justify"> For work with data tables, the number of read operations per second is extremely important, as it is to build a disk subsystem in an hour (Input Output Operation per Second, IOPS). At the same time, the ability to see a high streaming speed of data transmission (MBp / s) is much less important. A rather modest base with a volume of 200-300MB from 3-5 shortlisted cells can generate up to 400-600 IOPS in peaks. A base of 10-15 cores and a total of 400-800MB can see 1500-2500 IOPS, 40-50 cores of a 2-4GB database generate 5000-7500 IOPS, and a base of 80-100 cores can easily reach 12000-180.
As you know, the average demand for a disk subsystem can become 10-15% of the peak. Only the productivity itself is really important in the period of peak acquisitions: automatic acquisition of data from other systems, exchange of data from a distributed system during a re-performed period.
Modern disks in read and write operations with random access (Random Read/Write) can handle the following challenges one by one:
|
Intel 910 400 GB |
|||||
|
2400 - 8600 IOPS |
It's good to see that:
Single disks on database servers are not broken, only RAID arrays. For a further increase in the real productivity of the disk subsystem, it is necessary to pay a penalty (penalty) for writing to IOPS, as if the disk group was in RAID:
If you select 6 disks from RAID 10, then 2 IOPS of physical disks will be stored on a 1 IOPS record, and 6 IOPS disks will be used in RAID 6. In this way, in order to increase the capacity of a disk group for recording, it is necessary to add up the IOPS of all disks of the RAID group, and then divide them into a “penalty”.
Example 1: 2 HDD SATA 7200 RAID 1 write-free: (100 IOPS *2) / 2 = 100 IOPS.
Stock 2: 4 SATA 7200 RAID 5 write-free: (100 IOPS *4) / 4 = 100 IOPS.
Stock 3: 4 SATA 7200 RAID 10 write-free: (100 IOPS *4) / 2 = 200 IOPS.
Apply 2 and 3 to clearly demonstrate why, in order to save databases, for some, the typical read/write ratio should be 68/32, which is a shorter RAID 10.
From these three tables, it was clear why the productivity of a typical “gentleman's set” of 2 HDD SATA 7200 in RAID 1 is not enough for the server: at peak pressures, the number of bellies grows to the disk, the systems are short-lived, sometimes for a long time.
How to increase the productivity of the disk subsystem for recording? Increase the number of disks in a RAID group, switch to disks with a larger security wrap, select a RAID level with a lower write penalty. Kindly help cache by the RAID-controller from the enhanced write-back mode Write back. Data is not written directly to disks (as in Write Through mode), but to the controller cache, and only later, in batch mode, and in an orderly way - to disks. Depending on the specifics of the manager, the productivity of the record is increased by 30-100%.
Possibly low-volume or relatively small databases (up to 20GB) is an inexpensive way to increase IOPS - hybrid RAID from SSD/HDD. Larger and not needed files of the database for 3-15 koristuvachiv at the rozpodіlenіy structure on the kshtalt merezhі cafe or service station.
For large (200GB and more) databases with a long historical data loop, or servicing large databases, SSD cache may be effective (LSI CacheCade 2.0 or Adaptec MaxCache 3.0 technologies). For the operation of such systems, in the tasks of 1C for additional help, you can noticeably inexpensively and without major changes in the infrastructure, you can speed up disk operations by 20-50%.
The champion of swidcode in IOPS has transferred RAID arrays on server SSDs - both traditionally, instead of SAS RAID controllers, and PCIe SSDs. Two intermediate factors are important for their popularity: technological (the productivity of RAID controllers or the need to radically change the structure of the savings) and the price of implementation.
Okremo next to say about saving index files and TempDB. Index files are updated very infrequently (call 1 time per doba), they are read more frequently (IOPS). It is simply necessary to save such danim on an SSD, for reading ostentatious! TempDB, which is used to save time data, is usually small in size (1-4-12GB), but is capable of recording speed. Index and hour files will be sent so that you don't have to spend until you spend real data. Otzhe, the stench can be spread on an okremu (or rather - on two okremih volumes) SSD. I want to use the onboard SATA controller of the motherboard. From the point of view of reliability and swidcode, TempDB can be used to mirror (RAID1) from the SSD, it is possible on the on-board controller, and also to obov'yazykovym stash caches on the record. For this reason, desktop SSDs are also affected - for example, Intel 520-series, de hardware data compression when writing to TempDB will be more streamlined. The fault of these leaders from the global system of saving on the vision of the swedish subsystem is positively signified on the productivity of the system, especially at the moment of peak demand.
In case of failures, if it is possible to ensure the maximum speedy response of administrators in case of failures and if transport logistics, Virobnitstvo in UPP, volume exchanges in URBD), TempDB is to blame on RAMDrive. This decision allows you to win sometimes up to 4-12% of the overall productivity of the system. The failure is only responsible for restarting the server: if RAMDrive does not start automatically, it will be necessary to add an administrator for a manual start - otherwise the whole system will stop.
Another important component is the log files. The stench may be unacceptable for any kind of disk subsystem singularity - to generate a constant flow of dribnih animals for recording. It’s unfathomable for medium traffic, but it’s worse for the 1C server’s code for peak traffic. p align="justify"> It's wise to justify the log file (especially the SQL log file) to as many physical volumes as possible, which will be practically linear in terms of IOPS. For peace of mind, you can create a mirror from inexpensive and bulky SATA / NL SAS (for Full log), or inexpensive desktop SSDs all the same Intel 520 series (Simple log, or Full log, with free backup and cleaning).
In a nutshell, we can say that the arrival of an SSD at the server brought new possibilities for increasing the productivity of the mass servers - for the cost of the storage of data and the intelligent configuration of disk I/O.
The disk subsystem of the "ideal server pid 1C" looks like this:
1. Tables of data bases placed on RAID 10 (or RAID 1 for small data bases) from higher server SSDs with a hardwired hardware RAID controller. With high IOPS rates, you can look at the PCIe SSD option. For a large database, it is efficient to use SSD-cache storage of HDD arrays. However, the 1C configuration and data structure, which are victorious, are not capable of up to IOPS, and the number of cores is small - to extract a traditional array from HDD SAS 15K rpm.
2. Index files for a swedish and inexpensive single SSD, TempDB - for 1-2 (RAID 1) SSD or RAMDrive.
3. SQL log files (but also 1C) to see the vision of the volume (the same physical disk or RAID-1) on SATA / NL SAS HDD or inexpensive SSD, or a logical disk on a RAID array, on which the operating system of the server server is installed files/folders.
4. The operating system that data is stored on RAID 1 with HDD or SSD.
As the IT infrastructure is virtualized, it’s very bad, that SQL Server was installed not like a virtual machine, but directly on a physical server, on the naked eye. The price of food is from 15 to 35% of the productivity of the disk subsystem (deposit from the possession, drivers, virtualization and connection methods). A virtualized SQL server environment connects volumes with database tables, index files, and TempDB to the VM in exclusive Direct Access mode.
When operating systems 1C: Enterprise 8 for small and medium enterprises (up to 100-150 active employees at the same time), the next step is to minimize spending on network operations via the Ethernet interface. Ideally - to serve SQL Server, and "1C: Enterprise 8 Add-on Server x64", and 1C sessions in Remote Desktop with one physical server. From the point of view of the safety of stability up to vіdm, such a recommendation allows you to increase the maximum of the possession of that PZ, and for the fluctuations in the virtualization of virtualization, it gives the same level of safety and "repetition of the medium" on the other setting.
How to turn off Ethernet from the SQL server -> 1C:Enterprise 8 add-on server -> 1C:Enterprise 8 session? Merezhevy Ethernet interface, with iogo packaging of these data is notably small blocks for transmission, will always create additional jamming: when packing / unpacking traffic, and when transmitting itself (high latency). In 1C:Pіdpriєmstvo 8 to collect a large array of data, they are transferred for processing, and that is carried out according to the entire language, in certain situations - on the wrong side. When transferring data directly to one process to another within the server's RAM (on one server without virtualization), or through a virtual mesh interface (within the same physical server, with good server mesh adapters with transferred RAM blocks between VM ) zatrymki richly lower. Modern double-processor servers with large operational memory and a disk subsystem on SSD allow you to comfortably serve a 1C database for 100-150 active cores.
As for the quest of a database, a number of physical hosts is inevitable, it is necessary to connect all servers via 10Gb Ethernet. Or, at least, 2-4 aggregation 1Gb Ethernet connections with hardware accelerated TCP / IP (TCP / IP Offloader) and hardware support for virtualization.
The biggest cost of productivity on Ethernet ports suffer from budgetary decisions. It's no secret that 1Gb wired adapters that are soldered on most server motherboards are not designed to serve intensive wired traffic. Navite yakscho on the board є 2 or 3 ports of GbE, the stink of the implementation on the desktop chips. Sufficient management, stinks give rise to additional bills of lading from the service of merging exchanges, especially in the virtualized middle. The whole process of transferring data through such a chip is covered by the processor resources, operational memory and internal bus. Such chips do not give any accelerated transmission of IP traffic, if such transmissions are received, the Ethernet packet will mean a transfer to the processor. In a virtualized environment, the productivity of an interface can reach 25-30%. At the very least, you can not mention the need for the most appropriate interface with the help of monitoring. The central processor is responsible for the new one, and if it doesn’t work, then it’s just standing in the window of the mesh card. Ports on desktop chips should be switched off from the data stream in virtualized environments, after deleting the server management task. Add a discrete border card on the server chipset for intensive fencing traffic.
Discussing the productivity of the servers in the future is always accompanied by super-cooks about their reliability. The safety of durability to maturity will always require additional vitrates, especially for the support of uninterrupted virobnic processes. Without belittling the role of that space 1C, it can be said that the biggest dilemma of “productivity / superiority” is growing in different areas: for the first time, the optimization of hardware solutions is fought, for the other - the organization of processes and procedures. If the programs are marked critical, the main respect for the implementation of practicality is given not to the needs of individual server protection, but to the minimization of simple infrastructure.
Зрозуміло, для підприємств із відносно великою кількістю одночасно підключених користувачів (25-150) та розміщенням усіх додатків на одному сервері обов'язкове застосування джерел безперебійного енергопостачання, надлишкових блоків живлення самих серверів, кошиків гарячої заміни дисків та RAID-масивів із гарячим резервуванням. However, the availability of equipment does not replace the planned reservation of the data themselves. Mayuchi schodenny (more precisely, shonochovy) backup that operational file from Full SQL log, you can re-update DB 1C for a remarkably short interval.
Permissible idle time of the central system 1C for small and medium enterprises is 1-2 accidents per month, 1-4 years for the duration. Indeed, a great reserve of time - as before the resurrection, we are preparing for a long time. Necessary smart restart and availability of images of all virtual and physical servers in front of the VM on a real file/volume - to update the infrastructure part itself on the backup server. Obov'azkovy schodenny backup (and sozhnevyy that closing period) for the next physical attachment and Full SQL log for vypadkіv, if data entry "from the beginning of the working day" is critical and important manually. For the obviousness of the administration, you can invest in 1-2 years for the renewal of the practice by a zagal, high and with less productivity. Well, and there, de needing the uninterrupted work of 24x7, the first tasks will be the choice of a different architecture, possessing a minimum number of points of introduction and modern technologies of clustering. Ale tse already zovsіm іnsha іstoriya.
Original article: http://ko.com.ua/proektirovanie_servera_pod_1s_66779
I will allow the editor of the journal "Computer Review"
For the cob, I’m going to see the script for the work:
1.) Working with a file base through a wild resource (web server)
2.) Work with the file base at the terminal
3.) Working with server (MSSQL) database
Robot from the file base through a global resource (web server)
Everything is easy here. Yakscho tse zvichaynі form that 1-3 koristuvach. Then on the "server" (the machine on which the base is located is chosen:
The last moment, as a base on the cured forms. The axis here is just like everything is built, as it is described more, you can see the galma. Prote out є:
As for the base, you will be able to locally work out one coristuvach that is sufficient for your comfortable work, and the speed of the work through a raw resource will be so self-sufficient. Ale and here є out - a robot through a web server. On the Internet, you can find a large number of articles that describe how to organize a job with a similar rank, not judging in this article on this one. Alone, I will share with you my warnings: it’s better to improve the robot with corestuvachiv not through a web browser, but through a thin client (if we add a new base to the list of IB, on the side of the IB placement there is an item "on a web server"). Tse, for my warnings, quicker, lower through the browser. In addition, when working through a browser, pardons are heard in the interface (what did the PM and so on), there are none when working through a thin client.
Vlasne, speeding up with this recipe (ssd, processor with a high frequency, web server, thin client). It is possible to expand the myth "as the number of koristuvachiv is greater than 1 (for the current version greater than 0 :)) - a server base is required *.
*If you want, it’s safe to say that the price is not UPP or the base size is > ~ 4GB, and the number of coristuvachs does not exceed 4 (the maximum size of the base is the number of coristuvachiv, if it’s a big one, it’s possible to zustrich the fall, if through a file base with a file base did more people work?
Working with a file base at the terminal
Let's move on to the next option. We may have a terminal server and may have a file base. Here everything is similar to scenario 1 for the processor's fault:
Work with server (MSSQL) base
This scenario is the most complicated and, perhaps, it will require some additional statistics. I propagate within the framework of this article to look at just the basic principles that should be added to productivity
Nache everything. Like food / skargi / propositions - wellcome in the comments;)
