Performance requirements
The Ipesoft D2000 real-time application server can be used for a whole range of applications - from small SCADA systems (built on the Raspberry PI platform or on industrial computers with OS Windows/Linux) to large MES/EMS type systems with dozens of users, large application databases (storing terabytes of data), multi-terabyte archives and depository databases. In this chapter, we will try to summarize some best practices for designing and managing D2000 systems.
Virtualization
Today, the D2000 application server is often also operated in a virtualized environment (VMware, Hyper-V, Proxmox), especially for systems such as MES, EMS, SELT, and balance systems, less frequently for SCADA systems.
In this environment, resources are shared, resulting in two basic sharing problems:
allocation of sufficient resources
monitoring and diagnostics
The "hyper-converged architecture", in which we use powerful servers with local disks (in a RAID10 or RAID50 disk array), has proven to be successful. It is ideal when the virtualized D2000 applications have their own servers (which are not shared with other applications) and when the administrators of the D2000 applications not only have administrative rights to the virtual machines on which the D2000 is running but also have access to the virtualization environment so that they can perform operational performance diagnostics in case of problems.
In the case of several administrators (e.g. for network infrastructure and firewalls, virtualization, Active Directory, and application servers), we recommend the introduction of an operation log in which all operations that could affect the functionality of the servers will be recorded, in particular:
configuration changes in the network infrastructure
configuration changes in virtualization, adding virtual servers, changes in allocation and limitation of resources (CPU, RAM)
moving virtual servers, moving their disks
configuration changes in disk arrays, adding additional guests
configuration changes in the AD policy
changes in antivirus and antimalware software settings
software installations and upgrades
firmware upgrades (servers, switches, firewalls)
changes in server settings in BIOS (performance, security, other)
The subsequent analysis (usually if there is a slowdown and reduction in performance) is greatly facilitated by the existence of the operation log. To share the log, you can use e.g. SVN or GIT repository, SharePoint repository, and the like.
Allocation of sufficient resources in a virtualized environment
The resources that the D2000 primarily needs are three: memory (RAM), CPU, and disk space (we haven't experienced any problems with bandwidth limits on network interfaces yet). We recommend:
RAM - allocation of a sufficient amount of memory. It is ideal to reserve memory in a virtual environment so that memory is not shared between virtual machines (so-called ballooning).
Small D2000 applications need roughly 1 GB of RAM (the minimum for an application server on Windows/Linux is 4-8 GB), and large ones need several GB to tens of GB, depending on the number of configured objects, processes, and users. If more memory is available, we recommend allocating it to the SQL database for the archive (we recommend PostgreSQL) and the archive cache (we recommend several GB for the so-called isochronous cache). Applications using EDA technology (Energy Databank) benefit from several GB of memory allocated to the EDA server.CPU - the usage of CPU strongly depends on the nature of the application (constant CPU consumption for SCADA-type applications, significant peaks for balance systems or systems where user-triggered events take place - e.g. preparation of documents for monthly invoicing). In the case of physical servers, today's processors have enough necessary performance. In a virtualized environment, we encountered a case where VMware administrators artificially limited the maximum usable frequency for the balance system, because they thought it "consumed too much CPU". First, they caused significant user dissatisfaction (the preparation of invoicing documents took several hours instead of 30 minutes) and, on the one hand, the analysis showed that a significant part of the performance was consumed by the antivirus (ESET NOD) since it did not have configured exceptions.
For large applications, it is advisable to allocate more vCPUs (4-8-16) - the D2000 architecture allows good parallelization (parallel tasks within the D2000 Kernel, D2000 Event, and D2000 Archive processes [configurable]).Disk space:
for the partition with the OS, we recommend approx. 20 GB (Linux) or 100 GB (Windows)
for the partition with D2000, we recommend at least 40 GB for the start, while the archive database usually has the largest consumption - several GB to several TB (for SELT systems, the monitoring database - up to tens of GB)
if depository databases are enabled in the D2000 Archive (storage of historical data with unlimited depth), we recommend a separate partition for the depository databases (the size of the depository databases will gradually grow on it). Currently, there are customers with more than 20 TB depository databases, but newer versions of the D2000 allow you to turn on depository data compression, which usually has a compression ratio of 1:10 and better, which significantly saves disk space.
For the partition with the OS and D2000, we recommend fast disks (SSD), for the partition with depositories, slower HDDs or NAS are also sufficient.
Monitoring and diagnostics in a virtualized environment
In a virtualized environment, it is essential to have access to monitor operational parameters to ensure that the D2000 application does not suffer from resource sharing. We recommend that environment administrators monitor and be able to provide the following data (according to the graphs available in vCenter):
graphs of the CPU load of D2000 servers, other virtual servers within the host, and the total CPU load of the host (to diagnose if there is a lack of CPU power)
graphs of RAM consumption (proof that there is no ballooning - memory sharing between servers when there is a lack of RAM and subsequent swapping)
graphs of I/O subsystem load (metrics: reads/writes per second, read/write data amounts [kB/s], read/write latencies) - again, for D2000 servers, other virtual servers within the guest, and if the storage in question is shared, then a load of all hosts using disk storage (to diagnose whether there is a lack of I/O performance). Some shared storages provide their own load diagnostics from individual guests that can be used (LeftHand, 3PAR).
We recommend having all these graphs and data for them available for at least 3 months, for long-term performance monitoring.
In a virtualized environment, both the speed and latency of the disks are important for the D2000 Archive. It should be noted that when archiving, hundreds and thousands of database tables for individual archive objects are written in parallel.
Note: for older D2000 installations, we recommend increasing the PostgreSQL ODBC parameter BatchSize (starting with PostgreSQL ODBC version 12.02) from the default value of 100 to 10000 - the change can speed up interval calculations (RECALC) and the INSERTARCHARR action.
Graphics performance issues on virtualized Windows Server 2025
In case of reduced graphics performance (e.g. D2000 HI process) after migrating to virtualized Windows Server 2025, the following applies:
Within Windows Server 2025, 2 possible ways to increase display performance in a VM (virtual machine) are supported:
GPU Partitioning (GPU-P) – NEW - GPU performance is divided into "parts" and each such "partition" can then be assigned to a VM.
Discrete Device Assignment (DDA) - This is a classic "GPU Passthrough", where you assign the entire physical card exclusively to one VM - we recommend enabling this mode.
In both cases, we assume that the GDI api uses GPU hardware in the background. Details can be found here:
Troubleshooting guidance for Hyper-V GPU assignment, partitioning, and passthrough - Windows Server
In the case of using RDP, it is also interesting to consider the following: The problem with slow rendering of text and GDI elements via RDP on Windows Server 2025 is often caused by the system using the Microsoft Basic Render Driver (software rendering) by default instead of the physical GPU.
Forcing GDI acceleration via HOST-level registers: we recommend enabling.
GDI applications are often limited by the "remote driver". Try adding/editing this registry value on the server to force hardware rendering:Path: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Windows
Key (DWORD): DisableHWAcceleration (make sure it is set to 0)
Activating H.264/AVC 444 for plain text in GPO at HOST level: - we recommend enabling it. Details in the figure below.
Just for text rendering (GDI), enable H.264/AVC 444 mode in GPO. This mode prevents font blurring during image compression, which often looks like "slow redrawing", and uses the hardware encoder in your Xeon (Intel QuickSync).Enabling "Hardware Rendering" in GPO at HOST level: - we recommend enabling it. Details in the figure below.
For integrated Xeon chips, it is crucial to enable the "Use hardware graphics adapter for all Remote Desktop Services sessions" policy in the Policy Editor (gpedit.msc). Without this setting, the Xeon iGPU rests during RDP connections, and the processor does all the calculations (in software), which causes the aforementioned slowness.
Path in the Policy Editor:
Computer Configuration
→ Administrative Templates
→ Windows Components
→ Remote Desktop Services
→ Remote Desktop Session Host
→ Remote Session Environment
Details can be found here: