Operating system and server requirments

That is a difficult question since RepliStor can be configured to use memory in many ways. There are three basic parts to the replication "engine" of RepliStor (RS). First there is a Control Service which is a process which allows you to stop and start the other processes remotely. This process is very small and is not used a lot in general practice but is handy for controlling the RS Server Service. It takes about 1.5MB and has a Working Set of about 7MB.

The RS Server Service is the main UserMode workhorse of RS. Counting the linked modules it takes up about 16MB and has a Working Set of about 112MB in an average installation. The last of the non-GUI parts of RS is the Filter Driver, RepliStor.sys. The driver takes up a very small portion of Kernel mode memory but is perhaps the busiest of the three since it has to monitor the file system and it tells the RS Server Service which file operations need to be replicated. All told the driver would use less than 1-5MB of memory but it is using the Kernel memory space which is very small to begin with. Windows Kernel Memory is split into paged and non-paged pools and together would average about 2-300 MB of machine memory. The RS driver primarily uses non-paged pool memory and during heavy use will request and release memory as needed and the amount is governed by the OS.

There is one more process that uses memory in the traditional way. The RS GUI or console uses about 20MB of memory but is not always open.

That takes us to a different type of memory that comprises the bulk of the physical and virtual memory that is used by RS. That is called cache memory. There are two different types of cache memory used: KernelCache memory and Kernel Log memory.

The KernelCache is physical (RAM) memory that is set at the time RS is installed (configurable later) and would generally be in the 100-300MB range depending on the amount of memory in the computer to begin with. In operation RS watches for file operations that are contained within the user defined area of the local file system disk(s). This is configured by creating specifications. If a file operation is included in the specified portion of the local storage the RS driver will add it to the KernelCache.rdf (a file) as well as the KernelCache memory (RAM) allocated. This is called a split write and this is the queue of file operations that need to be sent to the Target node or nodes. The KernelCache is a finite size and if operations happen too fast these operations will overflow into Kernel Log files called OC$ files. Under normal conditions there may be no Kernel logs or there could be thousands. In RS 6.4.x each of these Kernel logs is about 100MB of file space.

So, how much memory does RS take? A typical installation on a machine with say 2GB of RAM would be 2-500MB of RAM and as much free disk space as required to cache the ongoing file operations. That is on a 32bit typical system. If you have a 64bit Windows 2008, R2 machine with 32GB of RAM then RS could be configured to take a bit more than 4GB of RAM. General wisdom indicates that the more KernelCache memory you have the faster the operations can be processed.

Up to a point that is true but remember that the speed of the connections between nodes and the speed of the file operations being written to disk greatly affect the processing of the operations and if RS cannot keep up with the transactions in KernelCache memory it will create Kernel logs. These are processed more slowly because of Disk I/O and other overhead.

There is no absolute answer to your question. I hope the above will help and maybe give you some general guidelines.

Many thanks jschrade, this was exactly what I have been look for.

Regards

Gary Mitchell

Infrastructure Tech Analyst

EMEA TI Regional Wintel Server Support

Citi

•Belfast: 028 90 409305

•gary.walter.mitchell@citi.com