The Raven Report

Ok so now we know what to look for how to we look for it?

Tools

Task Manager

Task manager can quickly provide general memory information. On the performance tab, you can quickly determine how much physical memory is in the system, how much physical memory is available, Kernel memory usage and how much memory the VMM has promised to running processes. With this information you can determine if further investigation is required.

The processes tab also has a wealth of information, but first you have to enable additional columns via the View à Select Columns option. On this tab you can quickly determine the amount of memory a process is using and how much of the paged & nonpaged pool is being used.  There many addition columns available, so don’t settle for the default view.

Process Explorer does everything task manager does and more. You can really dig into a process and figure exactly what it’s doing and what kind of resources it’s consuming.

Get it here:  http://technet.microsoft.com/en-us/sysinternals/bb896653.aspx

Poolmon is a great tool for tracking down kernel memory leaks or just determining what is consuming kernel resources .

More info here: http://support.microsoft.com/kb/177415  
& here: http://technet2.microsoft.com/windowsserver/en/library/0d302498-c947-4655-95af-719ae75acfb51033.mspx?mfr=true

Performance Counters

Memory\Available Mbytes (MB):  Indicates the amount of physical memory (in MB) immediately available for allocation to a process or for system use. Used with the Pages/Sec to determine if the system is low on physical memory.

Memory\Pages/sec: Pages/sec is the number of pages read from the disk or written to the disk to resolve memory references to pages that were not in memory at the time of the reference (hard page faults), and the number of pages that were redirected to memory outside of the working set.  This counter also includes paging traffic on behalf of the system Cache to access file data for applications, so a large number alone is not reason for concern just further investigation. This is the primary counter to observe if you are concerned about excessive memory thrashing.

Memory\Pool Nonpaged Bytes: This counter shows the current size of the Nonpaged pool in Bytes. Anything over 200MB should be investigated, and unless memory tuning options are in place (/3GB) then anything over 100MB.

Memory\Pool Paged Bytes: This counter shows the current size of the paged pool in Bytes. In general anything over ~250MB should be investigated unless memory tuning options are in place (/3GB) then anything over 150MB.

Memory\Free System Page Table Entries: This counter shows the number of page table entries that are available for use. Anything under 8,000 needs a closer look.  If the system is getting low consider using the /userva=XXXX with something like 3030 to start to bump these up.  

Memory: Cache Bytes: This is the size of the system cache. You can use this counter with the Pool nonpaged/paged & PTEs to determine who kernel memory is being allocated.

Registry Keys 

As with all registry modifications, use caution. Under normal circumstances you should not need  to modify these keys, but if you are running into kernel memory issues proper use of these keys can resolve the problem.

NonPagedPoolLimit: Specifies the maximum amount of system VA space that can be used by the nonpaged pool.

PagedPoolLimit: Specifies the maximum amount of system VA space that can be used by the paged pool.

SystemCacheLimit: Specifies the maximum amount of system VA space that can be used by the system cache.

SystemPtesLimit: Specifies the maximum amount of system VA space that can be used by I/O mappings and other resources that consume system page table entries (PTEs).

SessionSpaceLimit: Specifies the maximum amount of system VA space that can be used by session space allocations

PoolUsageMaximum: Determines the Max Poll usage before pruning starts.  The default is 80%. A value of 40-60 provides a workable solution to some paged pool exhaustion issues.

Windows 2003 64Bit can support a theoretical limitation of 16 exabytes or 2⁶⁴

·         Kernel Mode Address space - 8 TB

·         Paged Pool – 128GB

·         NonPaged Pool – 75% of RAM up to a maximum of 128 GB.

·         System Cache/PTEs - 1 TB regardless of physical memory.

As you can see, with 64bit Windows all of the previous memory constraints are blow out of the water.   For just this reason alone upgrading to a 64bit OS should be seriously considered.

Troubleshooting 32bit memory problems

For those who have legacy applications and 32bit hardware upgrading may not be an option. So in the following I will go over some tips and tools to identify memory bottlenecks a 32bit system.

Physical Memory Exhaustion - This is pretty much the straight forward if you are out of physical memory and the system is excessively paging to disk. More physical memory is most likely the answer.

Kernel Memory Exhaustion -  Memory problems in  kernel memory are a little more challenging to nail down.  Here are some of the signs:

Paged Pool & Nonpaged Pool

·         Sluggish or unresponsive user interface

·         Server has message or client processing failures

o   RPC Service Unavailable

o   Not enough Storage to process this command

·         Pool allocation failures:

System PTEs

·         Server fails to respond to I/O requests

·         Server fails to respond to network requests

·         Server has message or client processing failures

If you are experiencing these problems or other unexplained issues kernel memory could be your issue.

Your first stop should be the performance tab in Task Manager. From here you can quickly gauge if you are approaching the kernel memory limits of the system. Take a look at the Kernel Memory section and look at the amounts used by the page pool ( > 250MB or > 150 with /3GB ) and nonpaged pool( >200 or >100 with /3GB).  If we come up higher than these numbers further investigation is required.   We’ll start there next time.

Since a 32 bit OS can only access 4GB of memory, some magic needs to happen to allow the system to see additional memory.  Say hello to Physical Address Extension (PAE).  PAE allows the operating system itself to address more that 4GB of RAM. Each Process is still bound by the 4GB limits, but the VMM can utilize the all the Physical memory up to 64GB. How does PAE do it?  Starting with the Pentium Pro processor the x86 platform actually allows 36bits of address space, so  2³⁶ bytes  = 64GB addressable. PAE is automatically enabled on computers running Windows Server 2003 with Service Pack 1 (SP1) and Windows XP with Service Pack 2 (SP2) when DEP is enabled on a computer with a processor that supports the no-execute page protection feature.

The 2GB dividing line between user memory and kernel memory is completely arbitrary. 4GB memory tuning (4GT) is enabled by putting a /3GB switch in the boot.ini. Using the /3GB switch allocates 1 GB to the kernel and 3 GB to the User-mode space. But be warned, we are stealing from Peter to pay Paul here, and since Nonpaged Pool, Paged Pool & PTEs all reside in Kernel memory,  If the memory reduction in the pools is too great in a specific server installation, the server or the applications may generate an error or appear to stop responding.

You shouldn’t use 4GB Tuning in the following scenarios. The system has more than 16GB of memory, the server is a Terminal server, large file servers and Exchange are good examples when the /3GB alone does more harm than good

Exchange?? What do you mean Exchange? All the docs say to use the /3GB switch. What gives? Please welcome /userva to the party. The /userva=xxxx switch is designed to allow for more precise tuning of User-mode address space for applications that require more than 2 GB of User-mode space but do not require all the space that is provided by the /3GB tuning switch alone. Use the /userva switch with the /3GB switch in the Boot.ini file to tune the User-mode space to a value between 2 and 3 gigabytes (GB), with the difference (3,072 less xxxx) being returned to Kernel mode.

OK, so now the OS can see more than 4GB of Memory, but my application is still limited to 2-3GB of memory.  How can my applications use more memory?

As stated earlier by using PAE we give the OS the ability to address memory over the 4GB mark. Windows 2003 Enterprise and Datacenter Editions make use of the x86 36bit memories addressing for applications with an API called Address Windowing Extensions (AWE). The AWE API allows an application to transfer memory pages above the 4GB limit into the addressable memory area where it can make changes to the pages. This permits applications such as SQL Server that use large datasets to manage them in RAM rather than a slow paging file. AWE locks physical memory address space to an application. No other application can use this memory and it can’t be pages to disk.

What’s new in Vista and Server 2008 32bit?

Let say we have a Windows 2003 Server running Terminal Server  with the following Kernel memory utilization:

–      Paged pool: 50%

–      Non-paged pool: 70%

–      System PTEs: 98%

Even though there is still fair amount of memory in the paged & nonpaged pools, essentially this system is loaded to its maximum capacity because the System PTEs are fully exhausted. 32bit versions of Windows have relatively low maximum values for these memory areas, that all need to be allocated out of the 1-2GB of kernel virtual memory space. Up until Vista this allocation was only done at boot up.

 In Windows Vista/2008, kernel virtual address space is dynamically allocated. The sizes and locations of the paged pool, nonpaged pool, System Cache & PTEs are no longer fixed, but are dynamically adjusted according to the current operational requirements. This simple but effective change resolves the single pool exhaustion issue described above.  Now, not only can Peter steal from Paul, but Paul can steal back from Peter.

Virtual memory is the memory that applications use and can be much greater than the Physical memory installed.

Ok, so virtual memory isn’t limited by physical memory then how does it work? The Virtual Memory Manager (VMM) is responsible for mapping each virtual address into a corresponding physical address. The VMM performs the mapping by dividing the RAM into fixed-size page frames, creating system page table entries (PTEs) to store information about these page frames, and mapping them. VMM also manages kernel resources, such as the paged pool, nonpaged pools and the system cache.

User Mode Memory is the address space where all nonsystem (kernel) processes and applications run. Each process has its own virtual memory address space up to the maximum allowable on the OS Architecture. If you have 10 user mode process running on a 32bit windows system each will have its own 2GB address space

Kernel Memory, unlike the User Mode Address space, shares the same address space (2GB by default on a 32bit system), and all kernel allocations share the same memory space.

Kernel memory consists of four main areas:
     Paged Pool consists of virtual memory addresses that can be paged to disk.
     Nonpaged pool consists of virtual memory addresses that are guaranteed to reside in physical memory as long as the corresponding kernel objects are allocated.
     System Cache which is used to cache disk I/O processes for faster file access.
     System Page Table Entries (PTE)

A little more on PTEs, every process that runs on a 32bit Windows system has 4GB  of virtual address space for its use. Of this, the upper 2 GB address (kernel) is common to all processes running in the system. The lower region is called user address space. System PTEs are small chucks of kernel memory that are used to map virtual address space to physical memory. From the process perspective, each element of virtual address maps to a byte of physical memory. It is the responsibility of the Virtual Memory Manager to do this translation, and each PTE represents a page, and contains information necessary for the VMM to locate that page.

That's it for now. We'll get into the 32bit architecture in part 2