2008 aio application bandwidth blade bladesystem c3000 csUriStem dedicated disposal download edition email error Exchange express filter google hosting hp IIS IIS7 isapi iscsi KB linux log managed manufacturing memory microsoft mpio multipath oil performance proliant quota recovery release server SQL sqlserver storage studio system target togo web windows yahoo

Diagnosing memory issues with memtest86

So you got a nice new piece of hardware that you need to deploy at colocation facility or in business critical environment. But before you do you want to make sure that all your components are 100% in working order. We will discuss testing memory in this article.

There is a project called Memtest86 which tests PC based memory. How do you get it? You can download ISO (bootable CD) image from memtest86.com, or if you have any of the linux distribution flavors including CentOS, Knoppix and Fedora, you already have this functionality at your fingertip. When booting memtest from dedicated CD, you do not need to do anything, if you boot from linux CD, on initial prompt you need to type memtest or memtest86 depending on your distribution. As soon as memtest start, you can remove CD and if you wish, you can pop it into another piece of hardware and start test there.

Scroll down for additional info labeled What does memtest do and how does it do it

More info: What does memtest do and how does it do it

permanent link to article http://freecash.hogger.net/diagnosing_memory_issues_with_memtest86

At first, memtest does a barrage of tests on small memory location. When that suceeds, it relocates itself and then tests the rest of the memory. The tests are very lenghty as all caching by CPUs is purposely disabled (testing with CPU caches enabled will mask errors in actual memory bits as CPU will remember what it wrote to most recent locations and would not normally re-read from memory chips.) The tests include ECC checking if available on the system. During tests, a number of random and predefined patterns are tested that should easily detect bitwise problems.

Checking each memory region still may not be complete enough if multiple addresses are incorrectly handled by same area on chip (fused or insensitive bit address line) and for that reason, the addressing test is performed. During addressing test, if two different areas are mapped to same bits on chip, only one of the reads will respond properly and you will see long list of red errors appearing. For best result you must let at least one full pass to complete which can easily take overnight with 4GB or more of installed memory.