Face off part two: Windows vs Linux real world RAM and disk tests

David M Williams
Sunday, 27 July 2008 20:44

Opinion and Analysis

The results are conclusive. All browsers increased in memory use slightly over time but the Firefox 3 slope is closest to 0, ie a largely consistent amount of memory usage.

Although not relevant in the Internet Explorer vs Firefox argument, the peak amount of memory used by Firefox 3 is a lower value then the concluding size of Firefox 2, and this does demonstrate the work that went into memory management in Firefox 3.

However what is extremely relevant is that Internet Explorer 7 simply did not appear to give any memory back whatsoever, even when all windows were closed!

Additionally, as I found last week, Internet Explorer 7 ends up about 400MB larger than Firefox 3 - but this time on Windows Vista. This has several important ramifications.

Firstly, Firefox 3 clearly handles memory the same way irrespective of platform and demonstrates the same pattern on both Windows and Linux.

More significantly though, Internet Explorer’s memory usage is clearly a matter of its own devising. It is not a feature of Windows Vista that results in a missing 400MB. If it were this ought to be the case with Firefox.

This graph visually depicts Pavlov's results with these three browsers running on Windows Vista and their memory usage over time.



Pavlov explains that his group put in a lot of work to achieve the result they did. Perhaps Microsoft’s developers could take a peek. One of the actions they performed was to reduce the amount of memory allocations performed in the program. The thinking behind this was that long running applications – like a web browser – can waste a lot of space because the RAM fragments (just like you find fragmentation happening on a hard drive) due to variable sized amounts of memory being allocated and deallocated over time. Mozilla managed to reduce the number of allocations they used and also investigated which of the available routines to allocate memory fared best. The allocator they chose was jemalloc and they then worked with the developer to ensure ports were available for Windows, MacOS and Linux, permitting the same code to be used for all platform builds of the browser. Pavlov claims that switching to jealloc gave an immediate 22% drop in memory usage on Windows Vista.

Work was also expended to tuning caches, adjusting how compressed picture formats are stored, better detection and breaking of cycles – where two objects in memory refer to each other, thus keeping themselves both alive even when not required – and other items.

I believe the results are crystal clear. Linux behaves with a stability that Microsoft Windows cannot reproduce. It performs admirably under differing amounts of available RAM and it makes best use of RAM before turning to slower disk drives. The default web browser under Windows – Internet Explorer – visibly takes up ever increasing amounts of memory without releasing it. This is in sharp contrast with the default web browser under Linux – Firefox – which maintains a reasonably flat level of memory usage and then relinquishes it when terminating.

This is not due to Windows Vista trying to optimise performance, but because Internet Explorer’s development team do not appear to have demonstrated the same rigour that Mozilla put into practice.