That Next Step Is A Doozy...

Memory prices have dived through the floor recently, with DDR2-800 modules going for less than $25 for 1GB and $50 for 2GB.  (That's a single 2GB module, too.)  Usually when you see the low-end modules getting very cheap there's a high-end module that's still commanding a high price.  So I was wondering where the 4GB modules were.

Here they are.  At $581.99, I don't think they'll be selling too many into the SOHO market just yet.  Indeed, since the only reason you'd get them is to install more than 8GB of memory, it would be far cheaper to buy a server motherboard with 8 sockets and use registered memory.

Guess I'll have to make do with 8GB for now.

The reason I need all that memory is so that I can do away with a separate Linux server for my development environment, running the development and test environments under VMWare instead.  Which lets me do fun things like taking a snapshot of the entire server or create a new server on the fly to test replication, but does eat memory like candy.

Some testing today, though, showed that Linux under VMWare was only delivering half the performance of the same software off a Live CD.*  But then some more testing showed that it actually delivers closer to 75%, and my first test must have been screwed up somehow - probably because another process was running and hyperthreading** was interfering with accurate performance results.  But then, running configure for a Python build is enormously slower on my VMWare/Fedora 8 system than on my native Fedora 4 system, something I can't explain.

It's not all bad news, though.  After installing Fedora 8, the mouse pointer automatically tracked between Windows and Linux even without installing VMWare tools (a good thing, because I haven't been able to install VMWare tools).  And the clock problems that made standard Linux 2.6 builds all but useless under VMWare have completely disappeared.

Boring benchmarks follow.

* After which my PC refused to reboot for an hour, which was fun.
** My current desktop is a P4 2.6.
I've benchmarked the virtual systems on my Windows box against my Fedora Core 4 system.  I couldn't do a full set of benchmarks using the Live CD as it doesn't include any compilers.

First, let's configure the Python 2.5.1 source for a compile:

Fedora 8/VMWare

real    6m34.558s
user    0m53.891s
sys     5m19.352s

Fedora 4 Native

real    0m21.141s
user    0m12.305s
sys     0m8.716s

The Fedora 4 box is a little faster hardware-wise, an Athlon 64 3200+ vs. a Pentium 4 2.6, but that doesn't make up for a factor of 4 difference in user time.  (I'm assuming that the factor of 40 difference in system time is a horrible artifact of VMWare.)

I'll run a make on both sides now, and compare that:

Fedora 8/VMWare

real    7m8.188s
user    2m31.381s
sys     4m17.288s

Fedora 4 Native

real    1m40.223s
user    1m32.029s
sys     0m6.978s

Not too bad on the user time, but still the factor of 40 on system time.  Yuck.

Next, make install:

Fedora 8/VMware

real    2m2.966s
user    0m28.682s
sys     1m7.204s

Fedora 4 Native

real    0m15.004s
user    0m9.843s
sys     0m3.010s

And finally, we can run the benchmark with a custom-built Python:

Fedora 8/VMware

Loop: 3.007
String: 3.977
Scan: 0.940
Total: 7.923
 
real    0m24.595s
user    0m23.777s
sys     0m0.196s

Fedora 4 Native

Loop: 1.790
String: 1.987
Scan: 1.340
Total: 5.117
 
real    0m15.397s
user    0m15.364s
sys     0m0.007s

Which is a little slow, but not terrible.  It seems that process creation is killing me, so as long as I don't do that, I'll be fine...

I tried VirtualBox as well.  It's quite nice for a free virtualisation product, and it reported better times than VMWare on my Python benchmark.  But a quick check against a stopwatch indicated that this was because its clock was running about 20% slow.  Apart from that, it exhibits the same high system time as VMWare.

Posted by: Pixy Misa at 12:32 AM | Comments (3) | Add Comment | Trackbacks (Suck)
Post contains 640 words, total size 5 kb.