Re: Performance issues in 64-bit platforms
On Tue, Mar 20, 2007 at 02:17:27PM +0100, Dimitris Lampridis wrote:
> Hi everybody,
>
> although the following question is not debian-specific ( at least I hope so
> ), I'm posting it here, betting on the 64-bit architecture knowledge of the
> list.
>
> I have two 64-bit systems, both running Debian testing. One is a Core2 Duo
> laptop@2GHz, running a 32-bit Debian, the other is an AMD dual core 3800+
> desktop computer, running 64-bit Debian.
>
> I'm benchmarking both systems and I've run "hdparm -tT" on both. Now, the
> laptop has a typical laptop disk, namely a TOSHIBA MK1234GSX. The desktop on
> the other hand, has two Western Digital Raptors, 80G each, in software
> RAID0, and a slower/bigger typical desktop disk, namely a Seagate
> ST3160812AS, that is sitting outside the RAID configuration.
>
> The outputs of "hdparm -tT" are as follows:
>
> -----------------------------------------------------------------------------------------------------------
> 1 - LAPTOP:
> /dev/sda:
> Timing cached reads: 5788 MB in 1.99 seconds = 2901.82 MB/sec
> Timing buffered disk reads: 102 MB in 3.02 seconds = 33.78 MB/sec
33MB/s sounds right for a decent laptop drive.
Very nice ram speed on that (cached reads are a memory speed test, not a
disk speed test).
> 2 - DESKTOP WD raptors in RAID0 (root partition):
> /dev/md2:
> Timing cached reads: 1920 MB in 2.00 seconds = 960.95 MB/sec
> Timing buffered disk reads: 410 MB in 3.01 seconds = 136.14 MB/sec
136MB/s is rather nice for a raid0 of two drives each running 68MB/s.
Might be hitting a PCI bus limit or a controller limit at that level.
The memory performance is actually pathetic for an AMD. Are you running
single channel by any chance? Dual channel memory should get about
twice that (at least my old single core 3500+ with DDR400 does gets
1700MB/s or so). Make sure you are using a pair of identical memory
modules and that they are installed into each channel. Using crap
DDR2-533 ram is just a bad idea of course since that will really reduce
memory performance even further. But you certainly want to be using
dual channel.
> 3 - DESKTOP single WD raptor:
> /dev/sdb:
> Timing cached reads: 1938 MB in 2.00 seconds = 969.44 MB/sec
> Timing buffered disk reads: 206 MB in 3.02 seconds = 68.24 MB/sec
>
> 4 - DESKTOP single Seagate:
> /dev/sda:
> Timing cached reads: 1926 MB in 2.00 seconds = 963.75 MB/sec
> Timing buffered disk reads: 208 MB in 3.01 seconds = 69.02 MB/sec
> -----------------------------------------------------------------------------------------------------------
>
>
> Now, off to the questions:
>
> 1) Why are cached reads on the Laptop so much faster (case 1 vs. 2,3 and
> 4)??? The laptop has 1G of memory (probably a cheap one as well, to keep the
> cost low), the desktop has 2G of memory, fastest one my money could buy...
It sounds like dual channel on the laptop and single channel on the
desktop. If the desktop isn't the new socket AM2 and using plain DDR
ram, then your numbers make sense for single channel memory (they are
absolutely wrong if you have two channels of memory).
> 2) Shouldn't the "notorious" WD raptor greatly outperform the Seagate when
> comparing single disk performance (case 3 vs. 4)? After all, the WD costs 2
> times the price of the Seagate (or was it even more..?).
On seek time and random acces, yes. On raw transfer rate no. The
density and hence the amount of data moving past the head in a given
amount of time is not great. After all if you run half the density at
twice the speed, you still get the same amount of data going past the
head.
--
Len Sorensen
Reply to: