hdparm shows poor performance on cached reads for PATA drive
Konstantin Svist
fry.kun at gmail.com
Mon Dec 17 02:40:34 UTC 2007
John Summerfield wrote:
> Konstantin Svist wrote:
>
>>>
>>> For general performance, cached performance is pretty irrelevant,
>>> you're still constrained to the bus speed at best, and the general
>>> ability of the hardware (ATA motherboard electronics included) in
>>> general.
>>>
>>> I don't have the foggiest idea of what an HP ze4400 is; depending on
>>> its age your speed of 28.55MB is pretty good.
>>>
>>
>> I agree that cached performance doesn't matter all that much,
>> normally. However, it's a pretty large drop in performance - and I'm
>> trying to figure out why it happened. I wouldn't be worried about it
>> if it were off 5-10% - but it's just about HALF of the old value.
>> Also (and I'm not sure about this...) this might indicate some
>> problem with latency (which *would* be pretty important, especially
>> if it's this much).
>
> Latency a problem on a laptop?
>
> One of the tradeoffs one accepts in choosing a laptop is that it's
> performance is going to be less than on less portable systems.
>
> For starters, disks mostly spin at 5400 rpm instead of the usual 7200
> on desktops and 10,000-15,000 on other systems.
>
> Latency might be important in systems doing lots of short reads; then
> seek time matters too. When loading multimegabyte programs and data
> files into RAM so they can be paged out, forget it.
In general, I agree - but, again, this is not a 10-20% decrease, it's
almost 50%.
Plus, while investigating this, I've noticed that CPU performance
suffers heavily, as well (~50%).
Not only that, but hdparm says the cached reads are at ~175MB/s when I
drop into runlevel 1.
There's definitely something there that makes it so slow (I'm guessing,
a daemon) - and it doesn't affect my newer laptop.
More information about the fedora-list
mailing list