Does anyone know why iaStor is so much better (as measured by Sysmark) than msahci?
LOL… Is that a serious question?
Are you asking why the Sysmark score is better (less latency, more device utilization, faster completing of a given request) – which you could determine with a little research? Or are you asking “how does Intel do achieve whatever-it-is-that-makes-it-faster”??
In any case, whatever the reason for the better Sysmark score, I suspect it has something to do with iaStore being written by a specific vendor who can optimize their own hardware… and MSAHCI being a generic driver that implements a well known standard.
Peter
OSR
I’m curious to see the numbers.
It is supposed to offer enhanced performance in particular when there are
multiple drives such as in a RAID conifguration.Also it allows ATA drives to
accept more than one command at a time and dynamically reorder the commands
for maximum efficiency (c).
While it might offer better performance, on a lot of systems it’s the driver
with the highest ISR count, highest DPC count as well as the highest DPC
routine execution time making it less suitable for latency sensitive systems
such as those handling real-time audio.
//Daniel
wrote in message news:xxxxx@ntdev…
> Does anyone know why iaStor is so much better (as measured by Sysmark)
> than msahci?
>
I don’t know if the question was THAT dumb. I assume Intel has no magic potion which makes their driver faster so I was wondering if anyone had ever looked in detail what exactly they do which gives them such an advantage. I am seeing a 10-15 point increase via Sysmark between msachci and iastor. I would like to study this myself with a bus analyzer but I have not done so yet. Has anyone else ever looked in to the details of what they do differently?
On what test (io access pattern) iaStor has an advantage? Is it sequential or random, small or large requests?
No offense intended, seriously. It just wasn’t clear what you were asking… In my experience, one sentence questions are rarely sufficiently detailed to evoke well thought out and detailed answers. Such questions force us to guess at what you meant… and then either risk spending time (a) answering what we THINK was the question, or (b) asking you further questions to clarify your query.
From what you said in follow up, it seems to be clear that you’re asking…
“I’ve noticed that I get a higher Sysmark result running iaStor than I do running MSAHCI using the same drives, the same SATA controller, and the same system and system configuration. Before I go off and try to explore the reasons for this higher score, I was wondering if anyone on the list has any clues. Given that the actual DISKS can perform any faster, is the higher level of performance attributable to (for example) higher bus utilization under load or …? I realize I can try to measure this and come to some independent conclusions, but was hoping to save myself the effort if somebody could enlighten me (either on or off list). No sense reinventing the wheel, right?”
Maybe I’m the only one, but I didn’t get that from “Does anyone know why iaStor is so much better (as measured by Sysmark) than msahci.”
Peter
OSR
P.S. BTW, given the above question, I’m sorry to say that MY answer at least is “no, I don’t know.” – I *do* know, from numerous anecdotal observations and those of my colleagues, that at least SOME versions of iaStor can contribute to significantly greater DPC latency (system wide). While I’ve not measured it, I suspect iaStor is spending more time in either its ISR or DPC than MSAHCI. Mr. Terhell alluded to this as well. This could make sense. One would expect MSAHCI to be a model citizen (though, we certainly know that not all MSFT storage controller drivers have been so in the past). Whereas, we might expect a vendor-proprietary driver to be more concerned about the performance of that device over the overall interrupt latency and/or performance of the system.
PGV
Yes I should have put in more detail. I am running Sysmark 2007 on an HP EliteBook 8440p and varying only the driver (same disk drive, same OS, same everything else), MSAHCI vs. iaStor and I see a huge (15 points with Sysmark is huge) performance difference running the same test.
I have it on my todo list to run a subset of Sysmark and capture bus traces to try and discern what is so different but it may be a little while before I can free up that much time. Just wanted to see if anyone else had looked into this is greater detail than just my observation.
Thanks for the input all.
wrote in message news:xxxxx@ntdev…
> … I assume Intel has no magic potion which makes their driver faster
They have tricks up their sleeve, for sure. See this page:
http://www.intel.com/network/connectivity/vtc_ioat.htm
Some bits released as open source:
http://www.intel.com/technology/quickdata/whitepapers/sw_guide_linux.pdf
Not sure if any of these are in the Stor driver; there may be something
else.
– pa
interesting stuff, thanks for the links.