| Author |
Message |
Oliver S.
Guest
|
 Posted:
Sun Oct 23, 2005 8:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | Andy designs processors for a living, and was a central
architect for the P6.
|
Even experts sometimes say things that aren't 100% correct. |
|
| Back to top |
|
 |
Bill Todd
Guest
|
| Posted:
Sun Oct 23, 2005 8:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
Oliver S. wrote:
| Quote: | Andy designs processors for a living, and was a central
architect for the P6.
Even experts sometimes say things that aren't 100% correct.
|
And even fools sometimes say things that are. But betting on the fools
is on average not a winning strategy.
- bill |
|
| Back to top |
|
|
Guest
|
| Posted:
Mon Oct 24, 2005 12:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | AFAICT, Hoard requires at least two atomic operations and storeload membars
for every "uncontended" allocation? IMHO, this is unnecessary overhead...
|
Well, that's the way it was described in the original Hoard paper
(ASPLOS 2000), but as of version 3, Hoard uses per-thread local heaps
and so requires no atomic actions for most allocations and frees.
-- emery
--
Emery Berger
Assistant Professor
Dept. of Computer Science
University of Massachusetts, Amherst
www.cs.umass.edu/~emery |
|
| Back to top |
|
|
Chris Thomasson
Guest
|
| Posted:
Mon Oct 24, 2005 4:22 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | Well, that's the way it was described in the original Hoard paper
(ASPLOS 2000), but as of version 3, Hoard uses per-thread local heaps
and so requires no atomic actions for most allocations and frees.
|
That is good news. Now it kind of sounds like Oliver is basically
reinventing Hoard...
;) |
|
| Back to top |
|
|
Oliver S.
Guest
|
| Posted:
Mon Oct 24, 2005 8:14 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | That is good news. Now it kind of sounds like Oliver is basically
reinventing Hoard...
|
Do you think this feature mainly comprises HOARD or a whole other
allocator? |
|
| Back to top |
|
|
Andy Glew
Guest
|
| Posted:
Mon Oct 24, 2005 11:23 pm Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
Bill Todd <billtodd@metrocast.net> writes:
| Quote: | Oliver S. wrote:
Andy designs processors for a living, and was a central
architect for the P6.
Even experts sometimes say things that aren't 100% correct.
And even fools sometimes say things that are. But betting on the
fools is on average not a winning strategy.
- bill
|
Sounds like somebody was dissing me - probably in response to my post
about more complete hash functions likely impacting the L1 cache latency.
Unfortunately, the dissing post is not visible on my newsserver,
although the anti-dissing posts are. Was it cancelled? Can anyone
share it with me?
--
No worries: I agree about questioning authority and expertise, even
mine. Plus, if you have a solution for high speed cache indexing
using a more complete index function for the L1, I'd like to hear
about it.
But, I suspect that physics gets in the way, unless you are doing
tricks like indexing the cache not with parts of the actual address,
but with parts of data values used to form the address. I've gone
down that path, I think it can be made to work - but it is a lot more
complex than a virtually indexed cache, and virtual indexing is too
complex for many people (at Intel and AMD). |
|
| Back to top |
|
|
Kai Harrekilde-Petersen
Guest
|
| Posted:
Tue Oct 25, 2005 12:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
Andy Glew <andy.glew@intel.com> writes:
| Quote: | Bill Todd <billtodd@metrocast.net> writes:
Oliver S. wrote:
Andy designs processors for a living, and was a central
architect for the P6.
Even experts sometimes say things that aren't 100% correct.
And even fools sometimes say things that are. But betting on the
fools is on average not a winning strategy.
- bill
Sounds like somebody was dissing me - probably in response to my post
about more complete hash functions likely impacting the L1 cache latency.
Unfortunately, the dissing post is not visible on my newsserver,
although the anti-dissing posts are. Was it cancelled? Can anyone
share it with me?
|
There's not much sharing. Oliver's second post was quoted in full (a
single line) by Bill.
Oliver's first response to you was:
| Quote: | * yes, just XORing in more address bits *will* cost that much more time
Sure that this will heighten a critical path so that this will cost an
additional clock-cycle (although I believe that XORing won't help at all)?
I think that it's very likely that this won't do that!
|
Kai
--
Kai Harrekilde-Petersen <khp(at)harrekilde(dot)dk> |
|
| Back to top |
|
|
Chris Thomasson
Guest
|
| Posted:
Tue Oct 25, 2005 12:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
"Andy Glew" <andy.glew@intel.com> wrote in message
news:peypacgydbaa.fsf@pxpl2829.amr.corp.intel.com...
| Quote: | Bill Todd <billtodd@metrocast.net> writes:
Oliver S. wrote:
Andy designs processors for a living, and was a central
architect for the P6.
Even experts sometimes say things that aren't 100% correct.
And even fools sometimes say things that are. But betting on the
fools is on average not a winning strategy.
- bill
Sounds like somebody was dissing me - probably in response to my post
about more complete hash functions likely impacting the L1 cache latency.
Unfortunately, the dissing post is not visible on my newsserver,
although the anti-dissing posts are. Was it cancelled? Can anyone
share it with me?
|
Oliver seems to make frequent use of X-No-Archive: Yes. |
|
| Back to top |
|
|
Niels Jørgen Kruse
Guest
|
| Posted:
Tue Oct 25, 2005 3:22 pm Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
Andy Glew <andy.glew@intel.com> wrote:
| Quote: | Sounds like somebody was dissing me - probably in response to my post
about more complete hash functions likely impacting the L1 cache latency.
Unfortunately, the dissing post is not visible on my newsserver,
although the anti-dissing posts are. Was it cancelled? Can anyone
share it with me?
|
On the newsserver I am using, it was the other way around: your post was
missing but the followup was not. Both are present on groups.google.com
though.
--
Mvh./Regards, Niels Jørgen Kruse, Vanløse, Denmark |
|
| Back to top |
|
|
Alexander Terekhov
Guest
|
| Posted:
Tue Oct 25, 2005 4:15 pm Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
Niels Jørgen Kruse wrote:
| Quote: |
Andy Glew <andy.glew@intel.com> wrote:
Sounds like somebody was dissing me - probably in response to my post
about more complete hash functions likely impacting the L1 cache latency.
Unfortunately, the dissing post is not visible on my newsserver,
although the anti-dissing posts are. Was it cancelled? Can anyone
share it with me?
On the newsserver I am using, it was the other way around: your post was
missing but the followup was not. Both are present on groups.google.com
though.
|
Don't know what happened with Glew's post on the newsserver you're
using, but S.'s recent posts on google will disappear really soon.
http://groups.google.com/support/bin/answer.py?answer=7918
regards,
alexander. |
|
| Back to top |
|
|
Niels Jørgen Kruse
Guest
|
| Posted:
Tue Oct 25, 2005 4:15 pm Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
Alexander Terekhov <terekhov@web.de> wrote:
I don't see any X-No-Archive headers in these posts.
--
Mvh./Regards, Niels Jørgen Kruse, Vanløse, Denmark |
|
| Back to top |
|
|
Oliver S.
Guest
|
| Posted:
Sat Nov 05, 2005 1:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | That is good news. Now it kind of sounds like Oliver is basically
reinventing Hoard...
|
I'm currently switching from the HOARD-strategy to a 60% PHKmalloc-clone
with some cool locality-improvements; I bet that the resulting allocator
will be among the top allocators. |
|
| Back to top |
|
|
Oliver S.
Guest
|
| Posted:
Sat Nov 05, 2005 1:15 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | Don't know what happened with Glew's post on the newsserver you're
using, but S.'s recent posts on google will disappear really soon.
|
No, they won't; but I think I'll batch google-groups-deletes via wget
the next time. *g* |
|
| Back to top |
|
|
Chris Thomasson
Guest
|
| Posted:
Mon Nov 07, 2005 1:57 am Post subject:
Re: Micro-Benchmark for Memory-Allocators |
|
|
| Quote: | I'm currently switching from the HOARD-strategy to a 60% PHKmalloc-clone
with some cool locality-improvements; I bet that the resulting allocator
will be among the top allocators.
|
Looking forward to taking a look at your project. |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in