Brian Twichell wrote:
Hugh Dickins wrote:
Let me say (while perhaps others are still reading) that I'm seriously
wondering whether you should actually restrict your shared pagetable
work
to the hugetlb case. I realize that would be a disappointing limitation
to you, and would remove the 25%/50% improvement cases, leaving only the
3%/4% last-ounce-of-performance cases.
But it's worrying me a lot that these complications to core mm code will
_almost_ never apply to the majority of users, will get little testing
outside of specialist setups. I'd feel safer to remove that "almost",
and consign shared pagetables to the hugetlb ghetto, if that would
indeed remove their handling from the common code paths. (Whereas,
if we didn't have hugetlb, I would be arguing strongly for shared pts.)
Hi,
In the case of x86-64, if pagetable sharing for small pages was
eliminated, we'd lose more than the 27-33% throughput improvement
observed when the bufferpools are in small pages. We'd also lose a
significant chunk of the 3% improvement observed when the bufferpools
are in hugepages. This occurs because there is still small page
pagetable sharing being achieved, minimally for database text, when
the bufferpools are in hugepages. The performance counters indicated
that ITLB and DTLB page walks were reduced by 28% and 10%,
respectively, in the x86-64/hugepage case.
Aside, can you just enlighten me as to how TLB misses are improved on
x86-64? As far as
I knew, it doesn't have ASIDs so I wouldn't have thought it could share
TLBs anyway...
But I'm not up to scratch with modern implementations.
To be clear, all measurements discussed in my post were performed with
kernels config'ed to share pagetables for both small pages and hugepages.
If we had to choose between pagetable sharing for small pages and
hugepages, we would be in favor of retaining pagetable sharing for
small pages. That is where the discernable benefit is for customers
that run with "out-of-the-box" settings. Also, there is still some
benefit there on x86-64 for customers that use hugepages for the
bufferpools.
Of course if it was free performance then we'd want it. The downsides
are that it
is a significant complexity for a pretty small (3%) performance gain for
your apparent
target workload, which is pretty uncommon among all Linux users.
Ignoring the complexity, it is still not free. Sharing data across
processes adds to
synchronisation overhead and hurts scalability. Some of these page fault
scalability
scenarios have shown to be important enough that we have introduced
complexity _there_.
And it seems customers running "out-of-the-box" settings really want to
start using
hugepages if they're interested in getting the most performance
possible, no?
---
Send instant messages to your online friends http://au.messenger.yahoo.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
