Hi Julian,
I would suggest to rewords you initial comments to some thing like:
"apt was heavy reworked and optimised, probably your initial problem won't hit
you as hard as before.
Hashing optimisation can't currently be done for this release without braking
the ABI."
Any way, thank you for
Of course you don't have a CPU bottleneck with curl, it does not run 3
or 4 hashes over everything it downloads. APT needs to hash its
downloads to ensure they are secure, and it uses all available hashes to
do so, so if one hash's security is compromised, we can still hopefully
still rely on the
This seems to be a bit of a bikeshed on specific implementation options.
The problem here is downloads pin the CPU, and in my personal experience
I can download the same files via curl with nothing close to a CPU
bottleneck.
I do not know how you benchmarked APT but this may be a case of a micro
Tests run (for SHA256, add SHA512 as needed):
(1) time for i in md5 sha1 sha256; do nettle-hash -a $i < /tmp/test ; done
(2) time for i in md5 sha1 sha256; do openssl $i < /tmp/test ; done
(3) ./a.out < /tmp/test
where a.out is
#include
#include
int main(void)
{
Hashes h(Hashes::MD5SUM
New results (user time only, removing read() overhead):
OpenSSL: 6.8s or 9.8s = 147 or 102 MB/s = 1176 or 816 Mbit/s
Nettle: 8.8s or 11.8s = 113 MB/s or 85 MB/s = 904 or 680 Mbit/s
APT: 10.9s or 15s = 92 MB/s or 67 MB/s = 736 or 536 Mbit/s
As we can see, the improvement is not that
I think my laptop is too old to support AVX, though. But I'm not sure if
libnettle supports that anyway, and libnettle would be our choice.
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to apt in Ubuntu.
We can't use openssl for legal reasons, and also don't want it in
libapt-pkg for technical reasons anyway.
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to apt in Ubuntu.
https://bugs.launchpad.net/bugs/1123272
Title:
high
Results from my 5 year old laptop (X230, Core i5-3320M with 2.6 GHz),
for a 1 GB file:
MD5+SHA1+SHA256: 11s => 90 MB/s => 727 Mbit/s
MD5+SHA1+SHA256+SHA512: 15s => 67 MB/s => 533 Mbit/s
That seems like an acceptable baseline performance. You are not likely
to hit GBs worth of packages, so you
I am not satisfied with this response: I experience CPU bottlenecking on
ec2 server instances. Gigabit networking with CPUs sub-3ghz is
incredibly common place (read: entirety of server ecosystem), so I'm not
sure what you mean by "usual" network bandwidths.
--
You received this bug notification
Well, definition of "extreamly" is wired. In this bug report we have i5 with
20%!
I my case it was Intel Atom with 100% CPU usage and DSL 2000 Mbit/s!
Atom is slow, but not slowest used CPU.
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which
The hash algorithms should be fast enough these days in software to
handle usual network bandwidths. If you combine extremely high speeds
with extremely slow CPUs, that's of course suboptimal.
We can consider using libnettle at some point, I assume this might give
a nice speed up, but this is
** Tags removed: raring vivid
** Tags added: xenial
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to apt in Ubuntu.
https://bugs.launchpad.net/bugs/1123272
Title:
high cpu usage on download (not optimised SHA256, SHA512,
I believe this is still present in 16.04. Specifically I become CPU
bottlenecked (100% on one core by /usr/lib/apt/methods/http) when
downloading packages within AWS EC2 instances (better bandwidth with
download servers there). Besides the issues around locking up a core,
this makes several
** Changed in: apt (Ubuntu)
Importance: Undecided = Medium
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to apt in Ubuntu.
https://bugs.launchpad.net/bugs/1123272
Title:
high cpu usage on download (not optimised SHA256,
also get that issue on Vivid i386; mainly when the index is rebuilt when
using synaptic
** Tags added: vivid
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to apt in Ubuntu.
https://bugs.launchpad.net/bugs/1123272
Title:
Status changed to 'Confirmed' because the bug affects multiple users.
** Changed in: apt (Ubuntu)
Status: New = Confirmed
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to apt in Ubuntu.
16 matches
Mail list logo