Search results

I was doing some consulting work for a small VPS provider (who isn't on this forum, as far as I know), they wanted a custom kernel module written to log performance metrics. This was to be installed on the host node, so it could audit things like to see how 'busy' a vm was from the host node...

It wasn't like this when I was an employee with them. 

I do not understand your question; There is no error, I am just saying generally that 'dd' benchmarks on non dedicated hardware is silly because things like interrupts and timers are emulated from the host to the container(s) downstream

My only issue is the 'dd' benchmark; Benchmarking on VPS's with 'dd' is silly because the containers do not have direct access to hardware, the host node does some mmap() magic and other tricks plus some emulation of interrupts to<->from the containers.

Is buyvm reliable? I've never owned a VPS before outside of something I could at least 'control' ..

HAMMER v2 is coming and it's going to be really great once some more commits get pushed to git. 

Looking for US based Open/FreeBSD based VPS somewhere on the east coast, with IPv4 and IPv6 (please for the love of god no 6in4). Something small for web development and other things, not too expensive, please. 

Route reflectors, easier to use MEDs/prefix lists based on default zone (big cities) Just remember, SOME ASN's used to be OLD companies that were bought (adelphia <-> comcast)

sync is unnecessary, see vm.c in the latest kernel.. 3 is also unnecessary, a 1 will do, but the best way is to remount your filesystem so the entire page cache VFS layer is nuked..

Benchmarking with dd is silly for the following reason:   VFS cache. 

As a former CC employee, That is very shitty of you to post people's families on a thread that have nothing to do with the people in the original posting.

Benchmarking VPS's is utterly silly, interrupts are emulated, containers dont have access to some privileged cpu instructions (and they are emulated), timers are emulated from the host.. The only true way to benchmark a 'VPS' is to run stuff on the host node, when it's under load.. 

Like 'dd' is going to tell you anything...

I don't use VPS's at all, simply because I do not trust anyone enough with my data; not only that, I do not like the idea of not having control over ring0 :)

Try building squid without epoll to see if that helps, it should fallback to using select()..  I think there is a command line option to switch this on the fly, but I can't remember.

Most of it is probably syscall overhead from read(), and write(), to socket calls. You should attach strace to the process with the '-c' arguement and let it run for a few minutes, then see what it's spending time doing the most (if it's sycall load) and then optimize from there.

You should stay away from such tools, they will /never/ give you accurate results, ever. On virtualized environments, you'll get unusual results that do not match up with reality. Timers within virtualized environments have to be emulated due to containers not having physical access to hardware...

I've been using *BSD since 4.3 Tahoe (1980s), and I use it primarily for my development environment; I also use DFly with HAMMER (which is comparable to ZFS in some aspects)

You could also get creative and use your video card memory as a VFS backed filesystem.  I have one of them in production for squid, and it works /really/ well for small subsets of data (mostly html and others), plus I use it for mysql tmpdir :-)

Then don't.