Re: softraid crypto performance on Sun Fire T1000

2016-11-07 Thread Alexander Bochmann
Hi, ...on Sat, Oct 29, 2016 at 03:06:05PM +0200, Jonathan Schleifer wrote: > While a single core of the T1000 is quite slow, this just seems too slow, > making this setup unusable. openssl speed shows 10 MB/s for AES-128-CBC and 7 > MB/s for AES-256-CBC on a single core. So a single core is

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Jonathan Schleifer
Ah, thanks. I was just about to destroy my RAID-1 and see it that makes a difference :). So, the difference is pretty much the kernel locking: The fewer cores, the better the performance. But this still means that the softraid crypto performance is way below what openssl speed gives. I wonder if

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Stefan Sperling
On Sat, Oct 29, 2016 at 07:53:06PM +0200, Stefan Sperling wrote: > > Are you sure that LDOM was indeed using softraid crypto? > > Yes. Uhm, but the dd command wasn't :-) (the guest's root disk is sd2, not sd0...) Now our numbers align much better: # dd if=/dev/rsd2c of=/dev/null bs=10m

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Stefan Sperling
On Sat, Oct 29, 2016 at 07:39:29PM +0200, Jonathan Schleifer wrote: > > I have the 1GHz version with 4 cores (32 threads). > > Ok, so same per-core speed, so single-threaded performance should be the same. > (Btw, you have 8 cores, not 4. 8 cores @ 4 threads each.) > > > Otherwise it's probably

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Jonathan Schleifer
> I have the 1GHz version with 4 cores (32 threads). Ok, so same per-core speed, so single-threaded performance should be the same. (Btw, you have 8 cores, not 4. 8 cores @ 4 threads each.) > Otherwise it's probably similar to yours. > It's running 6.0 at the moment, yes. Some guests are running

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Stefan Sperling
On Sat, Oct 29, 2016 at 06:57:00PM +0200, Jonathan Schleifer wrote: > Oh, wow, these are *much* better than what I get. Which CPU do you have? I > have 6x 1 GHz (meaning 24 threads). Are you running 6.0? > > Thank you for these numbers, they make me much more hopeful about this > machine. I

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Jonathan Schleifer
Am 29.10.2016 um 18:34 schrieb Stefan Sperling : > On Sat, Oct 29, 2016 at 06:08:37PM +0200, Jonathan Schleifer wrote: >> Hm, my main problem seems to be that whenever I decrypt something from the >> disk, all other 23 cores seem to get stalled. >> >> So, would you recommend doing

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Stefan Sperling
On Sat, Oct 29, 2016 at 06:08:37PM +0200, Jonathan Schleifer wrote: > Hm, my main problem seems to be that whenever I decrypt something from the > disk, all other 23 cores seem to get stalled. > > So, would you recommend doing the following then: > > * Have a partition for the main system on a

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Jonathan Schleifer
Hi, > I run a T1000 which is segregated into a couple of LDOM guests (about 10). > Some of the guests use softraid crypto inside. The host does not. Yeah, I was planning on using LDOMs as well. However, since I wanted to put this into a datacenter (for a cheap price, so it not being the most

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Stefan Sperling
On Sat, Oct 29, 2016 at 05:12:51PM +0200, Jonathan Schleifer wrote: > Another thing I noticed: > > When running dd if=/dev/zero of=foo bs=65536, my SSH connection gets extremely > laggy. If I open 4 more in parallel, all go down to KB/s of writes, and SSH > becomes unusable. Now unusable as in

Re: softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Jonathan Schleifer
Another thing I noticed: When running dd if=/dev/zero of=foo bs=65536, my SSH connection gets extremely laggy. If I open 4 more in parallel, all go down to KB/s of writes, and SSH becomes unusable. Now unusable as in things need forever to start. Unusable as in I press a key and it takes forever

softraid crypto performance on Sun Fire T1000

2016-10-29 Thread Jonathan Schleifer
Hi! I just installed OpenBSD 6.0 on my Sun Fire T1000 (with 2 SAS HDs in a hardware RAID 1 that I set up from OpenBoot). However, I only get read rates of less than 2 MB/s from sd1a (the softraid), but reads from sd0d (the underlying partition of the softraid) get magnitudes more. While a single

Re: softraid crypto performance

2009-11-11 Thread Jan Stary
On Nov 10 23:36:42, Michael wrote: Hi, Am 10.11.2009 22:53, schrieb Jan Stary: Those 40 MB/s are limited due to the other systems read performance. However, softraid crypto seems (unreasonably ?) slow to me. Why do you even involve the network and some other system's read in

Re: softraid crypto performance

2009-11-11 Thread Jan Stary
On Nov 10 16:21:04, Alvaro Mantilla Gimenez wrote: On Tue, 2009-11-10 at 21:31 +0100, Michael wrote: Hi, when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without crypto, can do way

Re: softraid crypto performance

2009-11-11 Thread Jacob Yocom-Piatt
Jan Stary wrote: On Nov 10 16:21:04, Alvaro Mantilla Gimenez wrote: On Tue, 2009-11-10 at 21:31 +0100, Michael wrote: Hi, when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without

Re: softraid crypto performance

2009-11-11 Thread Marco Peereboom
Bonnie is not a realistic load, ever. Therefore the numberis are really not useful. If one insists on getting an idea of what crypto can run then do: dd if=/dev/rsd2c of=/dev/null bs=1m count=100 Where rsd2c is the raw crypto disk. At some point I will have another look to see if I can speed it

Re: softraid crypto performance

2009-11-11 Thread Jan Stary
when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without crypto, can do way more. Uh...that sounds wear to me. I just copy 70 Gb from a USB SATA HD to the local partitions under

Re: softraid crypto performance

2009-11-11 Thread Jan Stary
On Nov 11 07:09:36, Marco Peereboom wrote: Bonnie is not a realistic load, ever. Therefore the numberis are really not useful. If one insists on getting an idea of what crypto can run then do: dd if=/dev/rsd2c of=/dev/null bs=1m count=100 Where rsd2c is the raw crypto disk. This is how my

Re: softraid crypto performance

2009-11-11 Thread Robert
Warning, long post... This whole performance discussion made me curious, and since I had 2 identical SATA disks available I made some tests with softraid and crypto (dmesg at the end). summary (MB/sec): write readlayout 80 80 single disk / raw 68 80 single disk / file

Re: softraid crypto performance

2009-11-11 Thread Aaron Poffenberger
, it looks like my server's softraid crypto performance is pretty good, especially when taking into account that softraid(4) and the crypto discpline are very new to OpenBSD. One curious result is that my read speeds are lower than write. I was expecting reads to be a lot faster since the crypto

Re: softraid crypto performance

2009-11-11 Thread Marco Peereboom
You defeated read ahead with your RAID test and found out that kernel crypto is slow. Oh and you tested softdep pretty well too because that is what you were trying to measure, right? Lies, damn lies and benchmarks... On Wed, Nov 11, 2009 at 08:09:40PM +0100, Robert wrote: Warning, long

Re: softraid crypto performance

2009-11-11 Thread Robert
*) Softdep was put there on purpose, because this is most probably how the mount will be done by most of the users. *) My intention was simply to find out how those combinations will work on my system in the way that I would (from my knowledge so far) configure them - any advice on how to

Re: softraid crypto performance

2009-11-11 Thread Kenneth R Westerback
On Wed, Nov 11, 2009 at 02:30:25PM -0600, Marco Peereboom wrote: You defeated read ahead with your RAID test and found out that kernel crypto is slow. Oh and you tested softdep pretty well too because that is what you were trying to measure, right? Lies, damn lies and benchmarks... Lies,

softraid crypto performance

2009-11-10 Thread Michael
Hi, when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without crypto, can do way more. What I see during transfer in top/systat is a high interrupt load, however the interrupt load when writing

Re: softraid crypto performance

2009-11-10 Thread Jan Stary
On Nov 10 21:31:56, Michael wrote: Hi, when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without crypto, can do way more. What I see during transfer in top/systat is a high interrupt

Re: softraid crypto performance

2009-11-10 Thread Alvaro Mantilla Gimenez
On Tue, 2009-11-10 at 21:31 +0100, Michael wrote: Hi, when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without crypto, can do way more. Uh...that sounds wear to me. I just copy 70 Gb

Re: softraid crypto performance

2009-11-10 Thread Michael
Hi, Am 10.11.2009 22:53, schrieb Jan Stary: Those 40 MB/s are limited due to the other systems read performance. However, softraid crypto seems (unreasonably ?) slow to me. Why do you even involve the network and some other system's read in evaluationg your softraid's write? Maybe, because

Re: softraid crypto performance

2009-11-10 Thread Michael
Hi, Am 10.11.2009 23:21, schrieb Alvaro Mantilla Gimenez: On Tue, 2009-11-10 at 21:31 +0100, Michael wrote: when using softraid crypto with OpenBSD 4.6-current I never get more than ~10-11 MB/s disk writing speed even though the disk (WD Raptor 73 GB) itself, without crypto, can do way more.