[ale] Performance issue

[Jim Kinney]

I need to check on both of the expanders. I recall LSI soc but that was a
while ago.

Sector and block are complicated. Wound up using 512k at the raid level and
multiples for lvm and xfs. In the mix is Luks and it seems to be agnostic.

I do have a partition I let the system decide block size on and it also is
crap.

What the systems (yes. There are two) appear to do is read a block, wait,
write a block, wait, read, wait, write, wait... If there was a bad sector
alignment clash, I would expect constant but slow rw.

On Aug 21, 2016 8:50 PM, "LnxGnome" <lnxgnome at hopnet.net> wrote:

> Math says you should get your 12Gbps+ out of your 40TB array, and 35Gpbs+
> from the 104TB array, assuming 7.2K drives maxing at 170MB/s internal
> transfer.
>
> What kind of expanders are you using?
> Sector/block alignment?  IRQ conflicts?
>
> What's the single drive performance like?  other RAID configs?
>
> On 8/20/16 5:16 PM, Jim Kinney wrote:
>
> All,
>
> Getting DISMAL performance from new hardware.
>
> Twin 6-core Xeon E5-2630L v.2 @2.40 GHz
> 128GB DDR4 RAM
> LSI megaraid 2108 in a PCIe v3 slot
> Onboard LSI 3108
> Drives are all 6Gbps SAS2 4TB
> 2 arrays, RAID 6, one is 40Tb, other 104TB
>
> Everything hardware says I should expect a minimum through put of 12Gbps
> on either array.
>
> The max I'm getting from iotop is 784 Mbps.
>
> W. T. F!!!
>
> According to top, the system is basically idle. Ditto from iostat and
> every other tool I check. I'm doing a 'cp -a' from/to same array.
>
> Yes. New location is Luks encrypted. That process is totally asleep it's
> getting so little work. The rng is busy (dev/random is not running out
> (watch -n 1 cat /proc/sys/kernel/random/entropy_avail was always above
> 2048)) but not overloaded or asleep.
>
> Tested going unencrypted to unencrypted with similar performance.
>
> Changed tuned-adm to latency-performance from balanced and the read
> portion of the cp went to 0 bps for 20 seconds until I switched it back to
> balanced mode. So that was a fail. It improved greatly using
> throughout-performance mode (bursts of 250MB/s) but was still showing long
> periods of 0 writes. Load went up from a paltry 8 to 22 so it's working
> more.
>
> But top shows wa values across multiple cores hitting 100 so it's hitting
> a wall somewhere.
>
> Any ideas of more places to look?
>
>
>
>
> --
> Sent from my Android device with K-9 Mail. Please excuse my brevity.
>
> _______________________________________________
> Ale mailing listAle at ale.orghttp://mail.ale.org/mailman/listinfo/ale
> See JOBS, ANNOUNCE and SCHOOLS lists athttp://mail.ale.org/mailman/listinfo
>
>
>
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://mail.ale.org/mailman/listinfo/ale
> See JOBS, ANNOUNCE and SCHOOLS lists at
> http://mail.ale.org/mailman/listinfo
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ale.org/pipermail/ale/attachments/20160821/e8364aa9/attachment.html>