[TangerineSDR] Benchmarking Digital RF (HDF5)

[Phil Erickson]

Hi Bill,

  FYI, our software radar at MIT Haystack (Ettus USRP N210 hardware as the
source; GbE interface, running 25 complex megasamples / second at 16 bit
depth) goes to a recorder agent that formats to Digital RF.  This streams
about 191 MB/sec of HDF5 files produced by the Digital RF API to a ring
buffer inside a ramdisk, so going directly to memory.  A subsequent agent
processes power spectra with heavy reduction to a conventional spinning
disk set (RAID 6, I believe) with a GlusterFS** cloud overlaid.  This all
uses workstation/general purpose computing on Ubuntu platforms.  Note that
we directly run a cable between the USRP and the computer and do not have
it pass through any switches/routers.

  So I guess I don't have a data point because our RAM disk buffers any
slowdowns until we get it to spinning media.  We use a ~100 GB ram buffer
which provides probably 230 seconds of realtime soak per channel (x 2
channels for us).  I know that we tried writing the 191 MB/second to
spinning disks and it does not work - one needs SSDs at that rate.
Probably worth trying SSDs in your testing.

  Aside: we do find that writing continuously to spinning media makes it
die faster than normal since the manufacturers do not guarantee performance
with continual write unless you buy Enterprise/server grade hardware  - 10x
more expensive, naturally.  This is in line with Google's server farm
experience using JBOD or equivalent systems.

73
Phil W1PJE

** "GlusterFS is a scalable network filesystem suitable for data-intensive
tasks such as cloud storage and media streaming."  https://www.gluster.org

On Wed, May 29, 2019 at 10:08 AM Engelke, Bill via TangerineSDR <
tangerinesdr at lists.tapr.org> wrote:

> Benchmarking Digital_rf on Odroid
>
>
>
> There have been some deep concerns about the ability of a SBC to keep up
> with saving spectrum data using HDF5 (here using the MIT Digital_rf
> package, which uses HDF5 internally). I installed Digital_rf on my Odroid
> XU4 and ran the benchmarks that are included in the Examples folder.
>
>
>
> Writing to MicroSd card in the XU4 (SanDisk Ultra Plus, claimed speed 80
> MB/sec), benchmark-reported speeds
>
> 1.                     Writing raw binary data directly
>                                    133.36  MB/sec
>
> 2.                     Writing “simple single write” in HDF5
>                                 154.14 MB/sec
>
> 3.                     Writing “simple single write” in HDF5 with
> checksum        71.29 MB/sec
>
> 4.                     Writing      “        “    “  with checksum +
> compression L1        8.01 MB/sec
>
> Unexpected results here:
>
> a. HDF5 "simple write" is faster than writing raw binary
>
> b. top two speeds are faster than the claimed speed of the MicroSD card.
> Maybe the benchmark is measuring a burst rate during just the HDF5
> processing (?) Maybe there is some caching  or buffering going on.
>
>
>
> Writing to Western Digital USB 3 (spinning) external 4TB hard drive (incl.
> 64 MB internal cache) – benchmark-reported speeds
>
> 1.                     Writing raw binary data directly
>                                    103.66  MB/sec
>
> 2.                     Writing “simple single write” in HDF5
>                        96.04 MB/sec
>
> 3.                     Writing “simple single write” in HDF5 with
> checksum          55.02 MB/sec
>
> 4.                     Writing      “        “    “  with checksum +
> compression L1        7.67 MB/sec
>
> These results make a bit more sense than the MicroSD result. I can't find
> a claimed transfer speed for this hard drive on Western Digital's site.
>
>
>
> IF these results are representative (a big “if”) – then even a 4 TB
> spinning hard drive will be able to keep up with the PSWS data write rate.
> If anyone can confirm, reproduce, add to, or refute these results, I am
> most interested to know…
>
>
>
> -73-
>
> W. D. Engelke (Bill), Asst. Research Engr. AB4EJ
>
> Center for Advanced Public Safety
>
> Cyber Hall
>
> The University of Alabama
>
> Tuscaloosa, AL 35487
>
> Desk: (205) 348-7244
>
> Mobile: (205) 764-3099
>
>
> --
> TangerineSDR mailing list
> TangerineSDR at lists.tapr.org
> http://lists.tapr.org/mailman/listinfo/tangerinesdr_lists.tapr.org
>


-- 
----
Phil Erickson
phil.erickson at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.tapr.org/pipermail/tangerinesdr_lists.tapr.org/attachments/20190529/2b67525d/attachment-0001.html>