Re: Network very very slow... was iSCSI and jumbo frames

[Paul Ripke <stix%stix.id.au@localhost>]

On Wed, Feb 03, 2021 at 08:25:59AM +0100, BERTRAND Jo?l wrote:
> Michael van Elst a ?crit?:
> > On Tue, Feb 02, 2021 at 03:55:10PM +0100, BERTRAND Jo?l wrote:
> >> Michael van Elst a ?crit?:
> >>> joel.bertrand%systella.fr@localhost (=?UTF-8?Q?BERTRAND_Jo=c3=abl?=) writes:
> >>>
> >>>> 	But jumbo frames and Intel adapter don't increase NAS throughput :
> >>>>              204Mb           407Mb           611Mb           814Mb
> >>>
> >>> Where does this information come from?
> >>
> >> 	iftop.
> > 
> > But with what kind of disk operation?
> 
> 	All operations : read, write, by dd or bacula-sd. If I understand,
> bacula only writes sequential data files. Filesystem on this iSCSI
> volume is ffs (v2) with log (and of course without sync or async options).
> 
> > iSCSI naturally has a higher latency than a direct attached disk.
> > To utilize a link you may need to generate more requests in parallel.
> > 
> > For writing that's not that difficult as the client will buffer enough
> > data. For reading you need large buffers or significant read-ahead to
> > trigger concurrent I/O.
> > 
> > You may also need to tune TCP/IP on the client.
> > 
> > You could start a test with reading from the raw iSCSI volume
> > with dd. You need a read buffer of at least 1MB to get the
> > maximal concurrency.
> > 
> >> [ ID] Interval           Transfer     Bitrate
> >> [  5]   0.00-10.00  sec  1.08 GBytes   931 Mbits/sec
> > 
> > That looks reasonable for a Gigabit interface.
> 
> 	Of course, but with dd, I only obtain :
> 
> legendre# dd if=/dev/zero of=/opt/bacula/test.dd count=10 bs=10m
> 10+0 records in
> 10+0 records out
> 104857600 bytes transferred in 0.537 secs (195265549 bytes/sec)
> legendre# dd if=/dev/zero of=/opt/bacula/test.dd count=10 bs=100m
> 10+0 records in
> 10+0 records out
> 1048576000 bytes transferred in 53.396 secs (19637725 bytes/sec)
> legendre# dd if=/dev/zero of=/opt/bacula/test.dd count=10 bs=1000m
> 10+0 records in
> 10+0 records out
> 10485760000 bytes transferred in 1026.927 secs (10210813 bytes/sec)
> legendre#

How much RAM does this system have? I'm guessing you might be paging in
those last two operations. Instead, try:

dd if=/dev/zero of=/opt/bacula/test.dd count=100 bs=1m
dd if=/dev/zero of=/opt/bacula/test.dd count=1000 bs=1m
dd if=/dev/zero of=/opt/bacula/test.dd count=10000 bs=1m

ie. keep block size down to a sane level.

> 186 MB/s for the first file. 18,7 MB/s for the second one and about 10
> MB/s fort the third one.
> 
> And I don't understand. If iSCSI target or raid6 subsystem on qNAP were
> the bottleneck, CPU load should be greater than 1 and it's not the case.
> Maybe I will try to add a SSD disk as cache (but I'm not sure that this
> NAS supports cache over USB3).
> 
> 	Best regards,
> 
> 	JKB

-- 
Paul Ripke
"Great minds discuss ideas, average minds discuss events, small minds
 discuss people."
-- Disputed: Often attributed to Eleanor Roosevelt. 1948.