benefit of H2+ 2.5G Ethernet port

Post Reply
Ellicott
Posts: 16
Joined: Thu Apr 16, 2020 2:23 pm
languages_spoken: english
Has thanked: 4 times
Been thanked: 0
Contact:

benefit of H2+ 2.5G Ethernet port

Post by Ellicott »

My FiOS provider only offer about 800 Mbps Internet speed. How your guys take advantage of 2.5Gbps Ethernet port? Given that most NAS reading throughput is about 250 Mbps except a few NAS have 450 Mbps, how a local NAS can benefit from 2.5G Ethernet port?

I wonder,

PigLover
Posts: 50
Joined: Wed Nov 15, 2017 3:55 am
languages_spoken: english
ODROIDs: C2, XU4, HC1, MC1
Has thanked: 6 times
Been thanked: 7 times
Contact:

Re: benefit of H2+ 2.5G Ethernet port

Post by PigLover »

There are lots of uses for it. For example, change your NAS assumptions to use multiple SATA SSDs rather than spinning disks and 2.5gbps is rather easy to consume. Even easier if reading/writing to an NVMe M.2 drive (even with the PCIe 2.0 limitations of this board).
These users thanked the author PigLover for the post:
Ellicott (Thu Jun 25, 2020 1:04 pm)

domih
Posts: 330
Joined: Mon Feb 11, 2019 4:48 pm
languages_spoken: English, French
ODROIDs: UX4, HC2, N2, H2, C4, H2+
Location: San Francisco Bay Area
Has thanked: 114 times
Been thanked: 123 times
Contact:

Re: benefit of H2+ 2.5G Ethernet port

Post by domih »

There are various ways to bump disk I/O for leveraging network I/O:

(a) As mentioned above you can use SSD(s) (relatively economical), M.2 NMVe (more expensive). You can use the eMMC for the OS and thus reserve the full SSD or M.2 NMVe for storage.

(b) If you need a lot of a storage use for instance a 10TB HDD + SSD or M.2 NMVe as cache (Linux dm-cache, Linux bcache, etc...) I'm sure there is an equivalent for Windows.

(c) IMHO and the most efficient (and economical) is to run Samba and/or NFS in async mode. This means a file writing returns on the client BEFORE the writing is completed on the destination disk. In such a case the more memory (in dual channel for best speed) the better for the I/O buffers. H2+ can take as much as 32GB @ 2400MT/sec of memory. Put the H2+ behind a UPS to prevent data loss in case of power outage occurring during the few seconds or 1 or 2 minutes of file writing going on after the network operation itself completed. You can combine (c) with (a) or (b).

During some testing using 10 GbE I could see up to close to 8 times faster copies compared to 1 GbE. So using (c) I'm pretty sure you can see up to 2.5 times faster. 2.5 GbE on an H2+ is much less demanding than 10 GbE. Although this has to be tested. On my side, I'll do so in the next few weeks. Waiting for hardware right now.
These users thanked the author domih for the post:
Ellicott (Thu Jun 25, 2020 1:03 pm)

mad_ady
Posts: 8840
Joined: Wed Jul 15, 2015 5:00 pm
languages_spoken: english
ODROIDs: XU4, C1+, C2, C4, N1, N2, H2, Go, Go Advance
Location: Bucharest, Romania
Has thanked: 587 times
Been thanked: 531 times
Contact:

Re: benefit of H2+ 2.5G Ethernet port

Post by mad_ady »

Samba is async by default https://www.systutorials.com/is-samba-s ... or-writes/. Or do you mean other tweaks?

domih
Posts: 330
Joined: Mon Feb 11, 2019 4:48 pm
languages_spoken: English, French
ODROIDs: UX4, HC2, N2, H2, C4, H2+
Location: San Francisco Bay Area
Has thanked: 114 times
Been thanked: 123 times
Contact:

Re: benefit of H2+ 2.5G Ethernet port

Post by domih »

mad_ady wrote:
Thu Jun 25, 2020 1:33 pm
Samba is async by default https://www.systutorials.com/is-samba-s ... or-writes/. Or do you mean other tweaks?
Yes the original source referred by the link is from 2008. However, this deserves a second look because since 4.7 the default for strict sync has been changed from no to yes.

See https://wiki.samba.org/index.php/Samba_ ... ed/changed and look up strict sync.

<<...
The "strict sync" global parameter has been changed from a default of "no" to "yes". This means smbd will by default obey client requests to synchronize unwritten data in operating system buffers safely onto disk. This is a safer default setting for modern SMB1/2/3 clients.
...>>

Code: Select all

domih@ripper:~$ testparm -s -v | grep 'strict sync'
.../...

	strict sync = Yes
sync always default is no. So it depends on the value of FILE_FLAG_WRITE_THROUGH from the client.

My guess is that the "...This is a safer default setting for modern SMB1/2/3 clients..." or "...smbd writes to memory buffer, blocks until buffer contents are written to disk, which results in poor performance, but better data integrity..." refers to the fact that the client is basically told that the file has been received (implying it has been written) while the file on the destination volume could in fact end up absent or damaged in case of power outage (without a UPS to save the day) or destination local I/O error.

To ignore FILE_FLAG_WRITE_THROUGH if passed by the client, change strict sync back to no.

With NFS the async mode shows a huge difference when the speed of the network is significantly higher than that of the local disk I/O. So my assumption is that we should see something analog with SMB.

I have not yet tested for the difference at this point so I do not have a final theory. I also suspect that the results might differs depending on the client: Linux Samba Client vs Windows 10 because both are doing other caching but in different manners and the progress windows on both are kind of a "beautiful piece of art at the expense of actual facts" :-)
These users thanked the author domih for the post:
mad_ady (Fri Jun 26, 2020 1:09 pm)

Post Reply

Return to “General Topics”

Who is online

Users browsing this forum: No registered users and 1 guest