Syncing a Windows-SMB-share via App "Hybrid-Backup" extremely slow.

[UndSimQ]

@F.one - i understand your frustration and I guess 'you are welcome'

I will not write back on decades of IT experience (irrelevant to you), of running thousands of servers (irrelevant similarly), running many hundreds of TB of live data (irrelevant), running teams of engineers, and teams of teams of engineers (similarly irrelevant).

instead , I would just say if one of my 'guys' (men or women), people who work for managers who work for me, would come to be with this problem, the first thing I would ask them would be - what do we know about this server , what is the data on it, who is responsible , does it have to be there ,etc . yes, I can buy ** more hardware toys , have budget for it, and frequently do so - but why would I or my people f$ck around copying data that does not change at all because some dev team or some business group is too lazy to implement data lifecycle policy (aka archiving)? is this a smart task appropriately optimized for us or busy work because someone else is yes, lazy? This is how I run my teams and my people know they can get their stuff solved and we would go and require 'partners' to fix items on their end if they are on their end vs telling my engineers (sysadmins, DBAs, network admins, etc) that they are stuck trying to break through the wall with their head.. for me, this creates an environment engineers like myself can do best work and be supported, also it creates loyalty in people, both ways.

now, your management may be different and their degree of support may be different, and you may be in different position and timing place in your career so it is not an option. ok, cool. good luck anyway.

and yes, I remember tokenring, installing Win NT 3.51, OS2 4 and other fun things..

who knows , may be there is an issue that QNAP support can help you resolve and you can continue without changing requirements in any form that way...

[F.One]

@UndSimQ - Allet jut! You are welcome, too.

Sometimes one has to decide:

Are you willing and able to make people do their work with more discipline - or is it easier at the end to sync the f**king whole fileserver.
If the CEO himself decides that it's impossible to enforce any rule - he has to live with some other costs.
That´s all.

But hey - this way the CIO must get some bigger toys he would never could play with when everybody would work always as thrifty as he could do.

And to not being misunderstood: I appreciate that Qnap storage for its good overall performance and some features to a comparable very low price. I only doubt that HybridBackup is as stable / good as it should be.

PS:
You mentioned OS/2 - haha, reminds me at the times we had to implement terminal services from the AS/4000 and other iSeries in Eschborn via OS/2 into WinOS/2 at "Deutsche Bank", connecting the tokenring to good old "BNC-network", too - what a great idea of the management! Everybody knew - but it was a big decision you never had to ask questions for...

[UndSimQ]

you are completely right on the above. i guess I was fortunate to either work (or create) the environments where we get to ask questions/demand answers (rather than strong arming everything with infrastructure).

on QNAP , it is very likely you know more than me, i just bought mine and it is a toy for home , paired with 10G switch off ebay for $90 that is fun to learn turboiron on . so far I am happy but I do not have production requirements on it. in "real life" i deal with completely different systems/hardware.

"You mentioned OS/2 - haha, reminds me at the times we had to implement terminal services from the AS/4000 and other iSeries in Eschborn via OS/2 into WinOS/2 at "Deutsche Bank", connecting the tokenring to good old "BNC-network", too - what a great idea of the management! Everybody knew - but it was a big decision you never had to ask questions for..."

I actually liked OS/2 4 when it came out -real multitasking!! had to do REXX scripting on it for the radio phone management software (aka early cell phones) to know that my base stations are working properly and do verification. this is all mid 90s.

[F.One]

Yes, but "strong arming" sometimes is a really sustainable solution because you can construct a lot of things redundant - and if you are working on a real high level where there is nobody you could ask (but everyone asks you) you know how important this is for a good sleep at night.

And yes, too. Coming from Geoworks via WfW 3.11 I liked OS/2 4 a lot. Only disliked the decision to implement it with the decision at the same time to use MS Office 2.0 with tons of VBA to the Backend in Eschborn...

But we are completely off topic

[storageman]

(And isn´t a "mirror" always a one way backup? )

Not in my book.
A mirror is where the source and target contain exactly the same files.
If it is a one way sync the source is the master.
In a two way sync both the source and destination are both "masters", in which case there could be overwrite issues.

A backup, on the other hand, may contain deleted files from the source (in case you want to recover a delete mistake), and of course may include previous versions.

Going back your performance issue.
What is the speed doing a straight SMB copy from source to target?
Is the source in a VM? If so, this will affect performance, especially in a compare.

Also might be useful to check the internal read speeds to make sure nothing funny going on.
qcli_storage -t -> filesystem performance test
qcli_storage -T -> disk/raid performance test

The spec of that model should be about 400-450MB/s sequential 10GbE to a single host.

[F.One]

Thank you, storageman, you touched a good point!

Yes, the fileserver resides on a HyperV-host with 10GB eth

Copying to Qnap => HyperV-Host
1 big file => 250MByte/s
2.000 small files => between 50MByte/s and 200MByte/s

Copying from Qnap => HyperV-Host
1 big file => 450-500MByte/s
2.000 small files => 120M-350 MByte/s

Copying to Qnap => HyperV-VM
nearly the same results while testing on "Drive C:" of that VM - but with completely different results copying from/to the drive-letter where the content of the SMB share resides:

1 big file => ascending from 5 up to 80MByte/s
2.000 small files => between 1MByte/s and 60MByte/s, average: 15MByte/s

WTF?

So I checked the config of that VM
Both vhdx-files are mounted from an IBM Storewize V7000. Hm. What is the difference?
Diving deeper into this I found out that the vhdx with the VM-OS resides physically on LUN at a V3700 with 24x 10.000rpm SAS HDDs, being virtualized trough the V7000.
While the data source of that SMB fileshares resides directly on a LUN of the V7000 - but this DRAID was built up with nearline SATA-HDDs (!)

This is a huge difference and I will create a live copy via the V7000 to a SAS based storagepool with easytiering to flash right now and switch master / slave of that copy when it is finished.

Just for the records:
Output of "qcli_storage -t" is:
VolID VolName Pool Mapping_Name Throughput Mount_Path FS_Throughput
1 DataVol1 1 /dev/mapper/cachedev1 @?W?? /share/CACHEDEV1_DATA 836.60 MB/s
2 TimeMachine 1 /dev/mapper/ce_cachedev2 @?W?? /share/CE_CACHEDEV2_DATA 606.63 MB/s

So I will report back here if I would have to blame myself for not testing against the right volume of that VM! Simply did not looked where all those VHDX-files coming from the same source are placed in reality. My fault!

[storageman]

Good stuff, the plot thickens!
Yes a baremetal copy vs VM copy is a good test.
A further test 10GbE to 10GbE direct would also rule out switch issues.
As I said internal QCLI speeds are read speeds not write speeds but they are useful stat.

[Thisisnotmyname]

@UndSimQ: I guess some parts of your posting is more "opinion" than real experience.

like:
"...your current setup is already doing synchronization of 4-5 files a second against remote source , every day (60k files 5 hour runtime reported), that is not bad."

As I stated out:
1) The initial copy is done with some KILObytes per second when there is nothing to compare at all
2) It does not matter at all wether or not there are to be transferred 1Gig in thousands of small files or as a single big one
3) All resources are bored on both ends: cpu, disks (IO and throughput!), NICs

If you call a throughput of some kB/s under these circumstances "not bad" you are probably still on a PS/2 system with token ring. Otherwise it is just your opinion.

You do not know anything about the purpose, the structure and the usage of the source server - but you feel free to give advices to "archive" or even "better" call people stupid.
Sorry but I don´t like these kind of postings too much.

But just for the records:
I did set up a test job for a real small part of this server - and guess what? No "skyrocket" at all - bits still just trickle to the Qnap.


Anyway - thank you for your input.

The point that even QNAP support is unwilling / unable to answer my ticket for 2 weeks now is a sign to me to not use this solution (HybridBackup) anymore.
Maybe I will try another Synctoy or I can manage to virtualize that Qnap NAS via iSCSI behind the just ordered IBM V5030. Then I would live-migrate the source servers storage via fiberchannel to the v5030, too and can create real SAN snapshots.

Regarding the bolded item, you do realize that this is not QNAP's ticketing/support system, right? This is a community forum that they don't read or participate in. Participants here are just fellow QNAP customers like you. QNAP has an actual ticketing system at https://helpdesk.qnap.com

[F.One]

Sir, excuse me, Sir. But yes, I know that this is not a Qnap support system. That is why I opened a ticket at Qnap before I posted my question here.