Unsolved

This post is more than 5 years old

6 Posts

6707

May 19th, 2017 13:00

Problem: Copying files from Isilon to same Isilon = Slow (Using CIFS share and Windows machines.)

Questions in color.


Problem summary:
Copying files from Isilon to same Isilon = Slow (Using CIFS share and Windows machines.)

Problem details: In our workflow we copy millions of files from folder A to folder B (same Isilon, same share). This process is slow as we interface with our Isilon through CIFS and use Windows. As in when we copy files they are…
1. Sent from the Isilon
2. To our Windows machines
3. Back to the Isilon

Is there one or more ways to complete this work in a more efficient manner?

I was thinking of a few solutions… If they are even possible. Additional solutions are wanted!
A. We SSH into the Isilon and perform the copy work LOCALLY? Is this possible? How much faster would it be (a guess is fine)? How hard is it to setup AD permissions?
B. We mount the Isilon share via NFS and use a Linux server to perform the copies? Is this possible? How much faster would it be (a guess is fine)? Would we have to map Linux permissions to AD permissions?

Thanks!

2 Intern

 • 

300 Posts

May 22nd, 2017 07:00

furthermore it could be worth a try to upgrade to Server 2016 and use OneFS 8x. Then enable Server side copies which will cause the copy to be server-internal.

4 Operator

 • 

1.2K Posts

May 22nd, 2017 07:00

Most likely you will see best improvements by doing multiple copies in parallel.

Tools like emcopy and robocopy have capabilities for this built in, have you tried them yet?

I would not recommend to allow users log in to the Isilon cluster directly.

If you have Isilon SyncIQ licensed it can also be levereged for intra-share

parallel and efficient copies. Just note that an admin person needs to be involved

for manual copies; or some regular schedule needs to be set up for automation.

Using NFS just to manage data with Windows ACLs is probably not

work the hassle...

hth

-- Peter

3 Apprentice

 • 

624 Posts

May 22nd, 2017 10:00

SyncIQ will copy the files the fastest, but it will be a Read Only copy until you delete the SyncIQ policy and make the directory Read Write.

6 Posts

May 22nd, 2017 16:00

"Server side copies", interesting! Thank you so much for this! I'm going to research it!

6 Posts

May 22nd, 2017 16:00

Peter,

Multithreading: Yes, Robocopy is our go-to copy tool and most of us manually set multi-threading to 8.

SSH: To compensate for "SSH is not recommended" we were thinking about making a Windows application that handles the SSH session and copy commands, while at the same time hides the username and password. In other words the application would present the end user with a GUI that requests SOURCE and DESTINATION. Said application would have the SSH credentials built into it. Would something like this make a person in your position feel more comfortable?

Please keep in mind we are very attracted to the possibility of copying files locally on the Isilon. As a bonus, we are trying to find out if said copies would be symbolic links. As in they are not actually copies, just pointers to the original files.

NFS: When you mentioned NFS might not be worth the hassle. If it copies files 2x faster it would be worth it in my opinion. Again we might be able to make a GUI that handles the copy work. Nonetheless the sources below make me think the hassle wouldn't be that large. But my understanding of the Isilon and Linux are limited.

“Users who gain access to the cluster over SMB can access UNIX files stored over NFS with their LDAP accounts. Likewise, users who gain access to the cluster over NFS can access Windows files with their Active Directory accounts.”
Source https://www.emc.com/collateral/white-papers/h12417-identities-access-tokens-isilon-onefs-user-mapping-service-wp.pdf


“OneFS uses identity mapping to translate between Windows and UNIX permissions as needed.”
Source
http://doc.isilon.com/onefs/7.0.1/help/en-us/GUID-5A903ADA-A3CA-4DC4-B314-586583B0ACF0.html

4 Operator

 • 

1.2K Posts

May 24th, 2017 08:00

Hm... haven't heard "It depends" for some time now in this forum

You earlier mentioned "millions of files", so these presumably are small files, like couple tens of KBs?

With small files, most time is spent creating the new files,

rather than transfering data from source to new files.

Both network/protocol latency and file system "metadata" write operation contribute,

and getting rid of network/protocol buys you only one part of the job, maybe even

the smaller part compared to the network protocol. In depends... on physical

network latencies, protocol efficiencies (SMB2, SMB3 with server-side copies, NFS)

as well as metadata write efficiencies on file system level (metedata r/w acceleration on SSD,

metadata write caching/coalescing).

This is why most of the improvements I'd expect from doing operations in parallel.


It would take some efforts to build not only a safe "ssh wrapper" but also

a reliable parallel copy tool to run locally.


But it seem you are willing to spend some energy on the problem, so

why not first establish some baselines

- robocopy with higher thread counts (a google for other tools promising highly parallel copies)

- ... same with multiple instances on different directories!

- ... same with server-side copies enabled

- plain local copies with cp or cpio, also consider cp -c (cloning, an Isilon extension)

- (NFS -- nah, beware of the permissions/ACL complexity)

- SyncIQ (Yes! get a trial license from your account team)


Of course, carefully watch overall cluster performance when

stretching the test to highest levels(!)  Getting the millions

of files copied fast enough, at the expense of general

performance sufferings, looks like a less welcome suprise.

Let us know what you find, for your specific setup.

-- Peter

6 Posts

May 26th, 2017 14:00

Would Server-Side Copying work with Windows 10? I'm assuming so because it uses the same version of SMB 3.1.1?

2 Intern

 • 

300 Posts

June 1st, 2017 04:00

yes.

Windows 8 and later

And to correct myself it would also work with Server 2012 and with Server 2008 (but here only via robocopy).

6 Posts

July 21st, 2017 08:00

Question: Within OneFS 8x, can 'Server-Side Copy' be enabled per CIFS share or is it an all or nothing setting?

2 Intern

 • 

300 Posts

July 24th, 2017 03:00

checked on 8.0.0.x and it seems like it's all or nothing. afaik Continious Availability and ServerSideCopy are exclusive. since CA is configurable per share you could enable CA on a share you don't want SSC work with and enable SSC globally.but then you enabled CA on the Shares you didn't want to touch for SSC - so you didn't win a lot

4 Operator

 • 

1.2K Posts

July 24th, 2017 03:00

yeah, it's only in the  global SMB settings, like a couple of other SMB features...

6 Posts

August 31st, 2017 09:00

Is SSC on by default? Or does one have to manually toggle it on within OneFS 8x?

No Events found!

Top