14
9
I have a computer with large hard disks running Gentoo. I have to serve med/big files via a wired network to Apple devices (all of them running OS X).
Which protocol is the best for the following needs ? :
Kami
Posted 2010-02-08T20:44:14.803
Reputation: 3 108
7
You should find this article interesting:
iSCSI, AFP, SMB, and NFS performance with Mac OS X 10.5.5 clients.
It shows these test results:
(in seconds)
iSCSI 134.267530
AFP 140.285572
SMB 159.061026
NFSv3 (w/o tuning) 477.432503
NFSv3 (w/tuning) 293.994605
harrymc
Posted 2010-02-08T20:44:14.803
Reputation: 306 093
7
I have made some real world, non-scientific tests of i/o speeds of iscsi and different network protocols in OS X.
My setup:
The test was made by copying (cp) about 2gb of camera raw files (each about 20-25mb in size) to the device, restarting the device and copy the same data back to local SSD drive.
Write performance:
*) Only when using SMB I got very different write performance results from copying the files to device using cp or mv command!
Settings the async option for NFS greatly improves the read performance. I use the following mount command for the test:
mount -t nfs -o resvport,soft,intr,rsize=32768,wsize=32768,timeo=900,retrans=3,proto=tcp,vers=3,async server:/share /private/share/
Read performance:
The protocols seems to handle caching differently. This is the results I got when copying the files to the device and immediately back to the local SSD drive (without restarting the device)
Read performance - without restart
My conclusion: I will use either AFP or NFS since both protocols gives similar performance and flexibility (compared to iSCSI) for my purposes (Lightroom, backup, media streaming)
David Wincent
Posted 2010-02-08T20:44:14.803
Reputation: 71
3
Although they are data transfer protocol, I would like to remind you they don't accept the same characters as file name. For example, \ / : * ? ” < > | are not allowed in Windows NTFS and Samba.
It happens to Apple Talk protocol from my experience on MacOS 8.6 and Windows 95 running AppleTalk compatible service. Some characters in file names are allowed in MacOS are invalid to Windows.
Detail about my experience copying files from Linux Desktop to QNAP TS-212P running Samba and NFS can be found in Performance Comparison on Backup Linux Files to QNAP TS-212P. The test result in MB/s to what interests you are:
I have done a test with FTP, Samba, iSCSI, and NFS in File Transmission with Different Sharing Solution on NAS with QNAP TS-112. The test result in MB/s to what interests you are:
Therefore, Samba is faster than NFS from my experience. But some of my files contains illegal characters in NTFS and Samba, I choose to use NFS as my primary protocol.
Wish it helps!
Amigo
Posted 2010-02-08T20:44:14.803
Reputation: 201
1
basically most (if not all) protocols can be used on any platform, but some of them are more native then another one
alexus
Posted 2010-02-08T20:44:14.803
Reputation: 2 484
4Okay I already know that but what about speed, ease of use, limitations and security ? – Kami – 2010-02-08T20:49:46.883
Actually that's pretty useful for me. – Chris – 2012-01-01T22:15:11.803
Not sure why this would get down voted, except maybe that it's a touch simplistic; AFP has the advantage on Mac because it has the best driver, OS X's NFS support isn't great, and SMB is better on more recent OS X versions. So a big factor of performance is protocol support at both ends; a NAS with a good AFP driver will be best for Mac users in most cases, for example. – Haravikk – 2014-05-07T18:27:05.773
Thanks for the article it was useful ! Can I say that in my case AFP will be the faster and easier to use than the other ones ? – Kami – 2010-02-08T22:35:10.657
3The NFS v3 results reported are pretty lousy. In my moderately extensive work optimizing throughput for customers, NFS tends to be twice as fast or more than SMB.
With one server, one host over a gigabit network with no other traffic, I can get 900 megabits per second for NFS when CIFS struggles past 250.
That's not sustainable real-world throughput, to be sure, and doesn't invalidate these tests per se, but does make me quite suspicious about the results. – Jon Lasser – 2010-02-09T00:12:26.193
@Jon Lasser: The trouble with such tests is that they are usually only correct for the hardware used AND the nature of the test. In general, they are at best only weak indicators. – harrymc – 2010-02-09T07:15:33.660
@Kami: I can't answer that, as per my comment above to Jon Lasser. If you have the time, you might try to duplicate these test results on your hardware. If they differ in any way, it might be interesting for you to publish it here. – harrymc – 2010-02-09T07:17:50.227