....

Reported Speeds:

Array -> PC ~ 50MB/Sec

PC -> Cache ~ 74MB/Sec

PC -> Array ~ 31MB/Sec

 

.... suspect this is the reason for my ~32000 dropped packets 

 

Ummm, I would suggest that attempting any performance measurement is rather pointless when so many dropped packets are reported!

Bump

Bump

 

I'm not sure what you are expecting when there is clearly a hardware (or configuration?) issue on your network (from the high count of dropped packets).  Address that and then re-visit your transfer speeds.

Bump

 

I'm not sure what you are expecting when there is clearly a hardware (or configuration?) issue on your network (from the high count of dropped packets).  Address that and then re-visit your transfer speeds.

 

I'm well aware of what is causing my dropped packets, it's the shitty network cable I haven't bothered to replace. I'm not posting in general support, I'm more interested in what hardware setups other users have who are able to sustain r/w over 100MB/Sec.

 

I think unRAID is a fantastic product and I'm always looking for ways to increase performance (as speed is definately not one of unRAID's strengths). Even though it is marketed as a media server and speed isn't "required," the features and flexability allow for evolving unRAID into other markets.

I read that some of you were getting speeds above single GbE. How could you do that? I know that bubbaQ tried here to set up Link Aggregation but didn't succeed yet.

 

Are you using EtherChannel with multiple NIC? How to set up unRAID for EtherChannel use?

 

I'm very interested to hear from you guys

 

EDIT: Reply to my own question --> VM. This is a bit too tricky for me so I'll skip and save for 2 10GbE NIC instead.

I did some tests with my server and noticed that I can write to cache drive over gigabit network at ~105MB/s, however when reading from it I can only achieve ~50MB/s... I'm wondering why I can't also get ~105MB/s when reading from the server?

 

I did tested dd on my server and it's not HDD bottleneck for sure as my cache disk (wd red) can read ~150MB/s...

 

My unRaid server NIC is onboard Realtek 8111F, using the latest unRaid RC version, and my desktop PC NIC is a Marvel Yukon 88E8053, using windows xp... maybe related to one of these two onboard cards? Or software/drivers related? Unfortunately I don't have a 3rd PC with a Gigabit NIC at this moment then it's hard to identify in what system is the bottleneck...

 

Anyone else experienced similar behavior with such big difference in read vs write speed with similar NIC's?

Ok, did further tests and seems not really NIC related, I did now found that I can read from the server at ~100MB/s using FTP, the speed values I did posted on my post above were using smb. Seems I need to search if I can somehow tweak/optimize smb (not sure if at unRaid or at XP?) to get ~100MB/s on both directions... hints are welcome...

Hrm, I've not tried using FTP to speed test and I don't use a cache. It would be great to see higher write speeds on my system but I don't know what I need to tweak. I do have the later samba at least and I think I no longer have any slow green drives...

Ok, did further tests and seems not really NIC related, I did now found that I can read from the server at ~100MB/s using FTP, the speed values I did posted on my post above were using smb. Seems I need to search if I can somehow tweak/optimize smb (not sure if at unRaid or at XP?) to get ~100MB/s on both directions... hints are welcome...

 

SMB normally does not require much tuning for a LAN. Are you testing with the same files? How are you measuring the transfer speed?

Same file, I have the cache HDD empty with just two big test files, one with 2GB (the one I used for most tests). As I said before it's not HDD bottleneck at all as testing with dd on server it can read the test file at ~150MB/s at "first read"... and ~6GB/s after cached... server haves 8GB RAM... whole test file get's cached... then in almost all tests the test file was not even actually read from the HDD as it was already cached.

 

I can only guess that it must be related to how smb works, amounts of data sent in each "packet" before an ack or something like that... either at server or at XP machine, anyway strange why can only notice it in one direction, but maybe one of them haves slightly different behavior on size of sent "packets" or one replies acks faster, must be something... I did even tried to add some "socket options" on smb.conf that I did read that could help but no improvement.

 

Re how I measure it, for testing SMB I used most of time Teracopy to copy files to/from the server, Teracopy itself shows copy speed, and for FTP used Filezilla that also shows transfer speed, and... also used a very simple self-made network speed meter software (winpcap based) that measures total speed of data in/out in my windows machine network connection, independently of the software/protocol I did tested, and it did match the values on Teracopy/Filezilla. Also easy to see that a 2GB test file takes ~20s to copy from XP to server and takes ~40s to copy from server to XP.

 

P.S. HTTP download from server can also reach ~100MB/s, seems only SMB from server to XP (not from XP to server as mentioned) is degraded to ~50MB/s

Well, since you are doing a copy, it is a write test in the reverse direction.

Windows XP Network Connection Limit: http://support.microsoft.com/kb/314882

Just wanted to add that using FastCopy tool produces these results: xp->unraid = 105MB/s and unraid->xp 70MB/s (a bit better than TeraCopy but still not ~100MB/s, and yes tried to increase buffer size, no help).

 

Also tried W7 on the desktop PC instead of XP and... I was able to reach ~110MB/s in both directions using native windows explorer copy function! guess it is a lot more effective for such speeds than in XP... also it should use SMB2 with W7... Interesting though is that Teracopy on W7 is slower than native explorer copy reaching only 50MB/s in both directions (yes worst than in XP, in one of the directions at least...).

 

Maybe I need to live with that for XP...

This is my setup

 

ESXi 5.5 with unRAID as a VM (21 disks connected/passthru to unRAID).  We'll call this VM#1

 

Got another two VMs, one is a plex server (VM#2), and the other is running SAB/Sonarr/CouchPotato...etc. (VM#3)

 

I'm trying to improve the transfer speeds, since VM#3 downloads the content and stores it on a user share on unRAID via SMB share.

 

VM#2 (Plex), reads content via SMB share.

 

All the three VMs are connected to a vSwitch (vSW#1), and uplink of 1 Gbps to a small netgear switch, and the gateway for all the VMs is the IP of my router.

 

I've never seen any write/read more than 20-29 MB/s, which i'm assuming is low.

 

I have this plan: Create a separate vSwitch (vSW#2), add to each of those VMs a second NIC, that connects to this new vSW#2.  Then assign a different subnet IP to those NICs, and enable Jumbo Frames (9000). This new vSW will have NO uplink/gateway, so it will be internally within the ESXi only.

 

Then, I want to mount all the shares to VM#2 and VM#3 via a hostname (and edit the host file on each machine to use/communicate with unRAID via the new subnet by assigning dedicated hostname to the new subnet IPs).

 

Is this possible? will it help improve the transfer rates? what are your thoughts?

 

Thanks

 

Update 1: I just tested a copy from disk1 to disk2, an 8GB file, and it was taking around 44 MBps, as per TeraCopy.  Note that this was done via my Windows 7 PC.

 

Additional Info: The reason that i'm trying to improve the performance in the first place is, that when SAB post processes the downloads (e.g. par, unpacking, moving to final user share), it freezes my access to the network shares (user shares) of unRAID.

Those write speeds are normal if writing directly to the array.

 

You can do one of two things to speed it up.

 

1. Add a cache drive to vm1

 

2. Have sab keep the files it downloads to its vm until all post processing is complete, then move it over to the array.

I'm curious if anyone has done this yet, but my plan is to use a 3-drive striped raid array for the cache drive.  That will at least provide high write speed.

 

(Any recommendation on a raid card supported by unraid in the manner?)

I've tried RAID-0 pool cache drive in two setups: 4x2TB HDDs (two 7200 and two greens), and 2x2TB 7200 HDDs. Did not find noticeable difference so settled with  2x2TB 7200 HDDs.

 

Last card used is Addonics AD4SA6GPX2 4-port SATA-III PCIe 2X . The list of cards I tried you may find here.

 

Writing to such cache drive from laptop (1TB Samsung 840 EVO SSD) easily saturates 1GB network, of course. Why wouldn't it.

 

I am not looking for crazy speeds, but it would be nice to have at least 60 t0 70 MB/s transfer and write speeds to the array.  Will this be possible with either a 1tb 7200RPM drive or should i go for a SSD drive?

Disable parity for duration of transfer and you will get it.

I think the most I've ever transfer from my W7 PC to my server at once was 120 GB.  I mainly transfer 1 or two movies a day, which is less than 90GB.

Although I wouldn't do it for small transfers.

I am not looking for crazy speeds, but it would be nice to have at least 60 t0 70 MB/s transfer and write speeds to the array.  Will this be possible with either a 1tb 7200RPM drive or should i go for a SSD drive?

Disable parity for duration of transfer and you will get it.

I think the most I've ever transfer from my W7 PC to my server at once was 120 GB.  I mainly transfer 1 or two movies a day, which is less than 90GB.

Although I wouldn't do it for small transfers.

 

Thanks for the suggest.  I now change the way I transfer my files.  I transfer them while we are all asleep lol.  By the time I wake up everything is done.  Hell, I've been doing them via wireless for speeds ranging from 3 MBps  to 22 MBps depending on file size.