NAS Optimization

[Lancer]

I just got a Linksys NSS6100 and so far I am really amazed at the amount of features it has, but being my first NAS I was really underwhelmed by it's performance. After a bit of investigation I really had to come to terms with NAS is just not fast. I setup the port aggregation on the unit and have everything running through a gigabit switch. Right now I get about 6.5MB/s on the unit which is about average I have found. So on to my questions I am exploreing a few things to try and make this work better.


1. MTU - Average setting for MTU is 1500. According to Linksys this unit will support up to 9k. If I adjust my system to 9k MTU will this have an adverse affect on my regular network transfers (system to system)? Or does this only effect net traffic?

2. I am running DHCP from my router. This is a 10/100 router, it is connected to my gig switch. Since it is the DHCP server is this causing my network traffic to hit the router and then back out the switch creating a bottleneck? And if so would going from a DHCP environment to a static IP one fix this issue?

 

[Farva]

1) Each line drawn has its own dedicated bandwidth. You won't notice lag on any of your other machines.

2) Yes, the switch on router will be the bottleneck. No, a static IP will not fix the bottleneck. You need to put the NAS on the Gb switch in order to get the improved throughput.

 

[venm11]

IRight now I get about 6.5MB/s on the unit which is about average I have found.

Good lord. I don't know what drive & raid config you're using but that's intolerably slow.

If you ran raid 5, and gbe, shouldn't you be seeing between 60-120MB/s ? I get about 40MB/s on large transfers across gbe from a 1-drive pc/linux server.

 

[Lancer]

Well Tom's Hardware has a NAS Chart and while mine is still hitting on the lower end it's not as bad as many.
http://www.smallnetbuilder.com/component/option,com_nas/Itemid,190

System to system transfer's are no issue and are very quick I think I have figured out a few reasons for it.

1. CPU power much better on my gaming system and my Poweredge 1800

2. NAS supports 75 concurrent connections. There may be bandwidth limitations setup so multiple users are not killing the bandwidth.

For speed purposes I have all 4 drives in RAID 0 With both LAN ports set too aggregate and are connected to the switch. I don't have anything conencted to my router. Just a single line tieing it into the switch so my gaming system, the Poweredge, and both LAN ports from the NAS are all connected to the switch and the router is providing the DHCP service. For fun last night I decided to go ahead and start a second transfer to the NAS to see if it would slow down and even out the bandwidth. It didn't both transfers were running at 6.5MB which leads me to believe it is limiting bandwidth to share across the connections.

 

[unhappy_mage]

1. MTU - Average setting for MTU is 1500. According to Linksys this unit will support up to 9k. If I adjust my system to 9k MTU will this have an adverse affect on my regular network transfers (system to system)? Or does this only effect net traffic?

This can help, but you may need to set all of your machines to 9k MTU to get it to actually use that setting.

2. I am running DHCP from my router. This is a 10/100 router, it is connected to my gig switch. Since it is the DHCP server is this causing my network traffic to hit the router and then back out the switch creating a bottleneck? And if so would going from a DHCP environment to a static IP one fix this issue?

Unless the traffic would have to flow through the 100 megabit switch, it shouldn't cause any problems. DHCP is a rare thing, it shouldn't be doing it continuously.

 

[Madwand]

Well Tom's Hardware has a NAS Chart and while mine is still hitting on the lower end it's not as bad as many.
http://www.smallnetbuilder.com/component/option,com_nas/Itemid,190

The SNB NAS charts are dominated by cheap products, and the reason that they don't perform well is that they're cheap products. AFAIK, yours is not in this class and should be performing much better.

A cheap desktop with a consumer OS and a single drive can hit around 30 MB/s in file transfers without much trouble. So an expensive device with multiple drives in RAID 0 (!) hitting < 10 MB/s is pathetic.

Are you sure that it's connecting at gigabit speed to the switch (check the indicator light/color)?

Have you tried contacting the vendor and complaining about performance? I'd want my money back if the device was to blame and its performance couldn't be improved significantly.

Here's a performance test setup from Infrant. I'd increase the file size 10x and include 64k accesses in the test:

http://www.infrant.com/forum/viewtopic.php?t=265

 

[Lancer]

Well now I am even more confused. IOmeter paints a very pretty picture for the NAS and one I would be very happy to see, but as you can see between it and real world they do not line up.

I really do not believe this to be an issue with my system. Here is it's setup

Intel E6400 @ 3.2GHZ
Intel 975x BadAxe motherboard
2GB OCZ Platinum Rev 2 memory
PNY 8800GTX
36GB Western Digital Raptor
2x 300GB Seagate drives in RAID0
Windows Vista Ultimate

My server is seeing the same kind of transfer rates.

Intel 2.8GHZ Xeon EMT64
1gb ECC Registered single rank
2x 40GB Maxtor drives in RAID1
3x 300GB Maxtor drives in RAID5
Onboard Intel video

Both connections on the NAS are connected gigabit.

Any ideas on what could cause the dramatic difference in transfers?

 

[Lancer]

Did a little more digging and found my switch doesn't support 802.ad so I though I had link aggregation setup but alas no. still doesn't account for the slow speeds but at least I know I need to start looking for a new switch.

 

[Lancer]

Well thanks guys for your input I have a few contacts with Linksys from my work and they were able to get me with an engineer. They thought this sounded a bit off so they ran some tests for me in their labs. The results were surprising. And the winner for what my issue was........

Windows Vista.

Here are the speed tests they went through. Two notebooks each on XP pro and a workstation running Vista Ultimate. As messed up as it is the notebooks running xp were twice the write speeds of Vista. Very interesting and upsetting as I have 6 systems in the house 5 of which are Vista and one running Server 2003. So I'll test this tonight and see how it goes.

NSS6000 drag and drop throughput tests
RAID1 array
7/20/07
3PCs: Toshiba Tecra M5, HPXW4000, IBM Thinkpad R52A


Toshiba Tecra M5
WinXP pro - PC1
gigabit NIC - Intel Pro/1000PL
centrino duo
T2400 @ 1.83GHz
512MB RAM


write - 495MB
--> 15MB/s


read - 495MB
-->18.3MB/s


write - 2.01GB
--> 14MB/s


read - 2.01GB
--> 20MB/s




HP XW4400 workstation
Vista Ultimate
Intel Core2Duo
6600 2.4GHz
2GB RAM
NIC - Broadcom NetExtreme Gigabit Ethernet


write 495MB
--> 8.33MB/s


read 495MB
--> 11.8MB/s


write 1.16GB
--> 8.16MB/s


read 1.16GB
--> 11.3MB/s



IBM thinkpad
gigabit
1.86Ghz Pentium M
512MB RAM
Broadcom Nextreme Gigabit


write 495MB
--> 16MB/s

read 495MB
--> 19MB/s


write 1.16GB
--> 13.3MB/s

read 1.16GB
--> 18.75MB/s

 

[Madwand]

When it works well, Vista can significantly out-perform XP, but unfortunately, it has some "bleeding edge" issues like this, so sometimes performs poorly.

There are a couple of Microsoft hotfixes that can help sometimes. Disabling autotuning (and perhaps remote differential compression) sometimes helps. Search around for this stuff, or live with it and hope that new drivers, Vista SP1, etc. eventually clean up the mess.

Changes on the server side could also help -- esp. adding SMB 2.0 support (when it's finished, debugged and tuned).

The IOMeter performance shows that the performance problem could be related to Windows Explorer implementation. You could try third-party file transfer tools, e.g. xxcopy.

 

[Lancer]

Yeah I am going to try and dual boot into XP and see if I have the same results. As far as my standard Vista install I am going to try and hard set it to the 9k MTU setting to see if Jumbo Frames helps.

As far as IOmeter goes I learned a lot about that test after I ran that. The way that it was suggested to test it in the infrant forums is really misleading and not what you should expect real world. They have you set it to 100% sync transfers and really that is just not going to happen, but oh well I will be happy if I can narrow this down. I have a Netgear unit similar to this coming to test out and I am interested if we will see similar results between the OS's

 

[Toytown]

Its vista and i really think its the explorer application which is doing it. There is a post on another forum which is showing almost exactly the same issue.

Here are my results with my own NAS, I have all the updates from windows so far for it (Vista X64) Copying a 4GB Iso using windows explorer (except the FTP one)

Windows XP ----> NAS gets approx 10MByte a second
Windows Vista -> NAS gets approx 4MByte a second
Vista FTP -> NAS gets approx 10MByte a second

My NAS only has a 100MBit connection hence the speeds.

 

[Lancer]

This is interesting and a pretty big flaw IMO. I wonder why more people haven't really talked about it. It seems like something that they should be able to fix.