Unsolved
This post is more than 5 years old
6 Posts
0
86932
January 25th, 2014 10:00
Slow network performance R720
I have recently purchased 4 Dell R720's and I'm noticing a big network performance issue. These systems are running Windows Server 2012 Standard 16gb of RAM and 6x 1.2tb disks in a Raid 5. These systems are connected to a gigabit cisco switch. The issue i'm seeing is when I try to transfer files to these new servers im only able to achieve about 200mbps of throughput. The servers are under minimal load and should be able to max out (or get close) the gig connection between servers. I have already updated all drivers and tried disabled QoS on the nic. Any help is appreciated here! Thanks!
No Events found!


Daniel My
12 Elder
•
6.2K Posts
0
January 25th, 2014 10:00
Hello kylebryan
The hard disks may be causing the network connection to slow down. 200Mbps is 25MB/s, and that is about the write speed you will get on an entry level RAID controller that does not have cache when you run a RAID 5.
Which RAID controller do you have? When you perform transfers the other direction is it faster?
Thanks
kylebryan
6 Posts
0
January 25th, 2014 11:00
my write cache policy is set to Write Through. Can you give me a brief walk through on changing these settings? Is this something that can be done on the fly or will it require the server to be taken offline?
Daniel My
12 Elder
•
6.2K Posts
0
January 25th, 2014 11:00
What is the battery status? If it is set to write through because of the battery then it will automatically switch when the battery charges up. Yes it is something you can change on the fly.
Caching is a virtual disk option. It is not a controller wide option. If you have OMSA installed then go to the Virtual Disk section and in the task list for the virtual disk you can "Change Policy". Write policy is where you set the controller cache "Write Back". The Disk Cache Policy tells the controller whether or not to use the cache on the physical disks. Enabling the cache on disk will improve performance, but it is not battery backed so in the event of sudden power loss there can be data corruption if that option is enabled.
Thanks
kylebryan
6 Posts
0
January 25th, 2014 11:00
The array is not initializing and the network link auto negotiated at 1000mbps FULL duplex. I just performed a local file copy of a .pst file that was 13gb. when the transfer started it copied at about 170MB/s and after it got about 80% complete it dropped down to 10-15MB/s. so that leads me to believe there may be some read/write issue on these systems. Any perc configurations I should check out?
Daniel My
12 Elder
•
6.2K Posts
0
January 25th, 2014 11:00
The short answer is: Working as intended.
RAID 5 requires parity calculations. The H710 has a faster CPU and cache to process these parity calculations. The H310 has to slow down all I/O when it falls behind on data calculations. You will have good burst speed on short writes, but on big writes it will eventually fall behind and start throttling the flow of data because it doesn't have anywhere to cache it.
You have two options if you want better performance:
#1 Do not put a RAID 5 on this controller. Use 0, 1, or 10 only. If a parity array is present anywhere on the controller it will slow down all RAID arrays.
#2 Contact your sales rep and ask them if you can upgrade to an H710 and pay the difference. Normally any change on your order will require the entire system to be returned and reordered within the return period. Sales or customer care may be able to work something out where you don't have to return the entire system.
Thanks
Daniel My
12 Elder
•
6.2K Posts
0
January 25th, 2014 11:00
Also, go to task manager and check your disk activity. There may be a backup, WSUS update, windows updates, or any number of other things going on that are using a lot of network bandwidth or disk I/O.
kylebryan
6 Posts
0
January 25th, 2014 11:00
PERC H710 is what I'm running. 6x 10k disks. I should have no problem maxing out a gig link with this setup. Do you recommend I change the read write policy on these systems?
Daniel My
12 Elder
•
6.2K Posts
0
January 25th, 2014 11:00
That is definitely not the problem then. Even with all of the caching disabled on the H710 it should be faster than that. Is the array still initializing? When you create the array it will perform a background initialization that uses about 30% of available writes.
Copy a file locally on the server to see what the write speed is. This will help us determine if it is a write speed or network issue. Also, check the network pane to see what the link speed autonegotiated to.
Thanks
kylebryan
6 Posts
0
January 25th, 2014 11:00
it looks like I may have spoken too soon. It looks like im running an H310 controller and I don't see a battery... hmmm. What are my options now?
Daniel My
12 Elder
•
6.2K Posts
0
January 25th, 2014 11:00
Do you have OpenManage Server Administrator installed?
Check the write cache policy. It should be set to write back. Also, check the battery status. Write back is the default setting, but if the battery is below a certain charge threshold write back will be disabled and it will write through instead. It takes about 8 hours for the battery to hit the minimum charge from a 0% charge.
pcmeiners
6 Operator
•
1.8K Posts
0
January 25th, 2014 14:00
Get a controller with coprocessor/cache as Daniel suggests, will make a world of difference. This is the main bottleneck
After the new controller....
"performed a local file copy of a .pst file that was 13gb." If this is Windows, the OS is not optimized for copying large files, just the opposite, find a disk test which better matches typical data chunk sizes.
If you have a small network, < 70 users, in the NIC properties, turn off flow control and test.
With the new controller, I would recommend you install a defrag program which can do a boot time defrag bi monthly. Boot time defrag will maintain approx. 5-8% increase in speed as it defrags all open files which can be done by MS defrag.