UNSOLVED
E
EUtRWBdG1A12566
9 Posts
0
645
July 28th, 2014 02:00
Should we expect difference in Read and Write speed for File share
Hello Experts,
In our VNX5700, we are using CIFS for hosting our File Server Environment. Recently we experienced Performance issues(SR#63918522) wherein users were experiencing slowness in accessing the network shares which seems to have been caused by excessive re-transmission over the network.
In this regards, our Network team performed a series of tests over the network & found that the reads throughput(MB/s) (while opening a file on a network share) are always lesser as compared to the write throughput(MB/s) (copying a file over the network share).
My quesiton is: Is it expected that reads will always be lesser than writes for a File server Environment on a VNX5700?
We are only using 7.2K rpm NLSAS disks configured in Raid 6 Storage Pool for File.
I've attached server_stat output to this discussion as well. I've also raised a SR# 64608740 with EMC Performance team. Still awaiting their findings though.
Please let me know if more information is required in this regards.
1 Attachment
server stat output.txt
server stat output.txt
AdamFox
254 Posts
384
0
Posted July 28th, 2014 14:00
I wouldn't say it's "always" expected, of course the answer is "it depends" on your situation. Working with support on the reason why you are getting re-transmissions on reads is definitely where I would focus.
While I cannot say if this is your problem, one that I have seen before is in the case where the NAS server has a much larger pipe than the client and is subsequently over-running the client's switch port. Think about it. If the NAS server has a 10Gb interface and the client has a 1Gb interface and you ask for a large read, the server has the potential to send more data over its wire than the client could handle at a time being that the client's network is 1/10 the speed. So on a write, there is no problem because the client is sending data to the server at 1Gb/s and the server has plenty of bandwidth at 10Gb/s. But reverse the situation and you could have a problem because the server is sending data, perhaps not at full line speed, but at some multiple of 1Gb/s to the switch. The switch must then buffer any excess packets while it sends the packets to the client. If that buffer over-runs, most switches will drop packets and re-transmissions are the result.
Again, I am not necessarily saying that is happening in your case, just that I've seen it before. It's worth checking the switch to see if there are drops. If there are, especially if you can see it at the port level, then you may want to ensure that flow control is enabled both ways (many switches don't do this by default) so that the switch can tell the NAS server to pause if the buffers are filling too fast. This is better than re-transmitting which takes longer and should help latency.
Regardless of whether this is your particular issue or not, anytime you have a significant re-transmission rate, say > 1%, that is the first thing I would try and resolve because at that rate, performance will suffer. No network is perfect so a few re-transmissions are not necessarily a problem, but the more of them you have over the course of a test, the worse it gets.
Just a thought. Good luck and definitely keep the case going!