Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[drick]

My new NIC's FINALLY arrived from Amazon, and I'm now back into testing mode.

So here is my setup:

Vista x64 / Intel Pro 1000 PT Server NIC - node 1
Server 2K8 x32 / Intel Pro 1000 PT Server NIC - node 2
DNS 323 
 -FW 1.07
 -2x400GB HDD's in a RAID 1
 
All units have jumbo frames turned on to 9k, and I have all the bells and whistles turned on the NIC's as well.

So far the best throughput i can get is around 18 MB/sec sustained doing a file move from node 2 to DNS.

Has anyone else gotten better numbers than this, and if so how?

[fordem]

Server 2K3 x32, Intel Pro1000CT (integrated), 9000 jumbo frame.
DNS323, fw 1.06, 2x250GB RAID1

Sustained write - roughly 24MB/sec, sustained read - just under 31MB/sec (I'm going from memory here)

There are a couple of details that you haven't mentioned (and neither have I in the above - just so we're comparing apples to apples), I'll provide them now ...

Thos figures were achieved using a single 2GB file (actually NASTester) rather than many small files and the disk subsystem at the 'other end' was also 2x250GB RAID1 using an integrated Intel ICH5 SATA controller.

When doing a timed file transfer you MUST take into consideration the disk subsystem at the other end of the test, it is either being read from or written to and it can have a significant impact on throughput.  One of the reasons I'm going from memory is I have since upgraded the server, and I am now running W2K8 and 3x250GB RAID5 with an IBM ServeRAID7t controller (a relabelled Adaptec 2410SA) and the same test gives significantly lower read throughput - one of the short comings of RAID5 is the write bottleneck that it creates, so controllers try to deal with this by caching - if you're watching the actual bandwidth usage, you'll see the initial throughput at a pretty good level and as the controller's cache fills you'll see the throughput drop significantly.  With RAID1 the throughput stays more or less constant from start to finish.

[drick]

fordem,

good point on the details, but I wanted to get the string started and add them later.

for this test the source has an embedded Intel ICH8rR controller, and I used approx 10GB of Linux iso's (same data i used for the other testing on the other string). i also was doing a write from the server to the 323, but i didn't test the read back.

I will test the read back from 323, and post my results.

I will also test both methods against a real server this weekend, and I'm sure the numbers will go up significantly.

[drick]

ok, baseline test from file server to another server with a PERC5i RAID controller to see what the read speeds are of the source ICH8R controller and disks

copy from ICH8R to PERC5i (i actually copied these to 2x73GBx15k RPM SAS HDD's in a R1, so this should eliminate the source HDD's as the bottleneck)
9m @ 20,840,803 Bytes/Sec

copy from PERC5i to DNS (from the SAS HDD's)
14m @ 13,406,509 Bytes/Sec

all tests are run using MSFT Rich Copy 4.0, and the source data files are Linux iso's which equate to 10.5GB total.

this also tells me that my assumption that the DNS is the bottleneck is correct, would you agree with that?

[drick]

ok, trying this again with the new firmware.

i'm now getting just under 30MB/s (29.7) sustained in a copy from the 323 to a Vista WS.

SMOKING!!!

[drick]

now testing the data migration back to the 323 from the same host, and am only getting 11.5-11.9 MB/s sustained.

that's strange that it can only write 50% as fast as it can read. per the spec sheet for the hdd it should go a LOT faster:

Transfer Rate
Maximum Internal (Mb/s) / 1287
Maximum Sustained, SATA (MB/s) 105

[Arvald]

your hitting the speed of the processor in the 323 and since you are in RAID1 there is a slight overhead for that.

needless to say, you will never get full HDD speed on a unit like this.

[drick]

your hitting the speed of the processor in the 323 and since you are in RAID1 there is a slight overhead for that.

needless to say, you will never get full HDD speed on a unit like this.

thanks for the info arvald. is 12 MB/s truly the write limit for the ARM chip then? that is a good piece of info to know if so.