D-Link Forums
The Graveyard - Products No Longer Supported => DNS-323 => D-Link Storage => Beta code! => Topic started by: fordem on May 19, 2009, 08:57:56 AM
-
I discovered (by accident) that it is possible to have one drive formatted ext2 whilst the other is ext3. I had removed one drive in an ext2, RAID1 array and then formatted the remaining drive, first in ext2 and then ext3 and then replaced the "missing" drive to see if it would prompt for a format (it didn't).
Here are some read/write performance numbers using NASTester 0.4 - the test end-point is an IBM xSeries 206 server running Windows Server 2008 with 3 Seagate 250GB drives in a RAID5 array on an IBM ServeRAID 7t controller (a relabelled Adaptec 2410SA), 3GHz P4HT processor and 3 GB RAM, onboard Intel PRO1000CT configured for 9000 byte frames through a Netgear GS108T.
These tests are done by mapping a drive letter to each volume and then simply selecting the drive you wish to write to in NASTester, so no changes in hardware or configuration.
200MB file size ..
ext2 - writes at 19.59MB/sec average, reads at 14.66MB/sec
ext3 - writes at 14.69MB/sec average, reads at 14.63MB/sec
1000MB file size
ext2 - writes at 19.34MB/sec average, reads at 21.46MB/sec
ext3 - writes at 14.8MB/sec average, reads at 19.55MB/sec
2000MB file size
ext2 - writes at 7.43MB/sec, average reads at 22.13MB/sec
ext3 - writes at 6.55MB/sec, average, reads at 23.32MB/sec
I don't recall that significant a drop in write speed with a 2GB file size prior to 1.08 ???
-
Honestly that does not look like a huge performace drop to me. Also, there are a few factors to take into consideration as well. First and most important would be the kernel version which I thought was 2.6.11 or something.
There have vast performance improvements in EXT3 since at least 2.6.20 so I think that if the developers add a newer kernel performance should be vastly improved.
Just my opinion though
-
Honestly that does not look like a huge performace drop to me.
I'm not sure if we're referring to the same thing ...
Look at the write speed when the file size is 1.5GB (19.3MB/sec) and 2GB (7.43MB/sec), that is a significant drop, and if I recall correctly, prior to 1.08 it was not that big.
-
If I understand what your talking about is the difference in speed between ext 2 and ext 3?
-
Thanks for this testing. Finally, the difference on the speed are so little, i preferd sacrifice the speed, and get more stability....
-
For reference and in relation to Fordem's post there is no significant performance drop with 2GB files on v1.07/ext2/9k jumbo frames.
Running warmup...
Running a 200MB file write on drive z: 5 times...
Iteration 1: 17.25 MB/sec
Iteration 2: 19.05 MB/sec
Iteration 3: 19.02 MB/sec
Iteration 4: 18.96 MB/sec
Iteration 5: 19.19 MB/sec
------------------------------
Average (W): 18.69 MB/sec
------------------------------
Running a 200MB file read on drive z: 5 times...
Iteration 1: 25.86 MB/sec
Iteration 2: 23.49 MB/sec
Iteration 3: 26.28 MB/sec
Iteration 4: 25.35 MB/sec
Iteration 5: 26.23 MB/sec
------------------------------
Average (R): 25.44 MB/sec
------------------------------
Running warmup...
Running a 2000MB file write on drive z: 5 times...
Iteration 1: 18.66 MB/sec
Iteration 2: 18.34 MB/sec
Iteration 3: 12.78 MB/sec
Iteration 4: 18.46 MB/sec
Iteration 5: 18.65 MB/sec
------------------------------
Average (W): 17.38 MB/sec
------------------------------
Running a 2000MB file read on drive z: 5 times...
Iteration 1: 27.97 MB/sec
Iteration 2: 27.62 MB/sec
Iteration 3: 27.71 MB/sec
Iteration 4: 26.49 MB/sec
Iteration 5: 27.75 MB/sec
------------------------------
Average (R): 27.51 MB/sec
------------------------------
-
If I understand what your talking about is the difference in speed between ext 2 and ext 3?
The purpose of the exercise was to determine how much of a penalty was created by the journalling - or as you put it the difference in speed between ext2 & ext3, however, it also seems to have revealed an anomaly, either in the new firmware or in my testing.
With 1.08 firmware there is a decrease in write speed with 2GB files as compared to 1.5GB files that I do not recall being there with 1.07, I have not yet reflashed 1.07 to verify this.
-
I just read somewhere on the internet:
"EXT3 is a journalling file system, which tends to speed up file operations and recovery. Unfortunately it requires more memory and CPU time, so it is not so good on earlier machines. It is much faster on later machines, and more easily recoverable."
Maybe this is the reason why writing and reading with EXT3 is slower then with EXT2?
Also here some nice information about EXT3
http://www.linuxplanet.com/linuxplanet/reports/4136/1/
"Using the "journal" mode requires that an ext3 filesystem write every change to a filesystem twice - once to the journal, and then again to the filesystem itself. This can reduce the overall performance of your filesystem, but is the mode most beloved by users, because it minimizes the chances of losing changes to your files since both metatdata and data updates are recorded in the ext3 journal and can be replayed when a system reboots."
-
So what it comes down to is a performance hit or a reliability hit. I think I will go with the performance hit.
-
Tested the FTP file transfer on the DNS-323 with firmware 1.08 with different conifgurations
TLS/SSL activated
Gigabit wired LAN connection via DIR-655.
Two file sizes where tested. 150 MB file and a 1,2 GB file.
Made two to three transfers and too the average sped.
EXT2 - RAID0
Sending to DNS-323 = 20,09 MB/sec
EXT2 - RAID1
Sending to DNS-323 = 16,96 MB/sec
Receiving from DNS-323 = 7,72 MB/sec
EXT3 - RAID0
Sending to DNS-323 = 13,71 MB/sec
Receiving from DNS-323 = 7,32 MB/sec
EXT3 - RAID0 - Jumbo Frames = 9000
Sending to DNS-323 = 11,35 MB/sec
Receiving from DNS-323 = 8,03 MB/sec
EXT3 - RAID1
Sending to DNS-323 = 11,95 MB/sec
Receiving from DNS-323 = 7,95 MB/sec
Performance EXT2 vs EXT3
RAID0 - Sending files = -19,15%
RAID1 - Sending files = -29,53%
-
I'm curious why the slow performance reading from the drive? I have my DNS-323 on a gigabit network, 4k jumbo frames, and I get a LOT better performance from a RAID1 array! I just dragged a 642mb file
642 MB (673,941,504 bytes)
From the DNS-323 to my desktop, it was finished in 31 seconds. That computes out to 20.73 mbytes/sec reading from the DNS-323. I did copy it several times with identical results. Each time I copied another file to flush any buffers.
7.72 mbytes/sec sucks IMO! :o
-
gunrunnerjohn - does your post refer to the one immediately above it?
One possible cause is the write speed of the disks in the system receiving the transfer.
-
Correct, it was the post right above. It's hard to imagine a SATA hard disk that would limit the performance like that, I sure don't know of any. ???
-
Perhaps I should rephrase then - one possible cause is the write speed of the disk subsystem in the system receiving the transfer.
One of the desktops I use is a Dell Optiplex - 3.06GHz hyperthreading Pentium IV, 1GB RAM and onboard broadcom gigabit - equipped with an 80GB SATA Maxtor disk, copying from the DNS-323 to the Optiplex maxes out at 5~6MB/sec.
I have several of those Maxtor drives and the drive itself is not the bottleneck.
-
I'm still not getting how this is possible. Since you state the drive is not the problem, clearly the DNS-323 is capable of faster speeds in any disk configuration, and even with any firmware version. So, what is the explanation of the slow speed? I have never seen those speeds, even with 100mbit connections to the DNS-323!
-
The way I explain it is that the disk transfer is like a chain ...
You start with a file on a disk, and read that file, then you transfer the data through the disk interface, through the computers expansion bus, through a network interface, across a network, through a second network interface, across a second expansion bus, through a second disk interface, and finally write the file to the second disk.
The maximum throughput is limited to that of the slowest channel in the chain - things like network drivers can have an inexplicable impact on throughput - I've seen Realtek gigabit NICs max out at 70 mbps (that is not a typo) when used with a generic Microsoft provided driver.
The same no doubt goes for disk controller drivers, but how often do we pay attention to those?
-
Well, I suspect we can agree that the SATA disks or the DNS-323 is not the cause ot 7byte/sec transfers on a gigibit link. As far as the system at the other end, obviously, a serious bottleneck there can be a problem. Truthfully, I have never seen an example as egregious as you state, although I have seen gigabit NIC's and drivers that don't come close to delivering the performance you'd expect.
-
Hold on a sec - maybe you want try some "small file" tests - and be warned, the results are not going to be pretty.
I've used Microsoft's SyncToy to "back up" several gigabytes of mixed files, including a few thousand html pages - average size 36KBytes - throughput drops to something in the region of 0.5KByte/sec, and the limiting factor IS the DNS-323.
-
Well, I don't have enough small files like that to do a meaningful test, but if you really want to stack the deck to prove a point, it's always possible.
I made the mistake of thinking we were talking about the tests performed and reported on in this thread. In case you forgot...
Two file sizes where tested. 150 MB file and a 1,2 GB file.
Made two to three transfers and too the average sped
I don't see any tiny files there, so I'm still not buying your arguments.
-
I wasn't stacking the deck - back up (using SyncToy and other utilities) has been the primary function of my DNS-323 - the data just happens to be some 14GBytes worth of html pages.
When you have a "system" capable of transferring data from Disk_A in Device_A to Disk_B in Device_B at speeds exceeding 30MBytes/sec in a repeatable test, and "real world throughput" is to the order of 0.5KBytes/sec, some of us wonder why, and some of us investigate.
It also really doesn't matter whether you "buy" my arguements or not - I'm not argueing, just offering suggestions to explain the phenomenon you find curious - feel free to ignore me - you will, if your pursue it, eventually arrive at conclusions very similar to those I have arrived at, I've just had a few years headstart on you.
-
Again, we were addressing the scenario presented in this thread, you're dragging in a red herring that has nothing to do with the benchmarks provided in the thread.
Do you have anything constructive to add about this person's issue with performance? Let's forget about your zillions of little files and address the scenario here.
-
Perhaps you want to reread the thread in it's entirety ...
I'm the thread starter and posted the perfomance degradation I observed using ext2 & ext3 formatted disks - Geraner did likewise - the only person who has expressed any questions on performance was you, hence my offering you a potential explanation.
-
Well, you're right, I didn't realize that the other guy jumped in the middle, sorry. :)
Since I have no performance issue, and don't expect tons of tiny files to transfer at the same speed as large transfers, I'll just exit through the left hand door.
-
You may not, I certainly don't, but you'd be surprised at how many less experienced people do - which is why I took the trouble to point it out.