ddrueding
Fixture
If they are big files (and BD ISOs certainly count), you should be in the thick section of that graph (150MB/s+)
Windows Home Server / Home Server OS
- Thread starter Stereodude
- Start date
Stereodude
Not really a
Maybe...
It depends on the block size being used by the Windows file copy mechanism. Obviously small files are written using small blocks, but I have no idea about large files. Have you tried FastCopy?
No problem.
OK, if I concentrate on just one big file and use fastcopy over the network (via mapped drives) I see the same thing, starts out 75MB/s and after 30 seconds it is doing 18-25MB/s.
BUT if I do the same thing on the server from the mirror to the softRAID5 array with one big file I can get about 100MB/s. Maybe I didn't test this part very well previously? Now I am moving a similar set of files from the mirror to the softRAID5 array and still getting 96MB/s.
Now I am really confused, what problem gets good iperf results but bad actual file transfer results?
BUT if I do the same thing on the server from the mirror to the softRAID5 array with one big file I can get about 100MB/s. Maybe I didn't test this part very well previously? Now I am moving a similar set of files from the mirror to the softRAID5 array and still getting 96MB/s.
Code:
TotalRead = 69295.9 MB
TotalWrite = 69295.9 MB
TotalFiles = 713 (58)
TotalTime= 716.72 sec
TransRate= 96.68 MB/s
FileRate = 0.99 files/sNow I am really confused, what problem gets good iperf results but bad actual file transfer results?
I don't think iperf writes anything to the actual disk subsystem which is why you can see great network transfer rates, but different numbers when writing to the disk.
When I've tested the network component in my house, I used a RAM drive on both ends of the line when doing file-copy through windows machines. Once I knew those numbers, I then moved to using the RAID array to see what it was capable of in terms of file transfer. I never had good luck with Home Server, Server 2003, Server 2008, or Windows Storage server with network transfer performance. Only when I install OpenFiler and retested my transfer speeds did I see much better network file transfers rates.
Stereodude
Not really a
Are you pulling or pushing the data? I've found that can make a significant difference in copy speed. IE: Machine A pulling data from Machine B can be considerably faster than Machine B pushing data to Machine A.
How big of a buffer did you configure FastCopy to use?
In your test above it seems like the drive you are writing to can not sustain the data rate, so once the windows cache fills up it slows way down.
iperf doesn't put the data anywhere, so if you get good iperf results and bad results with a file copy your disks are the problem.
I am working on hard numbers to present right now but my theory at the moment is when I combine the network plus the softRAID5 I get crappy transfer. The network transferring to the server ICH9R mirror is fast and transferring from the ICH9R mirror to the softRAID5 is fast.
I am pushing the files from the client to the server, as I would do in normal use. Fastcopy defaults to a 64MB buffer, I haven't changed it.
I am pushing the files from the client to the server, as I would do in normal use. Fastcopy defaults to a 64MB buffer, I haven't changed it.
I guess this right here is the heart of the problem, but I am at a complete loss for what to do about it. All done with the same BD movie rip in file mode. Seems like it is only a problem with the network + softRAID5. Separately each is fine!!??
Pushing from client to server ICH9R mirror using fastcopy I can get this:
Pushing from client to server softRAID5 using fastcopy I get this:
From server ICH9R mirror to server softRAID5 using fastcopy I get this:
Pushing from client to server ICH9R mirror using fastcopy I can get this:
Code:
TotalRead = 35006.1 MB
TotalWrite = 35006.1 MB
TotalFiles = 898 (37)
TotalTime= 438.68 sec
TransRate= [b]79.80 MB/s[/b]
FileRate = 2.05 files/sPushing from client to server softRAID5 using fastcopy I get this:
Code:
TotalRead = 35006.1 MB
TotalWrite = 35006.1 MB
TotalFiles = 898 (37)
TotalTime= 1902.95 sec
TransRate= [b]18.40 MB/s[/b]
FileRate = 0.47 files/sFrom server ICH9R mirror to server softRAID5 using fastcopy I get this:
Code:
TotalRead = 35006.1 MB
TotalWrite = 35006.1 MB
TotalFiles = 898 (37)
TotalTime= 362.14 sec
TransRate= [b]96.66 MB/s[/b]
FileRate = 2.48 files/sStereodude
Not really a
What NIC are you using?
What are the results if you have the server pull from the client?
What are the results if you have the server pull from the client?
NICs
Client has two Realtek RTL8111D on the Gigabyte GA-X58A-UD5 (rev. 1.0)
Server has two Intel 82574L on the Supermicro X7SPA-HF
As an afterthought jumbo frames are disabled on the client, I don't see the option to turn that on for the server. Actually I don't see the option to turn any of the hardware LAN options on/off on the server inside device manager.
Will run the test when the client is free of doing other things, be a little while.
Client has two Realtek RTL8111D on the Gigabyte GA-X58A-UD5 (rev. 1.0)
Server has two Intel 82574L on the Supermicro X7SPA-HF
As an afterthought jumbo frames are disabled on the client, I don't see the option to turn that on for the server. Actually I don't see the option to turn any of the hardware LAN options on/off on the server inside device manager.
Will run the test when the client is free of doing other things, be a little while.
NICs
Client has two Realtek RTL8111D on the Gigabyte GA-X58A-UD5 (rev. 1.0)
Server has two Intel 82574L on the Supermicro X7SPA-HF
As an afterthought jumbo frames are disabled on the client, I don't see the option to turn that on for the server. Actually I don't see the option to turn any of the hardware LAN options on/off on the server inside device manager.
Will run the test when the client is free of doing other things, be a little while.
Client has two Realtek RTL8111D on the Gigabyte GA-X58A-UD5 (rev. 1.0)
Server has two Intel 82574L on the Supermicro X7SPA-HF
As an afterthought jumbo frames are disabled on the client, I don't see the option to turn that on for the server. Actually I don't see the option to turn any of the hardware LAN options on/off on the server inside device manager.
Will run the test when the client is free of doing other things, be a little while.
Stereodude
Not really a
I've never found any use for Jumbo frames.
Neither have I but I hear if they are on and not everything supports it that will kill performance.
Server pulling from client to softRAID5 using fastcopy.
Server pulling from client to softRAID5 using fastcopy.
Code:
TotalRead = 35006.1 MB
TotalWrite = 35006.1 MB
TotalFiles = 898 (37)
TotalTime= 644.98 sec
TransRate= [b]54.27 MB/s[/b]
FileRate = 1.39 files/sStereodude
Not really a
That's almost 3x faster by changing the push to a pull.
ddrueding
Fixture
I oscillate between local and network storage. Before GBe I was local, then went network until drive capacities got large enough that I didn't need a massive array. Then I needed a massive array regardless. Now I'm back to local because the array is back to only 6 drives and will outrun GBe. It also doesn't hurt that the workstations have enough RAM and CPU to handle background tasks without bother (BT, encoding, etc).
ddrueding
Fixture
Following up on the "made for RAID" subtopic above:
Hitachi 3TB drive: $200
Hitachi 3TB "made for RAID" drive: $350
That is way more than a minor premium, that is a full blown screwing. Same performance, same cache, same warranty, just a different firmware. Screw that.
Hitachi 3TB drive: $200
Hitachi 3TB "made for RAID" drive: $350
That is way more than a minor premium, that is a full blown screwing. Same performance, same cache, same warranty, just a different firmware. Screw that.
There has to be a megamaid suck to blow joke in there somewhere.
I think I will try the server 2008 eval I have, but I see no way in hell I will switch to such an expensive OS.
After that back to 2003 and PERC I guess.
I would be surprised if you see much difference with server 2008. I should have written down all my test numbers when I used it. If you do decide to go with a Perc, I bought two of the 6is from a guy named borito4 on eBay. He has a used Perc 5i going right now with about 15 minutes let.
There is a used Perc 6i for $150. That's what I paid about a year ago for a new one.
blakerwry
Storage? I am Storage!
This could be an interrupt sharing/contention problem or a chipset bug/limitation. Wouldn't hurt to ensure you have the latest bios/firmware/drivers (get NIC drivers from NIC manuf) for each system.
Could also be a windows setting... like the general performance option for applications vs services (my computer properties ->advanced ->performance) or the network performance setting for file server vs application server (somewhere in networking, perhaps properties for client for ms networks)
I think you are on to something there. Both NIC 1 and the Marvel SAS controller (AOC-SASLP-MV8 ) are IRQ16. I am going to find another cable and connect NIC 2, disable NIC 1 in windows and retest with that.
The IPMI stuff if hard wired to NIC 1 so I can't just abandon that NIC. I also found the advanced settings for the Intel NIC on the server, confirmed that jumbo frames were off on that and all the checksum offloading was on.
sdbardwick
Storage is cool
Adjusting or even turning off "interrupt moderation" (if such a setting is available for your NIC) can help also boost XFER rates. My early Intel desktop 1000BT NICs perform much better with it off.
Oh well, no difference on switching the NIC.
On the optimizations stuff it looks like typical server, background server and system cache for the computer. On the file and printer sharing it was already on maximum data throughput for file sharing.
It just occurred to me that I am totally hosing this Home Server thread when I don't exactly have home server anymore here and I am doing troubleshooting.
On the optimizations stuff it looks like typical server, background server and system cache for the computer. On the file and printer sharing it was already on maximum data throughput for file sharing.
It just occurred to me that I am totally hosing this Home Server thread when I don't exactly have home server anymore here and I am doing troubleshooting.
Still using NIC2 Turned off Flow Control and no difference. Still using fast copy here but not letting it complete, just waiting for it to settle down in reporting 20MB/s or less.
TX&RX flow control back on but turned interrupt moderation from adaptive to high, according to the built in help that could help high transmission rates. Unfortunately I think that actually made things worse, dropped to ~15MB/s
TX&RX flow control back on but turned interrupt moderation from adaptive to high, according to the built in help that could help high transmission rates. Unfortunately I think that actually made things worse, dropped to ~15MB/s
Mercutio
Fatwah on Western Digital
Replacement power supply for an Acer H340 = $120 by the time shipping is added.
Ouch.
Ouch.
ddrueding
Fixture
Very much ouch.
Bozo
Storage? I am Storage!
I guess shipping is very high when it come from BORG land. :-D
Mercutio
Fatwah on Western Digital
An interesting observation: Since I have this little Acer box here I've been playing with it. I've found that while the 2TB Seagate Green drives work like ass in RAID arrays, they're much better behaved with WHS. I stuck an eSATA card in and I presently have six 2TB drives connected. They all work just fine. That's particularly remarkable given that I couldn't get the same drives to even manage RAID1 properly.
Mercutio
Fatwah on Western Digital
Apropos of nothing, a Chenbro ES34069 is a mini-ITX chassis with 4 hot swap 3.5" bays, an internal 2.5" bay and a slim optical bay. They're sold with 120 or 180W PSUs and make a lovely foundation for a WHS or FreeNAS setup.
ddrueding
Fixture
That would make a great 9tb backup machine as well. Might have to do some of those.
Mercutio
Fatwah on Western Digital
That was my thought as well. Though not while 3TB drives are selling for $250 apiece.
There's a picture of somebody's appliance on the FreeNAS site. It just took me a little while to figure out what the chassis was.
Incidentally, if you drag a picture in to Google Image Search, Google will search for that picture and other things that are visually similar.
ddrueding
Fixture
Awesome. Thanks!
Mercutio
Fatwah on Western Digital
Both HP and Acer appear to have exited the WHS Appliance market. Home Server 2011 is a relatively new product but it doesn't look to me like any tier-1 OEMs are supporting it these days. Has anyone tried the 2011 version? I have a SOHO customer looking for remote desktop access to machines running Home Premium versions of Windows, which is something that IIRC Home Server provides. I'd rather point them to WHS than deal with potentially unreliable hacks to enable RDP.
Mercutio
Fatwah on Western Digital
I had the opportunity to test WHS 2011. A few things to say:
1. It seems to disable HDMI audio on every reboot. Which is hilarious for something that could potentially be used for a media center. Whoops.
1a. It's happening on both Intel and ATI hardware, so I don't think it's a "ha-ha bad ATI driver" thing.
2. There just aren't any interesting storage pooling options. You can do SoftRAID like Windows Server 2008. Windows 8 has interesting storage pooling options. So did old WHS. Just not this one.
3. Application support is essentially as per Server 2008. You didn't really want Antivirus software, right?
4. There seems to be a 10 connection limit just like Professional desktop versions of Windows. Which makes me ask what the hell the point is. It basically has all the drawbacks of desktop Windows and Server Windows and essentially none of the advantages.
4a. Yes it has TS Gateway and the automated backup craps. Neither of which are needed for my application.
So anyway, I'm going to chuck this and move it back to Windows 7. Thank crap I have Technet and didn't actually pay for this thing.
1. It seems to disable HDMI audio on every reboot. Which is hilarious for something that could potentially be used for a media center. Whoops.
1a. It's happening on both Intel and ATI hardware, so I don't think it's a "ha-ha bad ATI driver" thing.
2. There just aren't any interesting storage pooling options. You can do SoftRAID like Windows Server 2008. Windows 8 has interesting storage pooling options. So did old WHS. Just not this one.
3. Application support is essentially as per Server 2008. You didn't really want Antivirus software, right?
4. There seems to be a 10 connection limit just like Professional desktop versions of Windows. Which makes me ask what the hell the point is. It basically has all the drawbacks of desktop Windows and Server Windows and essentially none of the advantages.
4a. Yes it has TS Gateway and the automated backup craps. Neither of which are needed for my application.
So anyway, I'm going to chuck this and move it back to Windows 7. Thank crap I have Technet and didn't actually pay for this thing.
I have to say my nlited WHS 2003 has been a rock solid file server for me. Of course nlite removed the WHS pieces and basically left me with server 2003. I thought about putting server 08 on it from technet but I have no reason to, and I agree the things that home server adds for the 2008 version don't interest me either.
Mercutio
Fatwah on Western Digital
The thing is, it's ridiculously easy to make a decent file server. I understand that WHS doesn't really cost more than Win7 Pro, but the only compelling argument for it is the automated backup. It's a little like a solution in search of a problem to solve.
Is FreeNAS your recommendation for a decent file server or do you prefer something else?