Software RAID and fast SCSI discs
Discussion
Any feedback from hardware brainiacs would be much appreciated. This is a bit of a long shot for a car forum, but here goes...
I've just got access to our new application server. The hardware boys who bought it (with my money, without asking permission) are being rather defensive and I don't believe I'm getting straight answers. Here's the deal ...
The PC is a Proliant DL380. It has been fitted with six 15k rpm discs using the a Smart Array 6i SCSI controller (the built-in default option on that box). 4 of the discs have been set up in a RAID 5 array giving a useable storage capacity of just over 400GB. As you can imagine, those 6 discs cost a lot of money. If I copy a 200MB file from one folder to another folder on that RAID 5 array, it is taking over 20 seconds. I'm getting a data through-put of only 8MB/s. Now that is really very shite. My workstation with a single SATA disc can do the same thing at 50MB/s.
The hardware guys swear blind that that Smart Array 6i controller is a hardware RAID controller. As far as I can see it is a basic U320 SCSI controller and it doesn't implement RAID. If I'm right, then the machine must be running RAID 5 in software.
If that is true, would that by itself explain why these high speed and expensive discs are giving such crap performance?
Any thoughts or experience shared would be very much appreciated.
(Off to the pub now to cry into a pint.)
I've just got access to our new application server. The hardware boys who bought it (with my money, without asking permission) are being rather defensive and I don't believe I'm getting straight answers. Here's the deal ...
The PC is a Proliant DL380. It has been fitted with six 15k rpm discs using the a Smart Array 6i SCSI controller (the built-in default option on that box). 4 of the discs have been set up in a RAID 5 array giving a useable storage capacity of just over 400GB. As you can imagine, those 6 discs cost a lot of money. If I copy a 200MB file from one folder to another folder on that RAID 5 array, it is taking over 20 seconds. I'm getting a data through-put of only 8MB/s. Now that is really very shite. My workstation with a single SATA disc can do the same thing at 50MB/s.
The hardware guys swear blind that that Smart Array 6i controller is a hardware RAID controller. As far as I can see it is a basic U320 SCSI controller and it doesn't implement RAID. If I'm right, then the machine must be running RAID 5 in software.
If that is true, would that by itself explain why these high speed and expensive discs are giving such crap performance?
Any thoughts or experience shared would be very much appreciated.
(Off to the pub now to cry into a pint.)
I beg to differ:
http://h18004.www1.hp.com/products/servers/proliants
So, its hardware RAID (as I suspected).
Unless they've used NONE of the RAID features at all on the card and used software. Take a look in Disk Manament, that will tell you.
Greg
http://h18004.www1.hp.com/products/servers/proliants
So, its hardware RAID (as I suspected).
Unless they've used NONE of the RAID features at all on the card and used software. Take a look in Disk Manament, that will tell you.
Greg
All our DL380s have raid natively. I don't know for certain but the inclusion of the word array in the name of the controller suggests to me that it's hardware raid.
If you've got access to the box at console then go into array manager (program files > HP) and have a look at the config. Also, look at the disk section of Computer management, that should tell you if they're software raid.
If you've got access to the box at console then go into array manager (program files > HP) and have a look at the config. Also, look at the disk section of Computer management, that should tell you if they're software raid.
If it's a default HP/Compaq install go to http://yourservername:2301 and you can view the configuration of the RAID if you have the username and password.
The 6I is an integrated RAID controller on that model of server. It'll be basic but can be upgraded. If they have used a software RAID, then performance will suffer. TBH, it's harder to implement software raid than hardware on those boxes.
The 6I is an integrated RAID controller on that model of server. It'll be basic but can be upgraded. If they have used a software RAID, then performance will suffer. TBH, it's harder to implement software raid than hardware on those boxes.
agent006 said:
TBH, it's harder to implement software raid than hardware on those boxes.
And i'd get my arse kicked round the room if i did it on one of ours, and then would get laughed at for months afterwards.
They may have set it up wrong, but our distie always reckons the integrated raid on the 380 can bottleneck it a bit. Upgraded controllers are recommended by our distie.jamieboy said:Good point, often missed. You might assume if you copy a file from one place a server to another place on the same server, all the data transfer would occur on the server.
Are you timing the copy while logged on to the server? If not, i.e. if you're copying from a mapped drive on your workstation, then (I think) the time to get your 200MB over the wire to your machine and back wll be a factor.
On a proper operating system maybe, but not Windows. It will round trip, so your file copy will run at half network speed.
BliarOut said:
Just out of interest, which controller do you use?
Internal for app servers. Bigger storage needs get 6404's and all the 740's and 580/585's have 6404's too.
585's are king BTW especially with dual cores in. They walk all over the Intel dual core 580's to the point that it's embarrassing for Intel. With one of our major apps the Intel box tops out at 29,000 (of our 'transactions') and the AMD at 81,000. With software tweeks for the 585's NUMA memory architecture over 100,000 is possible.
ATG said:
My workstation with a single SATA disc can do the same thing at 50MB/s.
I'm dubious of this figure, reading/writing to the same disk at 50MB/s sounds unlikley so I'd test that again. Maybe the read was cached? Try it after a reboot and see...
Does this controller do background/on the fly builds, if it does has it definatley finished?
victormeldrew said:
Are you timing the copy while logged on to the server? If not, i.e. if you're copying from a mapped drive on your workstation, then (I think) the time to get your 200MB over the wire to your machine and back wll be a factor.
On a proper operating system maybe, but not Windows. It will round trip, so your file copy will run at half network speed.
Unix with NFS mounts is the same. I don't know of an O/S / filesystem that does, but I'm interested to be proved wrong.
ThatPhilBrettGuy said:Proper as in mainframe. I'm old you know
Are you timing the copy while logged on to the server? If not, i.e. if you're copying from a mapped drive on your workstation, then (I think) the time to get your 200MB over the wire to your machine and back wll be a factor.
On a proper operating system maybe, but not Windows. It will round trip, so your file copy will run at half network speed.
Unix with NFS mounts is the same. I don't know of an O/S / filesystem that does, but I'm interested to be proved wrong.
victormeldrew said:
On a proper operating system maybe, but not Windows. It will round trip, so your file copy will run at half network speed.
You must be shitting me. Seriously? Bwahahahaha. That is the lamest thing I've heard yet.
Arg, never mind
I thought that the fact that the data travelled via the client and back when moving stuff around sounded really silly, but I see phil brett's point that that is how most network filesystems work... I was sure that some had to be clever, but I have found no evidence to the contrary. I can't see it being a security issue, although there has to be some reason that it is done this way. It still seems odd, although that's probably because I'm missing something fundamental (again)
Ciao
Zak
ErnestM said:
If you copied the file via :shudder: command line through a UNIX terminal login (#cp /user/stuff/foo.bar /user/otherstuff/foo.bar) then it wouldn't round trip I believe...
there's something oher than command-line for copying? A window manager is only really useful for managing multiple rxvt sessions! Ciao
Zak
ErnestM said:Correct.
If you copied the file via :shudder: command line through a UNIX terminal login (#cp /user/stuff/foo.bar /user/otherstuff/foo.bar) then it wouldn't round trip I believe...
ErnestM
Most systems I use *still* work this way - generally I use telnet or a terminal emulator running via telnet. Mainframes (proper systems) run everything on the server - where else?
Even dare I say it (in contradiction to my assertion) Windows Terminal Server does the grunt on the server. There was life before Distributed File Systems, and believe it or not there still is!
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff