Print

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

[benn600]

On my Vista server I get absolutely horrendous speeds when I run speed testing programs.  I have seen as low as ONE Megabyte per second when writing a 128 MB file to my big RAID array.  I can easily get 30 on multiple computers simultaneously over gigabit networking.  Generally I get 10 -20 MB/second when running tests.  I have tried DiskBench and Disksped.

Running these same programs on an XP machine with a simple, single SATA/IDE drive will provide 20-50 MB/second rates.  You'd think that I'd be getting hundreds of megabytes per second.

I have seen two hundred megabytes per second before but it's usually in the OS and lately, my OS RAID 5 array has not been providing very good rates, either.  I have to wonder if it's Vista?!  Any ideas?!

[horse]

You have the Promise PCIe 16 ports SATA II card?
You are using the on board GigE controller?

If any of the above are not true, what card and what slot is it plugged into? (32/64 bit and speed 33/66/etc.)

I can easily get 30 on multiple computers simultaneously over gigabit networking

These are clients reading files from the same RAID that is slow locally, but across the network?
Read Only or running the disk speed utility on the RAID as a network drive?

Are all the drives SATA II (I assume it is the SATA Promise controller)

How is the cahe configured? Not knowing the Promise product I would hope you can enable write cache, but then it may (Like HP) only let you do that when the cache has Battery Backup connected to the RAID card.

Any errors in the Promise logs. (My guess is there will be no disk error reported in Vista as it will be hidden by the promise hardware.

If you have this card http://www.promise.com/product/product_detail_eng.asp?product_id=166 then it is an 8 lane card and should be getting close to 2GB on the bus. With SATA II drives, this thing should be smoking fast

[benn600]

You are exactly correct.  I'm using the Supertrak EX16350.  Sorry for not including more in my first post...woops.  But you're right on.
30 Megabytes per second (I meant to say) on multiple computers simultaneously, indicating 60 MB/second + read rates.

With SATA II drives, this thing should be smoking fast

That's what I was thinking!

How much difference do you think SATA II makes?  The card is reporting SATA II on all drives and I definitely believe that to be true.  The new drives I bought were definitely SATA II and the older ones I was almost positive that they were.

For cache settings, each drive has a "Write Cache Enabled" setting and an "NCQ Enabled" setting.  I have both set to enabled on all drives.

Here's some information on the Logical Drive, consisting of 16 500GB drives in RAID6 configuration.

Assigned Name   Logical Drive 1
RAID Level   RAID 6
Sector Size   512 Bytes
Stripe Size   64 KB
Parity Count   2
Capacity   6999.99 GB
Status   Functional
Background Activity   Idle

It's currently set to Write Through, which is slower but more secure.  Write Back adds the cache I guess.  I don't have the BBU but I do have a UPS on the server.

I'm confused about the drive cache and the card cache.  Is the card cache even used with the safer mode?  With both caches, that's 512MB total cache because each drive has 16MB.

It failed drive 16 two days ago and I replaced it (successful rebuild) but now I'm using the same drive in an external enclosure and it formatted fine and is holding data right now (450GB--almost full).

The server array is:
3.96 TB used
2.39 TB free
6.36 TB capacity
[62% full]


Back to the smokin fast assumption.  I would really think so, too.  Are there any other free speed test programs I could run?  Part of me just wants a number so I can know that it is performing as it should.  At one point I was getting 200 MB/sec duplication speed (read/write same logical drive).  If each drive is capable of 45MB/sec sustained transfer rate (a reasonable average), that would mean 630 MB/sec absolute max speed (this doesn't include the two parity drives).  So maybe I could hope for 500 MB/sec?  I really don't have any source or destinations capable of anything close to this so it's not a huge deal but it would be nice to know what it's capable of.

The biggest problem is that gigabit networking is the main hookup ... which is fast but can't compete with what this array should be producing.  Also, I'm usually reading or writing a single disk drive as I'm moving data around so the limiting factor is my single drive.  But, I should be able to get full speed access on multiple computers.

What is above gigabit ethernet?  If I buy a gigabit switch, what's the least expensive way to connect the server to the switch so the switch can be delivered more speed and then it would condense down to gigabit per computer?  What about 10Gbit to the switch?  I am absolutely not needing this but it would be interesting to learn more about what's next.  Or what about dual Gigabit connections?  Gigabit is about 125 MB MAX so 250 would be much closer to what I would say is a 500 MB max.  10GBit would be a gigabyte per second+...wow.

Addition: For errors in the logs...yes, I'm having some errors.  It is always Disk 4 and it says there is a bad sector (BSL) but it seems like it finds the BSL but then immediately clears it.  I see it during a Fix (Synchronization) and randomly.  Perhaps it's a specific section of the disk.  I've thought about taking Disk 16 that failed a few days ago (but is working fine externally) and putting it in instead of Disk 4 or simply buying a new disk.  I get sick of reading the error (and getting about 4 emails whenever it encounters it) but also it is probably an initial sign of failure.

It does bother me that drive 16 failed even though it's working fine now!  Any thoughts on this?

2nd Addition: The My Book's (Essential) that 10 of the drives came out of are, surprisingly enough, the WD RE2 series drives.  So apparently, WD puts their higher end drive (or put--haven't tried a My Book recently) in their external enclosure.  This is one of the drives that failed.  I think I had SMART disabled when the drive failed but I turned it back on for no defined reason.

[newsposter]

don't overlook the distinct possibility that a cheap mobo has a cheap mobo chipset which cannot handle the kinds of data transfers you're asking it to do.

[benn600]

lol.  It wasn't "cheap."  It was moderately expensive.  It's a Fata1ity board, too.  Remember, it's the difference between what I can get through the network port vs. what local speed test utilities report.  It must have something to do with Vista.

Addition: With a gigabit connection, I just moved 20 GB from my workstation to the server.  The speed fluctuated from 20% to 36% in the network monitor (of gigabit).  In the end, my calculated speed is 25 MB/sec.  If it would have stayed at 36% or so-and it does from time to time--it would have been more like 40 MB/sec.

But with a local hard drive speed tester on the server, I get under 20 MB/sec!

[ThoBar]

You used this...?

netsh int tcp set global autotuninglevel=disabled

I doubt it is the problem, but just to eliminate any Vista 'smarts' getting in the way....

[newsposter]

cheap has no relationship to price.

I've found that buying stuff because some celebrity endorses it is a waste of time.  Cars, phones, or computer hardware.  Remember that part of your $$ always goes back to the 'celeb' doing the endorsements.

[benn600]

I picked the board because it didn't have some of the junk I don't want (Parallel, serial).  And it has two 16x PCIe slots.  I didn't pick it for the name.

Plus, it was a moderately priced board with all the features I was after.

I don't care how cheap the board is, 1MB/sec is not normal.  Nor is 20!  It's capable of more.

Thank you confishy.  If I recall correctly, I already ran that but I will try again.  My server is currently running chkdsk because I wanted to check everything over as maintenance (I initiated it).  I selected the full boar run.  It's been running 10 hours and looks like it has a way to go still.

[KingSparta]

You Should Try To Isolate all the drives To See If The Speed Changes When Other Drives Are Disconnected From The Interface.

I Ran Into A Problem Once Where One Drive That Was Bad, Slowed The Speed Of All Drives Connected On The Same Interface.

In My Case It Was A External Drive Connected Using Firewire 800.

[horse]

So you can read from the array at 30MB/s, but the speed test program idicates a much lower rate.
Is the test program doing a read, write or read write test?
If disk 4 is have bad sectors, take it out of the array and see what happens.
If you have a UPS on the server, then enable the write cache, this will speed up write functions as there is no wait time. It will also potentially hide any issues with bad disks :-)
The reason for the BBU is to avoid corruption should data still be in the write cache and the server losses power. A risk, still with the UPS, but much lower.

TBH, I still suspect you are running into issues of insufficient power on one of the supply rails. (12v or 5v) that is causing writes to fail on the disks.

Also, upgrading to 10GigE, Ether Channel or Fiber Channel and a SAN (Storage Area Network) and using iSCSI or FCIP will not provide you any advantage with a single server. This is what we use in our Datacenters with PB (1000TB) of storage all virtuallized for hundreds of front end servers.

[benn600]

if I set the faster mode it reverts on reboot. Maybe I should get the bbu.

[benn600]

Okay, I realized some things.

First, the reverting back to the slower but more secure mode is apparently interlinked with the Vista Advanced Performance option in device manager on the drive.  I changed it and WebPam (for the RAID array) now shows the faster mode.

I also ran some tests simply taking a DVD of VIDEO_TS files (probably 6GB) and copying it from the array to itself.  Vista was saying 95 MB/sec.  So this would be read & write, providing 190 MB/sec total transfer rate, correct?

The only other drive in the system is the 640GB RAID5 array.  However, it is using the motherboard RAID card.  So this won't have very good performance.  I'm now running a copy of that DVD I've been copying around to the other array and am only getting ~14 MB/sec.  That's obviously very slow.  Does this seem possible?  Such a slow transfer rate?  I strongly believe there is other minor activity going on at the same time on this motherboard-driven RAID array.  My four cameras are saving their data to it at the same time.  I just stopped the camera recording and the speed did not change more than 1 MB/sec

For someone who lives around this RAID stuff, what speeds might I expect?  The RAID6 array is 16 drives so the theoretical maximum would be the sum of the speed of 14 drives, right?  There will obviously be overhead with the parity calculations, though.  So at 40 MB/sec * 14, that would be 560 MB/sec.

The biggest problem is I don't have a good source for writing or reading from the drive and just throwing the data out.  Every program I've tried reports horribly low numbers (< 20MB/sec).

And the onboard RAID system seems really slow so copying to or from it isn't a good test at all.  And a single drive via USB or anything is really slow as well.

Copying FROM the slower RAID5 array yields slightly greater speeds than RAID6->RAID6.  I get about 98 MB/sec.  It's 3 drives in RAID5 so that would make some sense that it would be two drives of speed...but now that I think about it, the parity is distributed so it could actually provide more than 2 drives of speed!  So this RAID5 array has good read speeds but not write?  That seems very logical but the write speeds seem too low.

-------------------
Addition: Too bad I don't have my data backed up so I could just delete the array and try things.  I am almost positive I was getting 190 MB/sec data duplication speed on this array during my initial testing phase with this card and Vista.  That's just over double the speed and would yield 380 MB/sec...a much more respectable speed.

[benn600]

[I'll keep adding single lines as I try more tests.]

In attempting to troubleshoot this, I have disabled NCQ and the write cache on each drive.

As I started disabling both on drives, the speed went way down to 30MB/sec.

NCQ and Write Cache disabled on all drives: 50 MB/sec

With write cache enabled on all drives but no NCQ: 75 MB/sec

Good: if I am self-duplicating data on the RAID6, note the speed, and then ADD a copying process from the other RAID5 drive to RAID6, the first speed doesn't change very much.  It looked like no change but I guess there is a very small change.  Now, after changing everything back to how it was (cache & NCQ enabled), it isn't running as fast as it was before the changes and reverted changes.

[benn600]

More big discoveries.  I am now using HD Tuner.  It looks like a much better program compared to all the other applications I've tried.  With it, I'm getting a minimum read rate of 97 MB/sec with a maximum read rate of 390 MB/sec.  That seems very nice and fast.

This is using the largest block size of 8MB in HD Tune with the most accurate sample.  With the fast sample, higher numbers are reported because it shortens the check time so it registers short bursts on the max scale.

Now I'm even more confused.  Some speeds are good some are bad.