Topic: 85% Speed Cut With W2K3 Server (from Vista) (Read 5841 times)

benn600 · « **on:** October 20, 2007, 12:05:39 am »

Questions at bottom in bold (PLEASE skip to the questions at the minimum)

So I should install W2K3 Server? OK! Just about exactly 24 hours later, I'm running Windows 2003 Server Standard Edition!

--STARTING with W2K3 Server ENTERPRISE x64--

Thursday, 6:00 PM: Started Major Backup Updates (everything I could)
Thursday, 11:00PM: Backups Finished
Thursday, 11:30PM: Fried Floppy Drive
> Mistakenly plugged cable misaligned (off one pin).
> Amazingly, I took 3 times as much care making sure I didn't block fans, etc....missed the power cable issue!
> Saw a spark, started smelling something
> Didn't want to cut power so waited a bit then decided to before the OS started loading (still Vista at the time)
> Floppy required for RAID drivers
Friday, 1:00 AM: Sick of being unable to boot correct RAID, OS reboots continuously, sick of floppies and extra hassle of opening case each time I want to format (add floppy)
> Plus, the mobo sucks in that it will NOT let me force it to boot the internal RAID so I always have to pull my Promise RAID card and it kinda works but I always run into issues
Friday, 2:00 AM: Decide to use single drive method--MUCH simpler but lose OS redundancy

-- any ideas here to help reduce the danger?
Friday, 5:00 AM: Sick of waiting on Service Pack / update downloads. TIME FOR BED!
Friday, 10:30 AM: Wakeup
Friday, 11:00 AM: Fight with PHP & Apache; simply cannot get PHP working correctly!
Friday, 3:00 PM: Started posting desperate questions on forums
Friday, 7:00 PM: Absolutely fed up, going to try Standard (x86) because Apache & PHP (plus others) are not x64 based so maybe there are conflicts
Friday, 11:00 PM: Completely finished! Completely updated, Apache, PHP, hMailServer, D-Link View Cam, Media Center, Juice Podcast Receiver, Shared Drives, Dynamic Update Client (no-ip), and whatever else I'm forgetting.

This has been one of the most interesting major server overhauls I have ever went through. Not to bore everyone but it's quite difficult to imagine.
We are planning a three family "major" vacation and, since our email is hosted here I had to setup a temporary server and switching meant downtime--and checking email was difficult since Thunderbird profiles are on server (now, down).

Not to mention being up till 5 AM! And destroying a floppy drive--but MORE importantly, I wonder if that could have destroyed more! It appears to not have but several hard drives are on the same cable the floppy was on! Then I was worried something would catch fire. I though: hmm, it's probably fine, I don't smell much. Then I realized that the very high powered fans were blowing over the floppy drive carrying any smell way away from me--so I had no idea how bad it was ... leading me to shut it down immediately.

Luckily the RAID data came right back. When I installed the driver and gave the drive a letter, I double clicked it and got "Access Denied" -- a MUCH more pleasant message than what the message could have said: "This drive is not formatted." So I just took ownership of the drive and it started flipping through all the data--phweww.

So I have some major questions.

I have not tried my MP101's yet but will at some point. I think the 1-track issue won't exist with W2K3 server now.

I was getting around 420 MB/sec read speeds on the array with Vista. I now, testing with the same program, get 60 MB/sec. That is absolutely not pleasant! The 16 drive array performs worse than the OS drive (single drive!) What the heck is going on?! I got the same speeds with both the attempted x64 and the current x86 operating systems.

Any other suggestions for what is in 2K3 Server Standard that I can take advantage of? I'm just using file sharing and general application running/server stuff. What's good that's built into the OS?

Now I just want to go rest...while listening to MC! I was so pleased (mainly this last time around when my work was leading to the finished product) when I opened MC and everything showed up--the server raid array was still perfectly intact!

benn600 · « **Reply #1 on:** October 20, 2007, 11:42:04 am »

It appears to have absolutely fixed the 1-track MP101 problem (Server 2003). And SHADOW COPY is absolutely amazing! I enabled it and am storing the copies on a separate, dedicated ~160GB partition. Anyone have any experiences with Shadow Copy? It basically keeps copies of files and folders at certain points in the past. I have it running every 4 hours and with 64 maximum versions, that is about 5 days as long as it doesn't fill up the drive. My only concern is that it may start backing up huge VIDEO_TS stuff (DVDs). But as long as I don't modify them and they are only created once as NEW files then they probably won't be backed up.

Every time I edit my Incorrect Rating playlist ([Rating] => [Ben's Rating]) my FLAC songs will change and they will obviously be copied over as a version. But they'll expire in 5 days.

I'm trying to figure out exactly how it works. It only scans for changes every four hours (my choice) so if I modify files, ....oh. I just figured it out. It checks the current file compared with its most recent copy. At that point it can backup the file. It knows some info based on the NTFS meta data (last modified, etc).

johnnyboy · « **Reply #2 on:** October 22, 2007, 07:39:37 am »

To give you maximum flexibility with stuff like your mail server, you could always just set it up running in a VM inside the server.
What you need with it is maximum uptime, resource wise it'd use next to nothing.

A fun experiment and an easy way to make sure something that crucial is always easily able to come up independantly would have it running in a VM image - then when your server is down you can just temporarily run it on a desktop or something.

That'd even let you use a *nix based e-mail program if you wanted or you could stick to a windows based one.

I've started using VM's alot at work for stuff and finding its a great way to seperate out functionality and configs between needs while still using the same box.

benn600 · « **Reply #3 on:** October 22, 2007, 05:46:27 pm »

That's an interesting idea--running the virtual machine on any other computer during upgrades.

As far as email servers go, I could not ask for anything more than hmailserver. It does everything I could ever want and the few things I want are being added by the developers.

I rearranged the LEDs. Unfortunately, my onboard network card doesn't appear to have an ethernet LED option but this case has two LAN LEDs. So, I hooked the first one up to the main OS drive and the second one to the Promise controller on the all-drives pin. This works well because now I have a light for the RAID array and the OS drives have an adjacent light. It shows all 16 drives independently but now I have that plus a single LED for the entire array.

All I want now is to solve the huge speed decrease problem!

horse · « **Reply #4 on:** October 25, 2007, 04:53:37 pm »

Benn,

Where are you running the speed test from? On the server or over the network?
Are you having problems with clients not getting the data fast enough?

Is the tool your using the same you mentioned before? If so, I can check mine, although most of my drives will not be as fast as yours as my 3ware controller is 1.5G SATA.

As for VMWare, yes this is very cool stuff, you need plenty of memory for multiple guest systems. But it does allow you to clone, build and move VM's around very fast. For disaster recovery it also makes for very fast rebuilds.

I don't run it at home due t the extra overhead on the box. Player or Server supports more hardware but needs a Host OS
At work I'm using ESX (no Host OS) and it runs very nicely even with 3 2003 server instances and 2 RH Linux.

LEt me know about the speed tool and I'll to a quick check for comparison

benn600 · « **Reply #5 on:** October 25, 2007, 06:42:47 pm »

I'm running it locally. It isn't even useful over the network cause it's even slower I'd think.

horse · « **Reply #6 on:** October 27, 2007, 04:45:45 pm »

Benn,

Just doing a quick check with HD Tune using pretty much any block size I get the following results: -

SATA 1.5 Attached RAID or Single disk 50 - 60 MB/s or 400 - 480 Mp/s
PATA 133 Attached single disks 70 - 80 MB/s or 560 - 640Mb/s

I then ran up two instances of HD Tune and ran them against different disks and if they where attached to different controllers there was no impact, if they where both on the same controller, it dropped 50% on each.

Without digging further it would indicate on my system the controller / bus speed is the limiting factor
Intel SATA 1.5 RAID on i7501 (built in)
Promise PATA on 66MHz PCI
3Ware SATA 1.5 RAID (not using RAID though) on PCI-X

Either way, with those speeds I can half fill the Gig Ethernet per drive and haven't seen any issue with playback on any of the client machines.

Will try the same test on another server with RAID using an LSI SAS controller and SATA II drives and let you know.

benn600 · « **Reply #7 on:** October 27, 2007, 05:33:16 pm »

It's the PRINCIPLE! Oh, and when I'm duplicating files directly on the server I like being able to duplicate an entire DVD in 10-20 seconds...yea, that's what I like. I like knowing what I paid good money for is performing as it is able to. That's interesting--it appears that PATA definitely is faster than SATA. I'm personally just trying to figure this out asap and I am discussing the same issue on another forum (2cpu) where another guy is having the same problem. I'm determined to solve it for me and the other guy.

horse · « **Reply #8 on:** October 27, 2007, 10:59:50 pm »

OK Benn, now you have my attention.

Just ran exactly the same test on a Win 2003 64bit Virtual Machine running in VMWare ESX 3.0.2
It starts of at around 50MB and after dipping down to 27, it shoots up to 580Mb for the majority of the test. Average was over 550MB. The CPU impact for the VM is much higher than on my other server (2.5% vs 34%)

The VMWare server is using LSI SAS SCSI controllers (on mobo) with SATA II drives.

This is weird . . . . . .

benn600 · « **Reply #9 on:** October 28, 2007, 02:19:10 am »

I didn't catch what kind of array you're using? 550MB? It's gotta be pretty big! At least 16 drives?

I'm also tempted to try virtual machines and such but anything out of the ordinary scares me. I'm a bit paranoid about losing data--especially right now when I'm working on several projects and I just can't lose the data for them...yes I back it up but don't want to rely on that!

newsposter · « **Reply #10 on:** October 28, 2007, 11:30:12 am »

Parallel ATA = 133 mbps x 10 bits per cycle
Serial ATA = 3 gbps x 1 bit per cycle

The math is left to the student.

horse · « **Reply #11 on:** October 28, 2007, 02:18:11 pm »

@newposter
The first server (home system) uses SATA II drives (3Gbps) but the RAID controller is only 1.5Gbps hence the PATA and SATA should have been close. Neither actually got over 1Gbps using this tool, which is strange.
The other (work) is SATA II drives and SAS which support SATA II at 3Gbps

@Benn

Sorry it was late when I typed the response, the 580M should have read 580Mbps.
The actual array is only RAID 1 and made up of 2 x 500G SATA II drives. Was not looking for performance, just didn't want to loose my VM images if I lost a drive. This is ony a feature test environment and not production / scale testing

As the HD Tune tool is indicating 580MBps I smell a rat as this is 4.6Gbps which just isn't possible at a sustained rate on the 3Gbps interface. If I had a RAID like yours with the data striped I could get the advantage of multiple 3Gbps interface, but mine is mirrored so no speed improvement.
I need to play some more, but am now intrigued! The number don't add up

benn600 · « **Reply #12 on:** October 28, 2007, 07:31:17 pm »

I think you mean 550 Mbps...that's around 62 MB/sec which, if you're mirroring, would make perfect sense. And there's no way you're getting 550MBps with only 1 drive mirrored.

horse · « **Reply #13 on:** October 30, 2007, 12:18:42 am »

Benn,

Hence why I smell a rat. The HD Tune prgram is telling me on its display that I'm getting 550MB/s.
And yes I TOTALLY agree I cannot be getting that unless VMWare has some code that alters the laws of physics.

The scale on the Y axis of HD Tune is MB/s and not Mb/s

Something is odd with how that program does throughput :-)

benn600 · « **Reply #14 on:** October 30, 2007, 12:57:29 am »

Whatever I was using to test I was getting ~50 Mega whatever's per second.

JimH · « **Reply #15 on:** October 30, 2007, 08:48:21 am »

Quote from: horse on October 30, 2007, 12:18:42 am

Hence why I smell a rat. The HD Tune prgram is telling me on its display that I'm getting 550MB/s.
And yes I TOTALLY agree I cannot be getting that unless VMWare has some code that alters the laws of physics.

A lot of, ahem, younger people seem to disregard the difference between MB/s and Mb/s.

A Mb/s is 1 megabit per second.

A MB/s is 1 megabyte per second

So a 100 Mb/s connection is about 10 MB/s.

Even an unamed and otherwise well educated person down the hall whose initials are m-a-t-t has trouble with the capital B thing.

Matt · « **Reply #16 on:** October 30, 2007, 08:53:47 am »

I'm taking lowercase mb back.

newsposter · « **Reply #17 on:** October 30, 2007, 02:30:12 pm »

42

And thanks for all the fish........

benn600 · « **Reply #18 on:** October 30, 2007, 10:13:51 pm »

No I am NOT using the 64 bit drivers. I made sure to download the latest correct bit number drivers for the two operating systems I tried. In fact, I thought the 64 bit OS was the problem in the beginning so I went back to the 32 bit version (OS & drivers)...no change.

I was reading on another forum that one guy who's having similar problems also has the Fata1ity motherboard. So now if I decided to replace the motherboard (YUCK) that could be a huge pain. First, I'm paranoid in that something would get screwed up and I'd lose my data...Second, it's a big hassle to pull everything with obvious risks. Third, I'd probably have activation issues for 2003 Server. Fourth--money. But if it would solve the problem I'd absolutely go for it.

I saw a 42-bay chassis for $4500. Now that's what I'm talkin' bout. Except I couldn't fill it up. Building a 40TB (40x1TB) server wouldn't make me magically gain tens of terabytes of data.

newsposter · « **Reply #19 on:** October 31, 2007, 03:10:09 pm »

for $4k-$5k you can buy a new to lightly used NetApp server with gobs of disk and be done with the experimentation.

horse · « **Reply #20 on:** November 01, 2007, 10:16:08 pm »

Hmm, not sure what's going on in this thread since I last checked.
Maybe up front I should state I've been doing Datacommunications and networking for 20 years (yep I'm old) and know the difference between a bit/s and a Byte/s. I can even show my age and we can discuss baud for those of us that enjoyed surfing the web with a text browser at a rocking 4800 - 9600 bp/s :-) I'm so glad I never had to install a MS SP in those days!

Anyway . . . . .
I'm actually suspecting HD Tune is the problem and not the OS based on the crazy results I got from the 2003 64bit STD edition running as a VM. It broke the laws of physics.

Anyway Benn, if I get time I'll spin up a 2003 STD 32 bit R2 server and check the results, but if they are as crazy as the 64 bit test, I will try another tool. I'll also compare it to a non virtual Vista machine (32bit at the moment) and see what it shows and if I get time, I will install 2003 native and compare on the same HW as the Vista machine (No VMware)

Things are busy at work, but I have most of the OS's running, just a matter of finding a coffee break (or late night wine break) to run a quick test.

benn600 · « **Reply #21 on:** November 01, 2007, 10:19:18 pm »

Remember that I have tried both 32 and 64 bit editions. Plus, I have simply copied and pasted files from and to the array. If I calculate the time it took I get around 50MB/sec so that would be about as fundamental as you can get without any other software potentially being the problem.

horse · « **Reply #22 on:** November 02, 2007, 07:55:57 pm »

Benn,

I did get a chance to run HD Tune on a Vista Ultimate 32bit machine and with 8MB block / most accurate and I get an average of 56MB/s. Burst was much higher, but that is the affect of the cache. This is on a X38 Intel Chipset based machine, with a Quad Core 2.4G and a single SATA II 3Gbps drive and controller.
This seemed a little low so I checked the mode it was operating in and found although the drive supports UDMA Mode 7 (Ultra ATA 512) it was operating in UDMA mode 5 (Ultra ATA 100) which is about inline with the relative speed to my home SATA I and PATA 133 speeds. Need to figure out how to pursuade the intel drivers to let the ICH9R run a little quicker, if it can. So much for 3Gbp/s :-)

I avoid doing a copy test on the same array/disk/controller as this causes strange results, especially if it is the same disk as the heads will be seeking back and forth as it reads and writes to different sectors on the same disk.

INTERACT FORUM

Author Topic: 85% Speed Cut With W2K3 Server (from Vista) (Read 5841 times)