Scsi Perf. issue & W2k Server

mubs

Storage? I am Storage!
Joined
Nov 22, 2002
Messages
4,908
Location
Somewhere in time.
Does W2k Server suffer from the "poor write perf." issue? If it does, is there a "system is on UPS, don't bypass write-cache" setting somewhere?

Thanks!
 

mubs

Storage? I am Storage!
Joined
Nov 22, 2002
Messages
4,908
Location
Somewhere in time.
Thanks, Pradeep.

Shit, then what are companies running W2k server doing about the performance hit? Every business worth something is on a UPS; penalizing them by doing a write-to-disk is nuts.

I know Fushigi's is an AS/400 shop, but maybe they have some W2k boxes? Anybody else have any comments about how businesses are handling this?
 

Fushigi

Storage Is My Life
Joined
Jan 23, 2002
Messages
2,890
Location
Illinois, USA
mubs said:
I know Fushigi's is an AS/400 shop, but maybe they have some W2k boxes? Anybody else have any comments about how businesses are handling this?
About 160 W2K servers in the Americas the last time I checked. Probably 300+ globally. Hopefully, 2004 will see some consolidation. I know we're planning to reduce the Exchange server count by about 2/3 through consolidation (90 to 30).

I just spot-checked a couple of our boxes and they have the write cache enabled checkbox grayed out. So we don't have the option to use it.

In a RAID situation, I think it's better to have controller write cache vs. drive cache. Under some circumstances, at least in the old days, drive write caches could get out of sync with the rest of the array, causing problems with the data. By disabling disk write caches, data is safer. Easily worth a small performance hit.

Seriously, I wouldn't worry about it too much.
 

Jan Kivar

Learning Storage Performance
Joined
Feb 3, 2003
Messages
410
I hope that it's included to SP5... IIRC there was some fix for W2k non-server, and that was "contact Microsoft" -type...

I guess running w/ SP2 is not an option... :mrgrn:

Cheers,

Jan
 

mubs

Storage? I am Storage!
Joined
Nov 22, 2002
Messages
4,908
Location
Somewhere in time.
Thanks for the reply, Fushigi! 65 raided drives on one host! 160 servers just for Exchange! I cower before your might! You're a giant compared to poor midget ole me...
 

Fushigi

Storage Is My Life
Joined
Jan 23, 2002
Messages
2,890
Location
Illinois, USA
mubs said:
Thanks for the reply, Fushigi! 65 raided drives on one host! 160 servers just for Exchange! I cower before your might! You're a giant compared to poor midget ole me...
System max for an AS/400 used to be 1000 drives, but I think they've upped that in the latest release. Something like 2000 73GB drives. Connection would be 18 drives/RAID adapter, U320 SCSI, controllers on PCI-X bus, 90 drives/frame, High Speed Loop (HSL) attach between frames (HSL=2GB fiber with 2nd connection for redundancy & path optimization). That'd more than likely be a box with 16 to 32 Power4 CPUs & 128+GB RAM. My little 4-way box with 14GB RAM and 65 drives pales by comparison.

Anyway, that was 90 Exchange boxes being consolidated into 30. 3 Active Directory domains (Americas, Europe, Asia-Pacific). Because some content is shared world-wide, the ADs are replicated across regions.
And I'm quite happy to not be a sys admin for those machines. :mrgrn:
 

mubs

Storage? I am Storage!
Joined
Nov 22, 2002
Messages
4,908
Location
Somewhere in time.
LunarMist said:
Did you try to use these?
Thanks for the link. I don't know, is it safe to use for W2k Server? I tried that a long time ago on my W2k Prof. box running Promise Ultra 100, and it craps out immediately saying
Code:
Error getting Write Cache value.
  (1117) The request could not be performed because of an I/O device error.
which is supposed to mean
Code:
This error message indicates that the disk device does not return information about its write caching status in response to the appropriate SCSI or ATAPI command. This error message is simply an indication of the capabilities (or lack thereof) of the device or its driver, and it typically implies that either the device does not support write caching or that the device driver does not support the commands that are required to query and set the device's write cache setting. 

To resolve this issue, contact the vendor of the disk device.
:(

I'm running the last driver Promise released for this card, circa 2002/09. I don't know if trying the built-in driver in Win2k will solve this particular problem or not. Is it safe to switch driver manufacturers like that?

I had forgotten all about that hotfix. Thanks for reminding me.
 

mubs

Storage? I am Storage!
Joined
Nov 22, 2002
Messages
4,908
Location
Somewhere in time.
Fushigi, that's awesome, I had no idea the AS/400 (can never remember all those new single-alphabet code names they dreamed up) could scale up that much! Maxed out, it probably approaches mainframes in configuration, performance and price, I guess?

For a few years, I managed a small AS/400 - a circa 1994 model, 24MB RAM, 1 x 2 GB disk drive, 1 x QIC. In 5 years of 24x7 power-on hours, it didn't hiccup once. Unfortunately, it ran an old legacy app. that was virtually unusable because of the botched mods that were made before I got there. For business reasons, I ended up implementing a Unix-based ERP running on a new HP Netserver I bought. (That's another stroy. That was the LS 133. Pentium Classic 133 MHz, 2 CPU's to a board, two boards max for a total of 4 CPUs! I bought it with 1 CPU, 320MB RAM, never upgraded it. No money. The CPU was passively cooled with an enormous heatsink, and a 92mm fan behind the front bezel that blew air over the heatsink from afar.)

When I was hired, I hadn't ever seen an AS/400, and had never used one. I had to figure out where the problems were, because the users kept freaking out with the errors. I had to swap out dead/dying terminals, and I had never worked with those before -- coax and all that (Ethernet I knew)! I had to figure out how to change the system clock the first time (daylight savings back to regular), how to properly shutdown the system. The manuals were all in a box that nobody knew the whereabouts of. It was a time of great trepidation and of the triumph of small victories.

Towards the end, the company moved, we pulled the plug on the AS/400, boxed it, and transported it to the new location with the intention of selling it as soon as we settled in. Then they laid me off......(they waited till the IT stuff was all working in the new location). I'm sure it's still sitting in the box on a shelf somewhere in the warehouse :mrgrn:
 

Fushigi

Storage Is My Life
Joined
Jan 23, 2002
Messages
2,890
Location
Illinois, USA
Most mid- to high-end iSeries nowadays easily outgun a mainframe (zSeries) in terms of raw MIPS. The mainframe is still a little more stable and has some cool partitioning that the 400 doesn't have just yet, but at this point a decision between the two would boil down to what your apps run on.

If it was newish in 94, that'd probably be an E or F model. Maybe an E10 or an F04. I bet the QIC drive was a 120MB or maybe a 250MB unit tops. I actually keep a QIC drive in my main machine for backwards compat with the early QIC tapes (just in case).

Reliability like you experienced is still one of the selling points. While MS is still estimating it's 99ish% uptime, IBM has measured AS/400 / iSeries uptime at 99.99+% by surveying over 80,000 system owners. Average unplanned downtime is, IIRC, under 3 hours a year. I've got 2 systems and have experienced, between both boxes, a total of 5 hours of unplanned downtime in roughly 10 operating years (1997-present on one machine, 1999-present on the other). Those 5 hours was a single event where we lost 2 disks in the same RAID set at basically the same time. We were able to pump the data from the drives and no data was actually lost.

One factor in the reliability was over-engineering. It was said, for instance, of the older systems like yours that the power supplies were good for 4 to 5 times their rating. After moving the system, you could feel how heavy it was. Lots of steel in the chassis so nothing flexes. Basically doing more than the minimum to get by. Check these out.

The smallest 400 I ever used had a whopping 8MB RAM and 2 320MB disks. Couldn't get out of it's own way. But it was just a test machine so it was adequate for the task. The first machine I ever worked on was a B50 that used something like 3 full racks to get 9.6GB of disk.

Like PCs, midrange hardware & software has come a long way. IBM uses a mere 3 iSeries machines to manage all of their manufacturing worldwide. Considering everything IBM makes, that's pretty impressive. And it'll continue for the foreseeable future as IBM has committed $500 million in R&D over the next 2 years, the pSeries and iSeries hardware will become unified (it's about 95% there already), and partitioning means a single frame can support AIX + Linux + OS/400 + Windows 2003 concurrently. Current hardware supports a max of 140 concurrent OS images; that'll go up next year. The POWER4 chips are pretty darn good, but POWER5 will be out next year, POWER6 is in development, and the feature set for POWER7 is coming along. That should hold us through about 2010.

iSeries = Integrated
pSeries = Performance
xSeries = IBM's X-Architecture
zSeries = Zero downtime

As you can plainly see, i > p, x, and z. :lol:
 
Top