RAID1 vs. RAID5 for performance

kjjdelist

Well Known Member
I've recently (last 6 months) changed our ENTERPRISE box from RAID1 to RAID5
to gain valuable disk space. Since that time, we've noticed a slow down in
the overall performance of our Enterprise box. I'm hesitant to say 'Eureka!
It's the RAID!' since we've made other changes and have added more load to
the box as part of our next go-live phase with CMS and Adv. Warehousing (we
are LIVE on Financials).

I have seven 9GB SCSI (10K RPM) drives configured as RAID5 allowing just
over 50GB of useable space. Should I see significant performance gains by
replacing that with four 36GB SCSI (10K RPM) drives in a RAID1
configuration?

The ENTERPRISE box is an IBM Netfinity 7000 w/Dual P-III 500 Xeons and 2GB
RAM. I have room to upgrade the memory and to add two more processors. Would
one of these options help more than the RAID configuartion?

Thanks, -Keith-
B7332, SP11.3, MSSQL 7.0, NT 4.0
 
Coming at this from a CNC perspective I can only comment on what our
techies have said when we had the same issue on our Netfinity Hardware.

What is the configuration of the RAID card itself? We found that we had
enough CPU, RAM, DISK, Disk Speed, etc... but the RAID card itself was
slowing us down. We replaced it with a faster card and things worked much
better. You might also want to look at how the disk is partitioned (your
note leads me to believe that it is not) in terms of what disks are
dedicated to OS, Swap File and Data Base. Got disk tools/monitors for
read/write activity??

Just a thought.

Regards and good luck!
Gerald.





kjjdelist
<kjjdelist@Hotmai To: [email protected]
l.com> cc:
Sent by: Subject: RAID1 vs. RAID5 for performance
owner-jdelistml@j
delist.com


07/25/2001 10:41
AM
Please respond to
jdelist





I've recently (last 6 months) changed our ENTERPRISE box from RAID1 to
RAID5
to gain valuable disk space. Since that time, we've noticed a slow down in
the overall performance of our Enterprise box. I'm hesitant to say 'Eureka!

It's the RAID!' since we've made other changes and have added more load to
the box as part of our next go-live phase with CMS and Adv. Warehousing (we

are LIVE on Financials).

I have seven 9GB SCSI (10K RPM) drives configured as RAID5 allowing just
over 50GB of useable space. Should I see significant performance gains by
replacing that with four 36GB SCSI (10K RPM) drives in a RAID1
configuration?

The ENTERPRISE box is an IBM Netfinity 7000 w/Dual P-III 500 Xeons and 2GB
RAM. I have room to upgrade the memory and to add two more processors.
Would
one of these options help more than the RAID configuartion?

Thanks, -Keith-
B7332, SP11.3, MSSQL 7.0, NT 4.0





--------------------------
 
I have a pretty similar problem.
I have IBM Netfinity 7000 w/Quad P-III 550 Xeons and 4GB
RAM prior go live. Since we go live and increase from 50 concurrent
user to 150 concurrent using mix Fat Client, Citrix, JAS, the system
seem to be slow down to half the speed. I was lucky that we have another
similar system was use for some other DB with the 770 Xeon 2mb cache. I
swapped out the CPU with the 550. The whole thing is running fast now.


Thai Nguyen
J.D. Edwards System Admin.
email: [email protected]
 
Unfortinatly, for performance reasons RAID1 is the better choice. The other thing is the partitioning of your disk. If you prefer to use RAID5 then you might want to consider putting 3 of your drives on a separate controller and use them for you system, pagefile, etc.. partitions and put the other 4 drives on a second controller for your Edwards software.

XE / SQL2K / W2KTSE
 
Not convinced of this answer. RAID5 will allow use of more spindles in
addition the binding and striping of the disk sets themselves. In a
previous, non JDE life, I found this to be the best way to enhance disk
performance. Of course you will need to consider whether to use fibre
rather than SCSI disk controllers too. Fibre obviously being more expensive
but giving much better results.

However, the administrative effort overhead of RAID5 with bound and striped
disk sets is higher because you have a more complex disk arrangement. Lots
to consider and this is my small contribution to the discussion.

Good luck.
Sid Perkins
Associate Director &
JDE Technical Project Manager
CB Hillier Parker Limited
Tel: 020 7882 8364
Fax: 020 7882 8720
Mobile: 0797 953 0506
www.cbhillierparker.com
 
Re: RE: RAID1 vs. RAID5 for performance

From a Compaq white paper:

"RAID 5 write performance is much lower than both RAID 0 or RAID 1 because for each write required for I/O operations, old data and parity information must be read and the new data and updated parity information must be written. At a 2:1 Read:Write ratio, RAID 5 configuration generates twice the number of I/0's as a RAID 0 configuration and 50% more I/0's than RAID 1. In Read only workloads, RAID 5 delivers roughly the same performance as RAID 0."

While yes I suppose I could see that RAID 5 could be faster than no raid at all, it is generally accepted that:

RAID 0 increases performance with no fault tolerance
RAID 1 gives fault tolerance (performance hit depends on controller)
RAID 0+1 combines benefits of both, but can be expensive
RAID 5 gives good "bang for the buck"

Since even high-end scsi hd's are pretty cheap these days, I pretty much say either 1 or 0+1.

Also, in these discussions we must always consider whether we will really see any performance difference before we start pouring out cash. For example, in that Compaq white paper, there is really no significant difference in overall performance until the benchmark is run with a couple hundred users. In other words, the difference for 1500 users waiting 6 seconds versus 3 seconds is everything, but for a lightly loaded system a save taking .04 seconds instead of .02 seconds will never be noticed (even though it's still twice as long). Did anyone follow that? :)


---------------------------------
OneWorld Xe SP15
Clustered Windows 2000 + SQL 2000
 
Re: RE: RAID1 vs. RAID5 for performance

Sorry for the ramble...

To answer the original question... you probably are seeing a performance issue because of the switch. Any of the changes you mention will help you, including the HD switch you mentioned. One thing I found when configuring my hardware, was that the bigger capacity drives were faster than less capacity ones. We've been brainwashed to think only RPM, but at least with compaq drives the 7200 RPM 36GB drives were about the same performance as the 18 GB 10k RPM drives (they didn't offer a 10k 36GB at the time). So, I think you'd see a big gain with the change you mentioned, both from the RAID improvement and from the faster drives.

---------------------------------
OneWorld Xe SP15
Clustered Windows 2000 + SQL 2000
 
RE: RE: RAID1 vs. RAID5 for performance

Dan :

High end SCSI controllers and its disks may be cheap north of Bravo River,
but not here
in South America! Many installations choose RAID5 because they don't want to
buy 2*N disks
instead of N+1, but they're fully aware of their pros and cons.
It's also to important to know how large is the Controller Cache and what is
the caching
strategy, 64 Mb cache seems to be fine for medium sized NT-SQL
installations.

Sebastian





B7321 to Xe, NT/W2K/SQL
JAS, Interoperability
MCDBA,MCP+I,MCSE,Citrix Admin
[email protected]
Grupo ASSA - Application Software SA
 
We are running RAID 5 with mirroring and are not having a speed problem. I would look at some hardware and software settings. Some months back, right after our Xe upgrade, we ran into some big speed issues. The problems manifested themselves most during full package builds. They all turned out to be software related settings. Here are some tips we found helpful.

1. Turn off real-time virus protection. JDE uses UNC pathing (\\server\volume\path\filename) for writing to files. This is true even when writing to local drives. Real-time virus scanning apparently thinks the information is coming in from an external source. It can slow the write processes down significantly. Folks, I'm talking hundred-fold degradation. I do not know if it impacts database writes. UNC = Universal Naming Convention

2. Load the following two LMHOSTS settings and nothing else on your Enterprise and Deployment Servers.

###.###.###.### EnterpriseServerName #PRE
###.###.###.### DeploymentServerName #PRE

###.###.###.### = IP address for respective server

These instructions preload the two entries for these related servers and reduce name resolution time. Keep additional entries to a minimum and do not include any comments. The whole file is parsed including comments on each lookup, so keeping entries and comments to a minimum will improve performance.

3. Remove all unnecessary network protocols from both servers and any workstations on which you want improved performance. Move the TCP/IP protocol to the top whenever possible.

More recently we discovered that some network adapter card settings can also cause similar problems. All these settings can be found by clicking on the Configure button of the Local Area Connection Properties dialog window and selecting the Advanced tab. Since these settings will be different for different cards and for Windows NT 4.0 and Windows 2000, I will not make any suggestions here.

You need to get the proper settings for the appropriate operating system from the adapter manufacturer. We had a few different cards but, fortunately, we were able to get information for each of them very quickly off the internet. Don't assume these settings are right because you have never changed them. I upgraded my workstation from NT to 2000 and the settings were still set for NT. The change in performance can be significant.

I hope this helps some of you.


Ray W. Justus
Kitchell Corporation
1707 East Highland Avenue, Suite 100
Phoenix, AZ 85016-4668
(602) 631-6157
 
>RAID 5 with mirroring

Huh?

The original poster stated that he noticed the degradation after changing the raid. The change in raid most definitely has contributed to the change in performance. None of us can say it is the main cause or not without knowing the other changes and seeing the system.

Back when 300MB drives were several thousand $, I ran lots of RAID 5 and it was plenty fast enough. But it IS slow-ER than the other options, period. We're talking relative here, not absolutes.

When I said HD's are cheap, I guess I was also talking relatively. If you're trying to get 100 GB useable because of some software system like JDE, the difference in 2*N vs N+1 is pretty minimal. Less than $5000 here (assuming 36GB drives), but even at triple that still pretty small potatoes compared to the price of a OneWorld implementation.

I'm not bashing RAID 5... I love it. But if you're going to be spending $500,000 to get a software package because you need that calibre of system, doesn't it warrant another $5k or even $20k in hardware to get the ultimate performance and fault tolerance?

---------------------------------
OneWorld Xe SP15
Clustered Windows 2000 + SQL 2000
 
Ditto.

----- Original Message -----
From: "boaterdan" <[email protected]>
To: <[email protected]>
Sent: Friday, July 27, 2001 12:03 PM
Subject: Re: RAID1 vs. RAID5 for performance


the raid. The change in raid most definitely has contributed to the change
in performance. None of us can say it is the main cause or not without
knowing the other changes and seeing the system.
it was plenty fast enough. But it IS slow-ER than the other options,
period. We're talking relative here, not absolutes.
you're trying to get 100 GB useable because of some software system like
JDE, the difference in 2*N vs N+1 is pretty minimal. Less than $5000 here
(assuming 36GB drives), but even at triple that still pretty small potatoes
compared to the price of a OneWorld implementation.
$500,000 to get a software package because you need that calibre of system,
doesn't it warrant another $5k or even $20k in hardware to get the ultimate
performance and fault tolerance?
http://198.144.193.139/cgi-bin/wwwthreads/showflat.pl?Cat=&Board=OW&Number=1
7643



JD Nowell
OW: B7332
ES: AS400 V4R4 CO: DB2/400 SP: 11.2
Users: 250 TSE Users: 100
 
Back
Top