Slow disks

[Peter Kunos_2]

Is it normal, when a new server (I Xeon 3.6 GHZ) with a Smart Array 642 and 10K SCSI disks (Raid5) is not faster than an ATA100 IDE 7200 RPM disk?
I made a test with the hdtach utility.
Server: Average read speed: 56.2 MB/s
My simple desktop reads 53 MB/sec.

With the Nbench utility (100 MByte, 4 threads, results in seconds):
Server:

Disk Performance, MBytes/sec
File size: 100.0 MBytes

thread: 0 1 2 3
write: 4.05 4.08 4.11 4.05
read: 14.75 14.35 14.38 14.81

My desktop:

Disk Performance, MBytes/sec
File size: 100.0 MBytes

thread: 0 1 2 3
write: 1.62 1.65 1.66 1.68
read: 1.19 1.20 1.20 1.18

Is this normal?

[Ranjith_5]

Hi Peter,

Your ATA performance will drastically come down if the number of users connected to the system is more. But SCSI can still give you good performance.

regards,
Syam

[Ian Vaughan]

Hi,
Remember that you are testing as a single user, the real gains will come when you have many users competing for resources on your server. At this point your IDE disk would be overwhelmed but your array of scsi disks will keep servicing the requests. All disks are just spinning lumps of rust - some spin faster, some have more cache, some support hot plugging etc.
I personnaly wouldn't use Raid5 for a write intensive application but that's another story.
HTH
Ian

[Ranjith_5]

Hi Peter,

I can't comment on the benchmark report that u have furnished, because differnt tools give you different values. What you should understand is that SCSI is far ahead than ATA. No doubt about it. You can only see the real performance difference in a multiuser environment.

If you want you can create a test environment.

Do the following on both server and ur desktop.

1. create a directory and give write permission on it.

2. try to write 1GB data from 5 different nodes on your network to this share.

3. During this time you run the benchmarking tool and get the report.

4.Sissoft Sandra is a good benchmarking tool.( go here to download its 8MB.
http://download.guru3d.com/sandra/)

I feel this test will show you a difference in the performance.


regards,
Syam

regards,
Syam

[Gary Artiaco_1]

I read the replies and don't you believe it.

HP does have a problem and although the systems I'm working on are different configurations, I have identified a scsi disk performance problem. I have not found the solution yet but I am working on it. We have two XW6000 workstations with 10K drives and Ultra320 SCSI integrated controller and my old W2K 1GHZ HP Vectra VL400 with an Adaptec 29160n and Maxtor 10K drive out performs them by nearly 50% ! There is a definitely something wrong. I have been in the computer business for 30 years and have been building and deploying mostly Intel based systems for the last 12 years and you would expect a 10K SCSI disk to give your system a performance kick.

We can compare notes if you would like.

gartiaco@midind.com