E200 with Linux

[Zoltan Juracz]

Hi!

We brought ML110G4 with E200 smart array 128MB BBU controller. 2x SATA HP HDD, and HP SAS cable.

firmware 1.66 (factory)

We installed Debian 4.0(etch) standard install nothing special.(very very slow formating) After we run some test, copy 1GB to HDD, and running Bonnie++, and a result very poor..
when start wrinting disk, speed: 45MB/sec buf few seconds later jumped down to 11MB/sec. this is normal maybe in RAID 1 config. but woory the load climbing to 11-12, wa 99% and a system totally slowly down.

we tried to different configs/settings, tried all setting combinations

list:

Kernel: 2.6.18,2.6.22,2.6.23.1
OS: debian etch, Centos 5.0(red-hat 5.0), Knoppix

Contorrer settings: RAID1, Single Logical
Array cache enabled
100% write, and 50-50%

computers: HP ML110 G4, other ML110G4, and Asus barebone.

All combinations has a same affect, norhing helped.
BUT in Windows 2003 the controller running OK, copy speed appr. 12MB/sec

We send controller to service, he is change controller to a new one, but same...


We only one think this controller dont work with linux. we tired 3 kernel version!
What do you think? We have no idea!

[Daniel Kipp]

I've got the same symptoms with a ML150G3 and the E200 128mb inside. Whenever copying a bigger file the writing speed is going down to 11MB/s, system slows down averall. Tried OS (all 64bit versions): ubuntu 6, 7 and 8 LTS, opensuse 10.3, etch.

Controller settings:
RAID1, 2HDDs
RAID5, 3HDDs
firmware 1.72

I have absolutely no clue for this...

[Daniele Palumbo]

I have same problem with ML310 G4, with E200.

debian, gentoo, and also suse is going to max 10MB/sec on writing.
on reading no problem.

called support, and on SLES 10 i have had installed PSP from hp.
i have installed all, i guess that the module that make the difference is cpq_cciss-3.6.18-10.sles10.i586.rpm.

my test:
dd if=/dev/zero of=/vm/bigfile count=10 bs=500M
(dd will say the medium speed, also used dstat to see the instant speed)

i am on it right now, so stay tuned.

[Daniel Kipp]

hello daniele,

yes, please keep me informed! as i wrote i tried different installations of os', but i must admit that i didn't tried to work with the proliant support pack (i think you ment this with psp).

if you have solution for this issue please let me know!

[Dirk Fries]

Hi @all,

this week we finally got our first HP Server ( usually we youse Dell ).
The new server is a DL 180 G5 with four 250 GB SATA Harddisks ( Raid 0+1 ). As OS we use Ubuntu LTS 8.04

While reading data is pretty fast (~160MB/s), we also have the problem that writing is pretty slow (~20MB/s).

So if anyone has a suggestion or a hint I'd be pretty grateful.

Greetings, Dirk

[Daniele Palumbo]

Yesterday we changed:
- controller
- cable
- backplane
- battery
and waited for fully battery load.

in the beginning the controller was really fast: 130-170 MB/sec. (dd 100M x 50 count from zero to partition xfs), then it slowly dropped to 30-40 MB/sec, and also sometimes the usual 20MB/sec.

i'm on it with HP... i guess.

this is the script i have used for testing, as you can see there is no fragmentation cause the partition is formatted at every cicle.

can you all run it, also without the 50GB test?
just to compare tests...
---
#!/bin/bash

umount /dev/cciss/c0d0p5

for fs in "jfs -f" "xfs -f" "reiserfs -f" "ext3"
do
echo $fs
mkfs.${fs} /dev/cciss/c0d0p5
mount /dev/cciss/c0d0p5 /vm
for count in 1 2
do
echo "100M x 50"
time dd if=/dev/zero of=/vm/100Mx50 bs=100M count=50
echo "100M x 50: copia"
time dd if=/vm/100Mx50 of=/vm/100Mx50copy
echo "10M x 500"
time dd if=/dev/zero of=/vm/10Mx500 bs=10M count=500
echo "10M x 500 copia"
time dd if=/vm/10Mx500 of=/vm/10Mx500copy
echo "500M x 100"
time dd if=/dev/zero of=/vm/500Mx100 bs=500M count=100
echo "500M x 100 copia"
time dd if=/vm/500Mx100 of=/vm/500Mx100copy
done
umount /vm
done
---

[Dirk Fries]

Here are my results :

100M x 50
50+0 records in
50+0 records out
5242880000 bytes (5,2 GB) copied, 223,428 s, 23,5 MB/s

real 3m50.957s
user 0m0.000s
sys 0m14.740s
100M x 50: copia
10240000+0 records in
10240000+0 records out
5242880000 bytes (5,2 GB) copied, 375,166 s, 14,0 MB/s

real 6m27.185s
user 0m2.430s
sys 0m36.410s
10M x 500
500+0 records in
500+0 records out
5242880000 bytes (5,2 GB) copied, 229,912 s, 22,8 MB/s

real 4m0.609s
user 0m0.010s
sys 0m14.960s
10M x 500 copia
10240000+0 records in
10240000+0 records out
5242880000 bytes (5,2 GB) copied, 333,477 s, 15,7 MB/s

real 6m0.541s
user 0m2.480s
sys 0m36.130s
500M x 100
100+0 records in
100+0 records out
52428800000 bytes (52 GB) copied, 2532,38 s, 20,7 MB/s

real 42m36.491s
user 0m0.000s
sys 2m30.200s
500M x 100 copia
102400000+0 records in
102400000+0 records out
52428800000 bytes (52 GB) copied, 3399,8 s, 15,4 MB/s

real 56m53.621s
user 0m26.000s
sys 6m1.600s
100M x 50
50+0 records in
50+0 records out
5242880000 bytes (5,2 GB) copied, 250,064 s, 21,0 MB/s

real 4m21.371s
user 0m0.000s
sys 0m14.770s
100M x 50: copia
10240000+0 records in
10240000+0 records out
5242880000 bytes (5,2 GB) copied, 350,158 s, 15,0 MB/s

real 6m17.199s
user 0m2.400s
sys 0m36.190s
10M x 500
500+0 records in
500+0 records out
5242880000 bytes (5,2 GB) copied, 237,968 s, 22,0 MB/s

real 4m8.269s
user 0m0.000s
sys 0m14.980s
10M x 500 copia
10240000+0 records in
10240000+0 records out
5242880000 bytes (5,2 GB) copied, 341,818 s, 15,3 MB/s

real 6m5.870s
user 0m2.400s
sys 0m36.210s
500M x 100
100+0 records in
100+0 records out
52428800000 bytes (52 GB) copied, 2532,13 s, 20,7 MB/s

real 43m26.097s
user 0m0.000s
sys 2m33.400s
500M x 100 copia
102400000+0 records in
102400000+0 records out
52428800000 bytes (52 GB) copied, 3388,48 s, 15,5 MB/s

real 57m52.583s
user 0m25.150s
sys 6m3.540s

[Daniel Kipp]

i can only test it with the simple "dd if=/dev/zero of=/var/vm/bigtestfile count=10 bs=500M" command, because on that partition are lots of data. and due to the low perfomance frequent backups are a real annoying thing.

here's the output:

10+0 Datensà ¤tze ein
10+0 Datensà ¤tze aus
5242880000 Bytes (5,2 GB) kopiert, 421,356 s, 12,4 MB/s

i also tested bonnie++ for benchmarking with following options:
bonnie++ -d /var/vm/ -s 8000 -m mybox -u 0

output:
Version 1.01d ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
mybox 8000M 10785 5 11814 1 9119 1 129792 77 108730 10 388.4 0
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 6957 99 +++++ +++ +++++ +++ 7235 99 +++++ +++ 23433 100
mybox,8000M,10785,5,11814,1,9119,1,129792,77,108730,10,388.4,0,16,6957,99,+++++, +++,+++++,+++,7235,99,+++++,+++,23433,100

as you can see the results are almost the same.

[Dix08]

Hello there ,


we have the same problem , with same controller anybody any ideas ??

Thanx a lot !

[Daniel Kipp]

i'm wondering why there are several owners/users of the e200 smart array controller with comparable and i reproducible results in writing performance under whatever distris and no prospective solution. i think i'm going to deactivate the running system and complain the whole system. i'm curious about hp's answer and i'll let you know.

[Daniele Palumbo]

i have had to put the server in production.
anyway HP was unhelpful, if i ask to solve the problem (that is not only mine or ours, as confirmed by phone), they say that i need a software contract.

i am very tired, this is most likely the first and the last HP server.

[e200problems]

We have the same problem. We're running Ubuntu 8.04,

# dd if=/RAM/ubuntu-7.10-desktop-i386.iso of=/data1/testspeed
1425016+0 records in
1425016+0 records out
729608192 bytes (730 MB) copied, 55.278 s, 13.2 MB/s

/RAM is a RAM disk, /data1 is one of the drives connected to our E200.
At first I thought the problem was with the fact that Drive Write Cache was disabled, but I haven't noticed any improvement after enabling write using hpacucli.

We're using the cciss modules that came with the distribution by default. Is anyone using any other modules ? (i.e. my drive is accessed using /dev/cciss/c0d0p2 for example).

[Adam Hansen]

I have the exact same problem - has anyone got anywhere with HP on this? I have a ML350 with an e200 controller with 128MB of RAM running firmware v1.72 with 6 250GB SAS drives. Help....

[Bryan Buck_1]

I am also having this problem running RedHat Enterprise Linux 5, is this specific to Linux? Has anyone tried using alternate file systems? I am using this system as a file server and the performance is so bad it is unusable, has anyone verified that the P400 actually works properly?

[Bryan Buck_1]

I forgot to mention that I have tried both the built-in cciss drivers as well as the HP provided drivers with no performance improvements.

[Adam Hansen]

Has anyone replaced the array controller with something else (e.g. Adaptec 3085 controller)? I agree, my server is unusable on Ubuntu 8.04.1.

Thanks,
Adam

[Daniele Palumbo]

I wish to try a P400.
i have mailed HP, no reply yet.

otherwise, if i have to buy something, i will get a 3ware.

but which is the cable connected?
i have to check inside the server to be sure.

bye
d.

[Timothy Carlson]

I'll add that I have performance problems as well. Around 25MB/s with Bonnie using an E200 on a DL185 with 4 500G SATA drives in a RAID 5 configuration.

This is with RHEL 5.2 x86_64. I've also found the kernel thread where there is talk about updating the driver because of performance. See the ChangeLog for the 2.6.26 kernel.

http://www.kernel.org/pub/linux/kernel/v2.6/ChangeLog-2.6.26

Search for cciss and you will find a patch submitted by Mike Miller of HP.

I tried the 2.6.26.5 kernel from kernel.org and still had serious performance problems. I think the performance was actually worse than what I was getting with the RHEL 5.2 kernel.



Tim

[Adam Hansen]

Be cautious of the 3ware adapter - I read someplace (ubuntuhcl.org I think) that one of the 3ware adapters didn't even work.

I also talked to Adaptec, they said they do not support Ubuntu, but the driver source is freely availalbe. Anyone feeling up to the challenge?

[Christoph Muster]

I have the same problem.

We just bought 4 ProLiant ML110 with Smart Array E200.
I have tried Ubuntu 8.04 with same slow write rates.

I hope HP will provide a driver or firmware update soon.

Thanks

[Daniel Kipp]

I reported the problem to my dealer and asked for help. due to a missing software contract my dealer could only offer to ask a personal contact from hp's storage arrea. i've got the answer, that the stated writing perfomance should be a realistic value for the e200. just can't believe it.

had anyone tested the e200 with a w2k3 installation? i think that'll be the next step for me, at least to prove that the controller should work faster than 11MB/s...

[Christoph Muster]

I can't believe it!
I just connected a single disk to the internal SATA controller and i have write rate of 35 MB/s....

[Uwe Zessin]

SATA disks often have higher bit densities or a different geometry and _can_ beat SCSI/SAS disks in pure MegaBytes/second spiral transfer-rate.

[Daniel Kipp]

sure. but in my case 3 sata hdds are attached to the e200 to work as raid5, another 2 sata's work as raid1 on the same controller, not scsi/sas. and i guess a couple of the above described cases also refer to sata-disc installations.

[Daniele Palumbo]

mine, for example.

i have 4 sata disk on a E200.

can anyone test the drive in md, aka the linux raid software?

if it works better than hardware E200...

bye
d.