I’ve been trying to solve the problem I have with my setup for a while without any luck. Hope you can give me help here.
There are performance issues with my server:
Asus M3A-H/HDMI + 4850e + 4GB
Internal NIC + 905c 3COM for DSL
WD GP drives: 1TB system, 2x500GB RAID, 2TB storage, 2TB storage (not yet in use, will be new system drive), 1TB USB HDD
OpenSuSe 11.2 - yes, this is old but cannot update to more recent just yet
HDD speed is not on the level it should be. Well, individual speeds are fine (hdparm -iI give good performance figures; 900+ MB/s and 90+ MB/s) but copy from HDD to HDD is not. The speed was around 13-18MB/s until I flashed a new BIOS to the mboard, changed SATA settings for ports 1-4 to ACHI and 5-6 were set to IDE (if all were ACHI then last disk was not visible and accible). After flashing and BIOS settings the speed rose quite a bit: 15-48MB/s. However, the fluctuation is severe and also it seems that occasionally the transfer stops for a brief moment and then continues. Still, new BIOS & settings resulted better performance but still far from good - half good at the best.
Also, the network speed is not at the level where it should be. iperf tests between server and WinXP machine and also between server and new machine with OpenSuSe 12.1 Live give speeds around 470…588Mbit/s. I’ve tried various different iperf settings with very little performance improvements. Also I’ve tried to increase mtu size to 8178 but result was very marginal. 8-port switch is Allied Telesyn AT-GS900/8 which should be sufficient to achieve 900+Mbit/s easily.
So, where to start so that the performance can be radically improved?
So you have had this computer for perhaps three years and been using openSUSE 11.2 for a couple of years. Did it just start running slow or when did you first decide it was running slow? I don’t think there are many suggestions for a two year old copy of openSUSE. Make sure your PC is clean and clear of all dust. Dust build up, dead cooling fans, dried up thermal grease, bad power supplies, increased heating and failing hard drives are the most common issues with this age system. I would suggest you look and see if any of these problems have come upon you.
Computer is few years older I haven’t actually paid attention to the performance before because I didn’t have that much HD space before. Also, I haven’t had time to use the PC much due house building but now that I start to have some spare time again I started to check if the performance could be improved and then realized that it’s not actually working well at all.
PC as such doesn’t suffer from overheating etc. PSU is good quality - same goes for the rest of the components. For the server PC I didn’t want to use the cheapest OEM components available at that time.
None the less, don’t discount the build up of dust inside of your computer. If you have cleaned it out in the last six months, then you are likely OK, but if its been longer than a year, its time to clean it out and good luck.
It’s been cleaned up few weeks ago the last time so it should be fine and been checked upon couple of days ago. Anyhow, I don’t think that is the right place to start looking for the source of the problem. I’m more leaning on kernel + chipset here for the origin of problems but cannot exactly pin point where to start looking for.
So, if a kernel update is what you want, I have a couple of scripts you should try. First is SAKC, which can compile any kernel source file posted and kernel.org. And SGTB which can fetch any released kernel source file back to 2005. The key word is stable release if you are looking for that.
>
> I’ve been trying to solve the problem I have with my setup for a while
> without any luck. Hope you can give me help here.
>
> There are performance issues with my server:
>
> - Asus M3A-H/HDMI + 4850e + 4GB
> - Internal NIC + 905c 3COM for DSL
> - WD GP drives: 1TB system, 2x500GB RAID, 2TB storage, 2TB storage
> (not yet in use, will be new system drive), 1TB USB HDD
> - OpenSuSe 11.2 - yes, this is old but cannot update to more recent
> just yet
Several years ago I decided to upgrade an older box with some new SATAII
hard drives. I discovered my motherboard was old enough that it failed to
understand the difference between SATA I and SATA II or III drives,
particularly the data transfer rate. And this was true even with a SATA
card that allowed the drives to be read.
I suggest you examine the specs on your hard drives and your motherboard to
be sure they are compatible.
I picked up a new motherboard and all was well. I also suggest not mixing
IDE and SATA drives.
Robert Smits wrote:
> paju2000 wrote:
>> - Asus M3A-H/HDMI + 4850e + 4GB
>> - Internal NIC + 905c 3COM for DSL
>> - WD GP drives: 1TB system, 2x500GB RAID, 2TB storage, 2TB storage
>> (not yet in use, will be new system drive), 1TB USB HDD
>> - OpenSuSe 11.2 - yes, this is old but cannot update to more recent
>> just yet
>
> Several years ago I decided to upgrade an older box with some new SATAII
> hard drives. I discovered my motherboard was old enough that it failed to
> understand the difference between SATA I and SATA II or III drives,
> particularly the data transfer rate. And this was true even with a SATA
> card that allowed the drives to be read.
>
> I suggest you examine the specs on your hard drives and your motherboard to
> be sure they are compatible.
>
> I picked up a new motherboard and all was well. I also suggest not mixing
> IDE and SATA drives.
Just to add to this. There have been a lot of changes to SATA support in
the kernel over the years, so I’d suggest it is very definitely
worthwhile upgrading to 11.4 or 12.1 before trying to address any SATA
performance issues.
jdmcdaniel3, robertsmits & djh-novell: thanks for the input!
It’s been several years since I compiled the kernel last time (something like Fedora 2 or so ;)) so I think I’ll pass the opportunity. This would actually be also yet another excuse not to upgrade the server OS to more current one. I just have to do the update in the very near future (after I get additional 4BG RAM and new 615e CPU).
One odd thing happened which pretty much seems to back up robertsmits comment. I added new 2TB drive to the machine, formatted it to ext4. This is connected to SATA5 port which is in IDE mode because ACHI mode for ports 5 and 6 leads to lost drives/ports. SATA 1-4 are in ACHI mode. Anyway, started moving files from a bit older 2TB drive to new one and now the speed is where it should be: ~95MB/s. So, most likely the mboard is just too old (780G chipset) to work properly with new SATA HDs. I don’t have any IDE devices connected in the machine. DVD writer broke down as well and new one will be USB connected.
Let’s see after OS update if the transfer performance (between HDs and network) improves after OS update. The plan is to install CentOS 6.2 as a base OS which will be long time supported (2017, IIRC). And under virtual PC I’ll install OpenSUSE 12.1 which I can shut down when needed and also updated more easily.
Update to the case. CPU has been updated to 6 core Phenom II and disto to CentOS 6.2. Network performance is now solid 645-650Mbit/s without fluctuations or drop peaks like before. HDD performance improved to 50+MB/s level but I’m not extatic about it - should be 100MB/s level. I’m pretty sure that the chipset is actually reducing the performance. There was one annoying thing with 2.0TB drives’ partitioning. When CentOS was installed the disk utility informed me that the partitioning of those 2.0TB drives is 512 bytes misaligned and causes severe reduction in performance. OpenSUSE did not tell anything about this - maybe it was too old (11.2). It took quite some googling before I managed to partition drives properly (tool that informed about the issue was not able to partition the drive correctly :-)).
In general I’m very impressed about the speed of CentOS 6.2 (64bit) compared to previous old OpenSUSE 11.2. Significant improvement in that area. However, server related admin tools are way behind OpenSUSE and also general readiness from services configuration readiness etc. is greatly behind OpenSUSE. It took quite a bit of googling before needed services were up and running properly but now it’s working pretty well. Some problems still to get graphics properly set for virtualization where I’m planning to use OpenSUSE as a workstation distro.
On 04/01/2012 03:36 PM, paju2000 wrote:
>
> Update to the case. CPU has been updated to 6 core Phenom II and disto
> to CentOS 6.2. Network performance is now solid 645-650Mbit/s without
> fluctuations or drop peaks like before. HDD performance improved to
> 50+MB/s level but I’m not extatic about it - should be 100MB/s level.
> I’m pretty sure that the chipset is actually reducing the performance.
> There was one annoying thing with 2.0TB drives’ partitioning. When
> CentOS was installed the disk utility informed me that the partitioning
> of those 2.0TB drives is 512 bytes misaligned and causes severe
> reduction in performance. OpenSUSE did not tell anything about this -
> maybe it was too old (11.2). It took quite some googling before I
> managed to partition drives properly (tool that informed about the issue
> was not able to partition the drive correctly :-)).
Yes, 11.2 is too old to know about the 4KB sector problem. That is fixed in
12.1, and I think in 11.4.
I have had the same issue running CentOS 5.5 or later on AMD based hardware.
Sometimes I would see catastrophic slow rebuild rates of a RAID1 subsystem using Hitachi 7K1000.D drives which can be pretty fast when put under testing.
What was the case in my setups (2 of them) was a bug in the kernel which drops disk performance to the floor.
Before my action plan I was getting max 30 mb/sec RAID rebuild rate. Same applied for “dd if=/dev/null of=somefile …” benchmark.
What I implemented as a fix is a setting in grub.conf:
After the setting I was getting like 105-108 Mb/sec disk transfer rates.
I wasn’t really able to get any detailed info on what excatly is the pci=nommconf setting, but it surely is a Redhad bug that should not have been present on the OpenSuSE system.
One other thing one needs to address is, that WD GP drives usually are terrifically slow. I have replaced 2 of those 2 GBs with Seagate Barracuda 7200.12 because of performance issues and after the change I could see a dramatic improvement in the IO responce times from the subsystem while running large “tar xzvf” operations from the disk to the disk (so this means read/write simultaneously).
I haven’t actually paid attention to the performance before because I didn’t have that much HD space before. Also, I haven’t had time to use the PC much due house building but now that I start to have some spare time again I started to check if the performance could be improved and then realized that it’s not actually working well at all.