Doomshammer's Weblog

Joerg Moellenkamp wrote a great 5-part tutorial about Solaris' Service Management Facility (SMF). Everybody who is working with SMFs but actually it not using efficiently should read Joerg's tutorial- it's absolutelly worth reading!

• Comments (4)
• Trackbacks (0)

This weekend I had to build a server for a customer. The hardware was given by the customer- I only had to set it up. Unfortunately I ran into a bunch of troubles, as Dell shipped the wrong hardware (sadly I noticed this much to late).

Given hardware was a Dell PowerEdge 1950 with Quad Core CPU and 4 GiB RAM. The server has 2 x73 GiB internal hard disks (SAS) and (currently) 4 x 500 GiB disks in a Dell PowerVault MD1000 (SAS) storage array. So far, so good... When I booted the machine the first time, I wondered why the systems disks were connected to a RAID controller (a PERC 5/i) - but I simply went on. The Solaris installer started and everything worked fine - at least until the installer reached the point where it wanted to partition the disks... "No disks found" was the message, the installer showed.

I rebooted and tried to disable the RAID-support of the controller, so that it only acts as SAS controller and passes the single drives through, but the PERC 5/i doesn't have a option for JBOD. So I looked up the chipset on the controller (LSISAS1068), downloaded the Solaris x86 driver from the LSI homepage and burned it on CD. Again the Solaris installer started and I chose "5" for "apply driver update", followed by "c" for CD. The installer recognized the driver on the CD, installed it, but when I executed "format", the disks still weren't found.

After some research in the internet I figured out, that the LSI 1068 controllers, that are shipped with Dell hardware, are confiugred with another firmware - which isn't supported by the LSI driver ... and of course Dell doesn't provide and Solaris driver.

Luckily, after some further research in tons of internet forums, I found an thread about Dell 2950 in the SUN forum. In this discussion, someone announced that he has written a driver for the LSI MegaRaid and Dell Perc 5 controllers for Solaris 10. The driver can be found here. I first tried to update the driver directly in the Solaris 10 installation DVD (or more specificly within the x86.miniroot) but this very annoying if you only have a VMWare with the correct Solaris version, so that you can build the driver. Additionally the driver would be only available during the installation - I would have needed to install it manually after the installation as well. What a luck that someone built an ITU image, which can be used for the Solaris 10 installer. I made a mirror of the package with the driver and the ITU images (as I don't know, how long the site will life).

Finally - with the ITU image - I was able to install solaris on the RAID-1 that I built on the 2 system disks. The installation worked w/o any problems... business as usual. But the next shock came quickly. The next task I wanted to do, was installing the MD1000 disks as one RAID-Z ZFS pool. But guess what... Dell again shipped the wrong hardware. The ordered SAS controller wasn't installed... therefor a PERC 5/E was installed - and of course this RAID controller isn't able to pass the disks through as JBOD neither. Well, what should I do? I set up a RAID-5 on the PERC controller, so that Solaris at least sees the disk(s).

The system is now up and running in the DC but it is really annoying that:
- Dell isn't able to ship the ordered hardware
- Dell needs it's fucking own firmware on standard LSI hardware
- Dell doesn't provide Solaris drivers for their own hardware
- The crappy PERC controllers don't provide JBOD-support

Conclusion: Solaris on a Dell 1950 with a MD1000 works, but it's really hard to get it working when a PERC controller is installed. I recommend to either not order a PERC controller but a standard SAS controller or to buy your own controllers.

Update: Yesterday the re-ordered SAS 5/E controller (original LSI SAS1068 chipset) arrived and I mounted it into the Dell 1950. The 4 x 500GB disks in the MD1000 were directly recognized by the controller and the controller was automatically initialized by Solaris 10's mpt driver. All 4 disks were found by "devfsadm" and I could quickly rebuild my ZFS pool. Now I'm happy with the box. So here again my advise- be very accurate with the hardware that you order in your Dell 1950. If you want to add a MD1000 and want ZFS to take care of the RAID, then be careful not to order a PERC 5/i or PERC 5/E controller but to take the SAS version (SAS 5/i and SAS 5/E) which passes the MD1000 through as JBOD to the Solaris operating system and which works out-of-the-box with Solaris' mpt driver.

• Comments (10)
• Trackbacks (0)

For one of my customers, I need to copy big data files from an external USB drive to a Solaris 10 server. Problem is, as the files are bigger than 4GiB, the FS on the hard disk of the USB drive needed to be NTFS formated (the customer uses Windows). So I searched for a solution to read NTFS file systems with a Solaris 10 box.

I found the FSWfsmisc package from the BeleniX project. It holds some tools to access ext2, ext3 as well as ntfs partitions. Simply download the package from this location, extract it and install it via pkgadd.

Now one can do a:

and my NTFS partition has been mounted to /mnt. Awesome! Copy speed via USB 2.0 to ZFS is really fast as well (18 to 25MB/Sek. As the partition seems to be mounted like a NFS share, one can't really access much informations from "mount" or "df". Therefor the package ships "xlsmounts" which shows detailed informations about the mounted drive.

• Comments (2)
• Trackbacks (0)

I finally had some time for playing with Solaris again - especially with ZFS. As I don't want to have another box standing in my living room and making noise, I decided to use a VMWare for playing instead. So I set up a VMWare with 11 hard drives and installed Solaris 10 11/06. Installation went smooth as usual.

First task was to set up a two-way UFS mirror between the first two disks, so that I have a failover when the first disk breaks (of cours, in a vmware this is stupid, as these are only vritual devices - but who cares... it's just for fun ). While creating the mirror, I ran into slight troubles caused by the virtual drives of VMWare. As I didn't choose to allocate the whole disk space during the setup, the disks weren't really touched when I mirrored the partition table with prtvtoc/fmthard. So the created metadb was lost after reboot and caused a OS panic. Deleting the metadb on the 2nd (not yet allocated disk) solved the panic. So I created a filesystem on the 2nd disk, so that disc space got allocated. This finally helped to resolve the metadb/metainit problem.

After that I created a Raid-Z to see how it performs within a VMWare - and though these are virtual drives, it performs like hell. All disk access is nicely striped over all 9 discs in the pool. Putting full I/O on the pool gave me this performance results:



I think these are very good results. Going to play a bit more now

• Comments (0)
• Trackback (1)

Indeed! I just am back from my trip to Hamburg. I was a really nice week. Hamburg is a very nice city. I really enjoyed my stay there. The training was very informative as well. Though I am happy being back at home now

Here a first impression of Hamburg. A HDR of the very interessting Docklands building in Oevelgoenne. It's a little bit "wiggly" as I took it free-hand, but the "mood" come over, though- doesn't it?

• Comments (6)
• Trackbacks (0)