Hello All!

First of all, wish you all a very happy and prosperous new year!

I was going through some email threads! I saw below comments observations from one of the engineer. I thought this would be the right forum for discussing the same.

Any inputs, feedback, comments are most welcome. This would help me as an engineer understand which OS is best and which OS suits what kind of business and business applications.

Thanks in Advance!

1. We cannot dynamically allocate new LUNs to linux hosts. This means that every time there is a need for new space, the host must be rebooted.

2. We do not have the ability to obtain core files from paniced or hard hung systems so root cause/diagnosis of issues is next to impossible.

3. Most hardware/OS issues require a reboot. They do not have the stability of Solaris to have a reboot be the last resort.

4. We do not have hardware monitoring on linux so we currently are not aware of local disk failures or memory errors.

5. We are having trouble getting reliable consoles on linux hosts. This means that if a host falls off the network, or hard hangs, we sometimes can't even get onto the console to see what's wrong.

6. There is a known issue with RHEL4 and SUSE 9 having time sync problems.
The clock skews from a few seconds to many minutes. There is a work around but time sync problems can serious corruption to databases and applications.

7. Linux scalability above 2 CPU's: linux does not perform well at 4 cpu's and above, making any cost savings a loss in performance.

8. Oracle doens't support VCS on linux so we have to use their RAC. To my knowledge, we have had no successful failovers of the RAC cluster when one of the two nodes had OS related issues. Also, because we cannot use Veritas with RAC, we don't have dynamic multipathing, so there is only one path to the storage. This creates single points of failure for connectivity to the data.

10. Network service resiliance - if linux systems lose network connectivity, they tend to not recover from it (if at all) as well as Solaris boxes do. We've seen multiple cases where linux hosts have lost their network connection and had to be rebooted to be usable again.


Regards,
Savith

Linux apps produce core dumps when they crash, as does the kernel. Could you explain what you're missing?

Most? Such as?

That's a limitation of the motherboard/chipset choice. Some motherboard have plenty of sensors and open-source drivers, others with open specs, and others are supported by the kernel's sensor module (recompile the kernel with them enabled). if you picked a board which doesn't have these options, be aware that there are plenty of "watchdog" and monitoring PCI cards on the market which are Linux-compatible. There may be a solution for you here. It's true that it won't be quite the same as running a Sun Ultrasparc, but then again, you shouldn't expect a non-Sun Microsystems computer to be a Sun Microsystems computer. Look at the traditional market for Sun Microsystems, and compare to the traditional market for PCs, and you'll recognize why PCs are lagging in the fault tolerance and prevention arena.

1. Be sure you pick a motherboard with real UARTs, or use a real serial card. Compile the serial drivers into the kernel rather than as a module.

Would a cron job syncing the time every 15 minutes to keep the clock synced within a few milliseconds be an acceptable solution? If not, then you may need to ask motherboard manufacturers about their realtime clock accuracy, etc. PC realtime clocks are not known for keeping good time.

Which kernel revision? Linux is actually highly scalable - some of the fastest supercomputers in the world running hundreds to thousands of CPUs are running Linux. Make sure that when you compile the kernel that you configure SMP support properly - and disable hyperthreading in the BIOS (if you're using Intel chips, obviously). Keep in mind that SMP on PCs is nowhere near 100% efficient (not that Sun is, but Sun is more efficient than typical PC chipsets in this regard). PC chipsets compete on price and the mindset is to make up for the difference by adding more boxes - since you can build multiple GOOD 4-way or 8-way PC boxes for the price of a quad Ultrasparc.

I've seen this with samba but not with TCP/IP itself. I'm skeptical on this point. :-/

Thanks a lot KimVette! Thanks a bunch for your valuable thoughts. I knew, Enterprise Linux is not as bad as it was reported, but wanted to hear from an expert and get to know more. I'll get back to you as soon as I can, with answers for your questions.

Thanks a lot again.