Recently we upgraded the hardware of a Linux system.

Old spec:
Intel S3420GPV mainboard, Intel Xeon X3440 CPU, 2 x 4GB 1333MHz ECC RAM,
3 x 1TB SATA HD

New spec:
Intel S1200SPS mainboard, Intel Xeon E3-1245 CPU, 4 x 16GB 2400MHz ECC RAM,
3 x 1TB SATA HD (from old system)

The system runs Linux kernel 4.14.11 32-bit.

After the hardware upgrade, the system boots up as usual.
No error, warning or failure was found in the "dmesg" listing.
All devices appeared to be working correctly.

During the first hour after booting, the performance remained reasonable.
The load average was about 0.4 (as reported by the "w" command), which was essentially the same as that before upgrade.

However, the longer it ran, the slower it became.
The load average varied between the minimum of 1.3 and the maximum of 5.24 in subsequent hours.

As a benchmark, the command "tar Jxf linux-4.14.11.tar.xz" was executed.
On the old hardware, it was finished in about 1 minute.
On the new hardware, it took 70 minutes to finish.

I cannot think of a reason for such performance decline.
Could someone kindly suggest a solution ?

Are you serious ?. Go read this from Linus `8 years ago.

1 members found this post helpful.

Then it should be possible to see which processes are waiting for CPU or I/O time.

What does top or htop say?

And I'd have to agree with the previous poster that running a 32-bit OS on such hardware makes little sense, but I don't see why that would lead to such severe performance degradation over time.

Thank you for the comments.

As a quick remedy, I rebooted the system with the 64-bit version of the kernel built with "IA32 emulation" enabled.
Its performance is much more "normal" than before.

This reminds me of an issue a client of the company I used to work for had.
Their SCO Unix system was going slow to dead slow.
They had not rebooted the system for such a long time that resource leakage was giving the system a slow death.
A reboot fixed the issue (when we finally managed to do so).
We modified the end of month procedure to include a system reboot.

For my own system, it was a pain to migrate from 32 bit to 64 bit Debian Linux versions, but worth it in the long run.

Some more incentive here perhaps. Have you confirmed a definite requirement for 32-bit pointers ?.

1 members found this post helpful.

Many thanks for the information.
The "64-bit kernel + 32-bit userland" combination is meant a transitional solution.
We shall definitely accelerate the migration to the native 64-bit environment.