[Beowulf] How to debug slow compute node?

John Hearns hearnsj at googlemail.com
Thu Aug 10 11:33:37 PDT 2017


Ten euros for me on a faulty DIMM


Sent from Mail for Windows 10

From: Andrew Holway
Sent: Thursday, 10 August 2017 20:05
To: Gus Correa
Cc: Beowulf Mailing List
Subject: Re: [Beowulf] How to debug slow compute node?

I put €10 on the nose for a faulty power supply.

On 10 August 2017 at 19:45, Gus Correa <gus at ldeo.columbia.edu> wrote:
+ Leftover processes from previous jobs hogging resources.
That's relatively common.
That can trigger swapping, the ultimate performance killer.
"top" or "htop" on the node should show something.
(Will go away with a reboot, of course.)

Less likely, but possible:

+ Different BIOS configuration w.r.t. the other nodes.

+ Poorly sat memory, IB card, etc, or cable connections.

+ IPMI may need a hard reset.
Power down, remove the power cable, wait several minutes,
put the cable back, power on.

Gus Correa

On 08/10/2017 11:17 AM, John Hearns via Beowulf wrote:
Another thing to perhaps look at. Are you seeing messages abotu thermal throttling events in the system logs?
Could that node have a piece of debris caught in its air intake?

I dont think that will produce a 30% drop in perfoemance. But I have caught compute nodes with pieces of packaking sucked onto the front,
following careless peeople unpacking kit in machine rooms.
(Firm rule - no packaging in the machine room. This means you)




On 10 August 2017 at 17:00, John Hearns <hearnsj at googlemail.com <mailto:hearnsj at googlemail.com>> wrote:

    ps.   Look at   watch  cat /proc/interrupts   also
    You might get a qualitative idea of a huge rate of interrupts.


    On 10 August 2017 at 16:59, John Hearns <hearnsj at googlemail.com
    <mailto:hearnsj at googlemail.com>> wrote:

        Faraz,
            I think you might have to buy me a virtual coffee. Or a beer!
        Please look at the hardware health of that machine. Specifically
        the DIMMS.  I have seen this before!
        If you have some DIMMS which are faulty and are generating ECC
        errors, then if the mcelog service is enabled
        an interrupt is generated for every ECC event. SO the system is
        spending time servicing these interrupts.

        So:   look in your /var/log/mcelog for hardware errors
        Look in your /var/log/messages for hardware errors also
        Look in the IPMI event logs for ECC errors:    ipmitool sel elist

        I would also bring that node down and boot it with memtester.
        If there is a DIMM which is that badly faulty then memtester
        will discover it within minutes.

        Or it could be something else - in which case I get no coffee.

        Also Intel cluster checker is intended to exacly deal with these
        situations.
        What is your cluster manager, and is Intel CLuster Checker
        available to you?
        I would seriously look at getting this installed.







        On 10 August 2017 at 16:39, Faraz Hussain <info at feacluster.com
        <mailto:info at feacluster.com>> wrote:

            One of our compute nodes runs ~30% slower than others. It
            has the exact same image so I am baffled why it is running
            slow . I have tested OMP and MPI benchmarks. Everything runs
            slower. The cpu usage goes to 2000%, so all looks normal there.

            I thought it may have to do with cpu scaling, i.e when the
            kernel changes the cpu speed depending on the workload. But
            we do not have that enabled on these machines.

            Here is a snippet from "cat /proc/cpuinfo". Everything is
            identical to our other nodes. Any suggestions on what else
            to check? I have tried rebooting it.

            processor       : 19
            vendor_id       : GenuineIntel
            cpu family      : 6
            model           : 62
            model name      : Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
            stepping        : 4
            cpu MHz         : 2500.098
            cache size      : 25600 KB
            physical id     : 1
            siblings        : 10
            core id         : 12
            cpu cores       : 10
            apicid          : 56
            initial apicid  : 56
            fpu             : yes
            fpu_exception   : yes
            cpuid level     : 13
            wp              : yes
            flags           : fpu vme de pse tsc msr pae mce cx8 apic
            sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr
            sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm
            constant_tsc arch_perfmon pebs bts rep_good xtopology
            nonstop_tsc aperfmperf pni pclmulqdq dtes64 monitor ds_cpl
            vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid dca sse4_1 sse4_2
            x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand
            lahf_lm ida arat xsaveopt pln pts dts tpr_shadow vnmi
            flexpriority ept vpid fsgsbase smep erms
            bogomips        : 5004.97
            clflush size    : 64
            cache_alignment : 64
            address sizes   : 46 bits physical, 48 bits virtual
            power management:



            _______________________________________________
            Beowulf mailing list, Beowulf at beowulf.org
            <mailto:Beowulf at beowulf.org> sponsored by Penguin Computing
            To change your subscription (digest mode or unsubscribe)
            visit http://www.beowulf.org/mailman/listinfo/beowulf
            <http://www.beowulf.org/mailman/listinfo/beowulf>






_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20170810/c1717584/attachment-0001.html>


More information about the Beowulf mailing list