Skip to Content.
Sympa Menu

perfsonar-user - Re: [perfsonar-user] Question regarding performance issues after Meltdown and Spectre mitigations

Subject: perfSONAR User Q&A and Other Discussion

List archive

Re: [perfsonar-user] Question regarding performance issues after Meltdown and Spectre mitigations


Chronological Thread 
  • From: Andrew Lake <>
  • To: ,
  • Subject: Re: [perfsonar-user] Question regarding performance issues after Meltdown and Spectre mitigations
  • Date: Thu, 25 Jan 2018 07:56:20 -0800
  • Ironport-phdr: 9a23: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

Hi,

The core perfSONAR development team has not observed any impacts on performance to this date. This includes both performance of the network measurements themselves as well as the software. We were concerned in particular about iperf3 and other throughput tests since they potentially use the most CPU, but at least on the systems we have access to, we could not find any indication of the problem. There have been some third-parties that have run iperf3 benchmarks as well (e.g. https://www.phoronix.com/scan.php?page=article&item=linux-more-x86pti&num=1) and have also not seen any issues.  I have seen benchmarks done by others that say postgresql takes a hit (used by pscheduler and esmond), but I have not been able to find any noticeable difference in how it runs on any of our perfSONAR boxes. We obviously don’t have access to every combination of hardware out there, so its possible there are cases that were affected, but so far we haven’t been able to find any. As we indicated in our original email about the vulnerabilities, if anyone does encounter any issues we’d be interested to know about them.

Thanks,
Andy



On January 25, 2018 at 4:36:03 AM, () wrote:

Hello,

 

Has anyone noticed an effect on network performance on perfsonar metrics after the mitigations have been applied on their node?

 

Unfortunately our node was recently reinstalled (November 2017) and previous data was for different hardware so I can’t compare properly. Having said that, I don’t see a major drop in performance comparing with a month before the mitigations were activated: Mitigations went live, if you had automatic updates enabled, on January 5 (or around that date), and required a reboot. In some cases (pre-Skylake processors) it requires a BIOS update as well. Our node, perfsonar01.jc.rl.ac.uk, has all mitigations active according to this detector script: https://github.com/speed47/spectre-meltdown-checker

 

Kind regards.

 

--

Athanasios Kanaris

Research Infrastructure Group – Scientific Computing Department

Science and Technology Facilities Council – RAL Harwell Campus

tel: 01235 567197 - ext: 7197

email/skype:

 




Archive powered by MHonArc 2.6.19.

Top of Page