[sdnog] issues with high IO systems

[Daniel Shaw]

Hi,

This is not an easy one-answer issue for me. I think you may have to try a few options and look deeper to try to find out what the real cause of the delays are.

A few areas to consider:
1. Maybe improve NFS at higher layers: look at udp vs tcp and NFS versions. Also rsize and wsize at mount.
2. Maybe improve NFS at lower layers: Are there any possible NIC or switch limits in play? If you do NFS over the same interface as the web or other services, and/or the snmp, could you dedicate a different NIC to NFS only.
3. Other ways of sharing storage? Perhaps experiment with iSCSI instead of NFS.
4. Consider other alternate options for collecting logs: Syslog supports logging to a remote syslog server for example. And many services like Apache that don't log to syslog by default can be configured to do so.

Sorry I don't have a quick or easy solution, but I hope the above provide food for thought.

> -------- Original Message --------
> Subject: [sdnog] issues with high IO systems
> Local Time: 28 November 2017 2:14 PM
> UTC Time: 28 November 2017 10:14
> From: samir.s.omer at hotmail.com
> To: sdnog at sdnog.sd <sdnog at sdnog.sd>
>
> Hi all
>
> I have 5 servers which are sharing an NFS monted storage for access logs. the log files are huge about 0.6 TB of Daily logs.
>
> my issue is that the shared storage is keeping the machines busy and not replying to my cacti snmp polling which keeps sending down alarms.
>
> any advice on how to go about this issue ?
>
> Regards,
> Samir
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sdnog.sd/pipermail/sdnog/attachments/20171128/97ff79c2/attachment.html>