Hi everyone,
Background:
My Freenas finally landed in its final location where it is fed with a 100 Mbits/sec Down and 50 Mbits/sec Up WAN fibre connection (effectively tested at ~95 and 47 Mbits/sec). The modem-router is the Internet-provider supplied 'box'.
Problem:
The problem is that the connection bandwidth varies a lot from extremes 50 Mbits/s to 1 Mbits/s and both in download and upload. Typically I have now 2.5MB/sec download (from NAS) and 300KB/sec upload (to the NAS), while yesterday, at the same time of the day, I had respectively 1.9 and 5.5 MB/sec .
Strangely, several transfers in parallel can be established to increase overall bandwidth up to saturating (almost) the theoretical bandwidth.
Setup:
NAS is made of an E-350 CPU, 8GB of RAM, 5 WD EARX 2TB in RAIDZ1. It runs Freenas 9.1.1 . Non standard (aka not 22) ports are 'properly' opened on modem-router, both for GUI and SFTP (obviously because both are working). When on LAN, NAS is able to achieve 200-300 Mbits/s on single connection/transfer. All services are deactivated except SSH, DynDNS and SMART.
Tests are made from a 'professional connection' (~500-800 Mbits/s Up and Down), and from a 'standard' ADSL connection (~1-12 Mbits/s Up and Down). Win7 with WinSCP, DokanSSH and WinSSHFS are used while OpenSSH under Ubuntu.
Tests already ran:
_Reporting does not show any CPU saturation (and LAN showed 200Mbits/sec bandwidth).
_I have been playing with the Sysctls by taking inspiration in This thread and some other internet sources (attached picture). No real influence if activated or not.
_ I enable SSH logging but did not revealed any error or collision (but maybe I am doing it wrong?)
_ Different encryption cipher (AES and blowfish) through WinSCP give different transfers rate between them, but main problem remains.
_ I tried 'standard ftp' and had almost exactly the same transfers rate (so mostly excluding CPU).
_ I tested at several times of the day (thinking that it could be the 'global internet traffic') but it is not really reproducible (and problem can be 'dropped' by establishing several // transfers)
... some other tests I may have forgotten...
I suspect more something in the RAM and/or buffer things... I have to look at this, but any help would be appreciated ;)
EDIT: I added a picture to illustrate the phenomenon.
Background:
My Freenas finally landed in its final location where it is fed with a 100 Mbits/sec Down and 50 Mbits/sec Up WAN fibre connection (effectively tested at ~95 and 47 Mbits/sec). The modem-router is the Internet-provider supplied 'box'.
Problem:
The problem is that the connection bandwidth varies a lot from extremes 50 Mbits/s to 1 Mbits/s and both in download and upload. Typically I have now 2.5MB/sec download (from NAS) and 300KB/sec upload (to the NAS), while yesterday, at the same time of the day, I had respectively 1.9 and 5.5 MB/sec .
Strangely, several transfers in parallel can be established to increase overall bandwidth up to saturating (almost) the theoretical bandwidth.
Setup:
NAS is made of an E-350 CPU, 8GB of RAM, 5 WD EARX 2TB in RAIDZ1. It runs Freenas 9.1.1 . Non standard (aka not 22) ports are 'properly' opened on modem-router, both for GUI and SFTP (obviously because both are working). When on LAN, NAS is able to achieve 200-300 Mbits/s on single connection/transfer. All services are deactivated except SSH, DynDNS and SMART.
Tests are made from a 'professional connection' (~500-800 Mbits/s Up and Down), and from a 'standard' ADSL connection (~1-12 Mbits/s Up and Down). Win7 with WinSCP, DokanSSH and WinSSHFS are used while OpenSSH under Ubuntu.
Tests already ran:
_Reporting does not show any CPU saturation (and LAN showed 200Mbits/sec bandwidth).
_I have been playing with the Sysctls by taking inspiration in This thread and some other internet sources (attached picture). No real influence if activated or not.
_ I enable SSH logging but did not revealed any error or collision (but maybe I am doing it wrong?)
_ Different encryption cipher (AES and blowfish) through WinSCP give different transfers rate between them, but main problem remains.
_ I tried 'standard ftp' and had almost exactly the same transfers rate (so mostly excluding CPU).
_ I tested at several times of the day (thinking that it could be the 'global internet traffic') but it is not really reproducible (and problem can be 'dropped' by establishing several // transfers)
... some other tests I may have forgotten...
I suspect more something in the RAM and/or buffer things... I have to look at this, but any help would be appreciated ;)
EDIT: I added a picture to illustrate the phenomenon.
