10-21-2018 05:31 AM - edited 03-05-2019 10:59 AM
Hi to all,
A PC (Windows10) ihas 10Gb NIC connected to 10Gb interface on 2960X Switch (via SFP+).
2960X is connected to another 10Gbe switch (Quanta 10Gbe) via SFP+
A Nas is connected to Quanta switch via SFP+
Result is : PC < SFP+ > 2960X < SFP+ > QUANTA < SFP+ > NAS
All the chain is 10Gbe connected with SFP+
When i try to transfert files (NetBIOS transfert), I'm not be able to do more than 1Gb transfert rate (120Mo/s)
At the begining i start at 600Mo/s but some few second after it go down quickly to 120Mo/s...
It seems that the Cisco device is reducing the transfert rate...
How can i change this state ?
Best Regards.
10-23-2018 12:11 AM - edited 10-23-2018 12:12 AM
Hi,
I do a reload but it doesn't seem necessary
I set it, it's a little bit more quick (150Mo/s) but it's very far of what we can do.
See the begining of the file transfert speed...
10-23-2018 12:22 AM
Hello,
even 150MB is still way too low...
Can you do a ping with different packet sizes and the don't fragment bit set from your PC to the server you are downloading from until you don't see the 'Packet needs to be fragmented but DF set.' line ?
So e.g.:
ping -l 1472 -f x.x.x.x
Pinging x.x.x.x with 1472 bytes of data:
Packet needs to be fragmented but DF set.
Packet needs to be fragmented but DF set.
Packet needs to be fragmented but DF set.
Packet needs to be fragmented but DF set.
Then lower the size to e.g. 1450 and check if it is still being fragmented...
10-23-2018 11:22 AM
Ping result
c:\>ping -l 1450 -f 192.168.1.228 -t Envoi d’une requête 'Ping' 192.168.1.228 avec 1450 octets de données : Réponse de 192.168.1.228 : octets=1450 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1450 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1450 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1450 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1450 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1450 temps<1ms TTL=64 Statistiques Ping pour 192.168.1.228: Paquets : envoyés = 6, reçus = 6, perdus = 0 (perte 0%), Durée approximative des boucles en millisecondes : Minimum = 0ms, Maximum = 0ms, Moyenne = 0ms Ctrl+C ^C c:\>ping -l 1400 -f 192.168.1.228 -t Envoi d’une requête 'Ping' 192.168.1.228 avec 1400 octets de données : Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1400 temps<1ms TTL=64 Statistiques Ping pour 192.168.1.228: Paquets : envoyés = 11, reçus = 11, perdus = 0 (perte 0%), Durée approximative des boucles en millisecondes : Minimum = 0ms, Maximum = 0ms, Moyenne = 0ms Ctrl+C ^C c:\>ping -l 1300 -f 192.168.1.228 -t Envoi d’une requête 'Ping' 192.168.1.228 avec 1300 octets de données : Réponse de 192.168.1.228 : octets=1300 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1300 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1300 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1300 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1300 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1300 temps<1ms TTL=64 Statistiques Ping pour 192.168.1.228: Paquets : envoyés = 6, reçus = 6, perdus = 0 (perte 0%), Durée approximative des boucles en millisecondes : Minimum = 0ms, Maximum = 0ms, Moyenne = 0ms Ctrl+C ^C c:\>ping -l 1200 -f 192.168.1.228 -t Envoi d’une requête 'Ping' 192.168.1.228 avec 1200 octets de données : Réponse de 192.168.1.228 : octets=1200 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1200 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1200 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1200 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1200 temps<1ms TTL=64 Statistiques Ping pour 192.168.1.228: Paquets : envoyés = 5, reçus = 5, perdus = 0 (perte 0%), Durée approximative des boucles en millisecondes : Minimum = 0ms, Maximum = 0ms, Moyenne = 0ms Ctrl+C ^C c:\>ping -l 100 -f 192.168.1.228 Envoi d’une requête 'Ping' 192.168.1.228 avec 100 octets de données : Réponse de 192.168.1.228 : octets=100 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=100 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=100 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=100 temps<1ms TTL=64 Statistiques Ping pour 192.168.1.228: Paquets : envoyés = 4, reçus = 4, perdus = 0 (perte 0%), Durée approximative des boucles en millisecondes : Minimum = 0ms, Maximum = 0ms, Moyenne = 0ms c:\>
10-23-2018 11:30 AM
Hello,
You have to start with a higher value. Start with 1500 and decrease by 2 until there is no more fragmentation...
10-23-2018 11:46 AM
1472 is the value who is working
c:\>ping -l 1472 -f 192.168.1.228 Envoi d’une requête 'Ping' 192.168.1.228 avec 1472 octets de données : Réponse de 192.168.1.228 : octets=1472 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1472 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1472 temps<1ms TTL=64 Réponse de 192.168.1.228 : octets=1472 temps<1ms TTL=64
1473 is saying that it must be fragmented
c:\>ping -l 1473 -f 192.168.1.228 Envoi d’une requête 'Ping' 192.168.1.228 avec 1473 octets de données : Le paquet doit être fragmenté mais paramétré DF. Le paquet doit être fragmenté mais paramétré DF. Le paquet doit être fragmenté mais paramétré DF. Le paquet doit être fragmenté mais paramétré DF. Statistiques Ping pour 192.168.1.228: Paquets : envoyés = 4, reçus = 0, perdus = 4 (perte 100%)
10-24-2018 11:06 AM
What do you think about these result ? does it help us ?
10-23-2018 12:26 AM
don't get fooled by the diagram you show us.
this behavior is a windows-thing.
windows read data fast until a buffer gets full, and "empties" the buffer over the network at a "normal" speed.
(while filling the buffer with new data)
as a result it looks like the transfer starts at a higher speed and then drops.
can you do the same test copying data between two different disks on the same machine?
or better, is it possible to do the test with a direct connection between two machines?
10-23-2018 11:38 AM
Hi,
I use SSD, so there is no disk limitation issue for doing data transfert ...
Here are a copy from C to D drive letter.
I join a screenshot of my network NIC state.
I'm really limited by the network for unknow reason.
10-25-2018 12:01 AM
ok so you've got
- a host with ssd storage capable of transferring at Go ( in French which I assume is Gbyte in English?)
- a quanta 10G network interface
- a 2960X 10Gb
- and where does the data go to after the 2960X?
10-25-2018 03:12 AM
PC <> 2960X <> QUANTA10GB <> NAS SYNOLOGY 3614XS
10-25-2018 03:32 AM
RS3614xs/RS3614RPxs delivers ultra-high performance at 3,518 MB/s throughput
I suppose the "," is a Thousand separator and this comes down to 3,5 GBytes/sec ??
(but this does not necessarily mean this full capacity can be used for a single data-stream.)
please give more information about the 2960 and the NAS configuration
or to bypass the 2960, e.g. is it possible to connect the server and the NAS directly?
10-25-2018 03:42 AM
this thread is relatively old, but thos the version apply to your NAS?
10-25-2018 10:54 AM
Jumbo frame will only help me to take +10% of bandwidth, it doesn't explain me why i'm not able to reach around 9Gb instead of 10Gb... currently, i'm just doing 1Gb instead of 9Gb with normal MTU, that it sound bad.
10-26-2018 12:14 AM
where did you enable jumbo frame? only on the NAS?
you should configure the NAS, both switches, and maybe also your windows host to use jumbo frames.
you have a simple config where the 2960 only forwards on L2 it does not any routing !!!
look at this thread that explains that "ip routing mtu nnnn" is not the right command to use in your config.
it is for routed ports, following this, on the 2960 you need both: system mtu jumbo 9198 and system mtu routing 9000
to use jumbo frames on L2
can you try this?
Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: