Hello Xinke,
I believe I've found the culprit. It took me longer than I thought it would, but I have a full Litecell 1.5 network up and running now. I continued to see poor throughput though, even with the Litecell instead of a B210! Over the last few days I worked through the different components, and discovered that it seemed like the bottleneck was in the PCU. That lead me to take a look at the osmo-pcu issue tracker, where I found the following ticket: https://osmocom.org/issues/1524 . It looks like the PCU has trouble communicating with the basebands of some phones, particularly phones with Mediatek chipsets. I was able to get ahold of an iPhone8 for testing (Qualcomm baseband), and was able to get a more reasonable 400kbps of throughput (with healthy PCU packet counters, vs my lab test phone that had large numbers of PCU TBF timeouts). It just so happened that my main lab test phones were all low end models with Mediatek chipsets, even though they were in very different form factors and from different manufacturers. Hopefully this is helpful-- let me know if you've found another possible explanation for your observed low throughput.
Regards,
-Matt J.

On Thu, Apr 12, 2018 at 3:43 PM XINKE ZHANG <xinke.zhang@hotmail.com> wrote:

Hi Matthew,


I'm sorry for late reply. 

I have also tested the latest split components running on a xubuntu 14.04 machine, but the throughput is very limited. I buy a test phone, and find that the dl bler is so high. In my opinion, the high dl bler causes the low throughput. Nowadays, I'm trying to locate issues, but there is no progress.


Best Regards,

Xinke


From: Matthew Johnson <matt9j@cs.washington.edu>
Sent: Wednesday, April 4, 2018 16:54
To: Neels Hofmeyr
Cc: XINKE ZHANG; osmocom-net-gprs@lists.osmocom.org
Subject: Re: Data Rate of EGPRS is Too Low
 
Hello Xinke,
I'm a graduate student at the University of Washington, and just brought up an EGPRS/GPRS network with an Ettus B210/osmo-trx. My B210 has a GPSDO as well. I'm just getting started on this project (so I'm unfortunately not the expert you are looking for), but I can corroborate that I am also seeing very limited single user throughput in EGRPS and vanilla GRPS (6-8kbps maximum measured via iperf). My network is built with the latest split components running on a debian9 virtual machine via virtualbox. I'll attach my configs here if they're helpful. I also have access to a litecell1.5 and will be bringing up the network on it in the next few weeks, so I'll be able to report if it behaves any better. The litecell setup will also be running on dedicated hardware to rule out any strangeness from virtualbox. Sorry I don't have any other clues at this time, but I can at least confirm Neels's analysis above and rule out the impact of running the old nitb and ggsn.
Cheers,
-Matt Johnson

On Mon, Mar 5, 2018 at 2:15 PM, Neels Hofmeyr <nhofmeyr@sysmocom.de> wrote:
On Sat, Mar 03, 2018 at 12:18:52PM +0000, XINKE ZHANG wrote:
> I'm sure that B210 has a GPSDO. I execute 'uhd_usrp_probe' to prove it.

I'm not familiar with it, but if you say so.

> My entire config files are attached here. Pls help check it. Thanks.

It all looks good to me. Two minor notes:

- the pcu config once needed 'two-phase-access' to function reliably, but this
  is no longer needed.

- you seem to be running an old version of the ggsn. The new one is called
  osmo-ggsn with a config file style and vty interface like the other osmo
  programs. However, this should not affect your downlink throughput.

- you also seem to run osmo-nitb, we have moved to separate programs, being
  osmo-hlr, osmo-msc and osmo-bsc. Refer to
  http://osmocom.org/projects/cellular-infrastructure/wiki/Osmocom_Network_In_The_Box
  This should, however, not affect your downlink throughput, either.

So yeah, sorry, but I have nothing of weight to add. Maybe someone else here
has deeper knowledge of EGPRS and SDR based BTS?

~N