Weird GPRS and UMTS latency

Bad designer wrote an interesting comment to my yesterday’s post: 2G and 3G networks have huge latency issues. As Stefan reported almost a year ago, GPRS is intolerable, UMTS is awful and HSPDA is reasonable but still not what one would hope for.

The latency does not seem to be associated with serialization delay. UMTS gives you reasonable transfer rates and significant latency and GPRS gives you an order of magnitude higher latency than ISDN with comparable transfer rate. If anyone knows enough about the mobile technologies to explain this phenomenon (or at least give me useful pointers) I’d really appreciate your comments.


  1. Reuben Farrelly14 January, 2010 08:55

    Totally agree, to the point that I'd argue that 3G wireless is just not even really a viable alternative as a backup for any sort of wired/fibre link...simply because of the latency.

  2. Unsure what the cause is but here in Oz about 6 months ago we had a major carrier make some invisible change on their 3G network and had the latency drop from 400ms down to sub-200ms. We didn't change anything - it's still not great but it is better.

  3. Many years ago I worked on getting a Vendors Packet Core devices to adhere to the 3gpp standards. I no longer do but have kept an interest in the evolution and so I can slip back into the work should I want to (sadly its always at the expense of my bread & butter MPLS Carrier Design roles so I probably won't)

    Anyway, here are a couple of links discussing the issues. Its very much to do with the radio states and channels. (The air interface)

    both start up latency and your regular latency are discussed.

    1. (Read the comments, more of an business analysis on mobile blog)

    2. (Read his articles he links too -- This is a great blog for watching the 3gpp standards evolving and he explains them too.)

    3. (Not to do with latency but in depth descriptions of the standards..)

    An interesting aspect is everyone saying LTE having an all flat architecture with less devices etc in the path will help.. I dont buy it. Even in 2000 there was no congestion (1gig edge 2.5 gig backbone) and as such from the packet core entry point just propagation latency. That firmly lays the blame downstream to Radio Area Network (RAN) and the Air interface (Radio)

    That used to be mobile--air--3g--nodeb--atm--rnc--atm--sgsn--gige--ggsn--gige--mpls_backbone (Pos).

    Now is air-hsdpa-nodeb-(Vpls)-rnc--gige--sgsn---10gige--ggsn--10gige--mpls_backbone (Pos/10GE)

    LTE will be air--OFDM/LTE--enhanced_nodeb--ethernet_backhaul--enhanced_Serving Gateway---etherent--enhanced_packet_gateway---MPLS_backbone (40gig) (Here you see the devices collapsed)

    apologies for any mistakes in the above but i did this quick :-).

    When discussing the latency in 2000/1 (Yes I noticed immediately and raised it) with my colleague Vendor Radio guys, they said "You IP guys will just never get a handle on how radio is working"... I said then it was unacceptable. Glad to see they learned.

  4. but one simple question...why my IP packets got latency starting from ~150ms to ??? and when i phone someone via UMTS (only Voide but also videophone) i dont register any latency ?

    I don´t how a phone handles the video...but i assume that the same SGSN/GGSN and so on combo is used as for data ?

  5. voice doesn't go to SGSN/GGSN. The voice equivalents are MSC (2.5/3g) MSS/MGW(3g/hsdpa inc VOIP) and there isn't yet specified for LTE any voice solution ;-) -- some proposals are One Voice AKA IMS Lite, IMS itself, CS Callback OR VOLGA (Voice of LTE Generic Access)

    afaik, Voice and data use different radio channels (Logical/Physical). Voice is continuous and data is bursty so switches channels.

  6. I have had good luck with Verizon, latency around 200 ms. Bad luck with ATT though and from what I have read in USA Today it's because of the IPHONE crushing their network. A lot has to do with signal strength and location of the antenna.

  7. I don't know these technologies specifically, but I did a course in coding theory not that long ago where radio type environments came up. Since you're more likely to get
    whole chains of errors or missing stuff in a radio type link than the single bit errors
    you get in most other contexts, one approach (also used on CDs) is to send one bit
    each of a bunch of codewords, then another bit of each, then another, and so on. so if you lose multiple bits in a row all that happens is you've slighting damaged each of several different codewords and can recover them all. However, this approach massively increases serialization of any single codeword.

    As I say, this may have nothing whatever to do with the problem you are seeing.

  8. latency is, for the most part, due to packet processing as it goes through the mobile network. lots of magic to make this stuff work. it is getting better as fewer devices are required to process traffic but until LTE expect up to 300ms.

    a quick google search turned this up

  9. Looks like this is not relevant; it describes startup latency. With GPRS and UMTS you're encountering latency on every packet.

  10. I get 92ms as we speak.

    I've worked with GGSNs and other parts of the mobile data network, and the delay is not in the network (between GGSN and NodeB). I would think it's due to the encoding/decoding of the radio signal what with the spread spectrum and signal bouncing. It would be interesting to know in which direction the signal is slowest.

    Also, it's good to keep in mind that the uplink speed is much slower than the downlink speed.

    But I never worked with the radio parts, so I don't know any details.


You don't have to log in to post a comment, but please do provide your real name/URL. Anonymous comments might get deleted.