How fast is GPRS?

[marlof]

<a href="http://www.ppcw.net/stories.php?story=02/09/24/7007368">http://www.ppcw.net/stories.php?story=02/09/24/7007368</a><br /><br />People who have used GSM phones to wirelessly connect to the Internet and have switched to GPRS will feel a relief when it comes to GPRS speed. People who were only used to broadband net access and use GPRS will feel disappointed when it comes to GPRS speed. Next to that some report to get 40k connections all the time, others peak at 20k. So, how fast is GPRS? What are the limitations? Read an -older, but still great- <a href="http://www.ppcw.net/stories.php?story=02/09/24/7007368">article</a> on PPCW.NET on the subject.

[CoffeeKid]

I'm an early adopter when it comes to GPRS - I had it maybe a month or so after Fido rolled it out in Canada, paying at one point up to $100 or more a month for bandwidth-limited access.

Today, I'm on one of the best GPRS plans (pricewise) on the planet - $50 a month, unlimited bandwidth, N. America wide access at no extra charge (via Fido and Voicestream / T Mobile). It may be worldwide, I'm not entirely sure.

I recently wrote an article on my personal site about GPRS speeds:

http://www.spiffle.com/rants/10-08-2002_07-30pm

To precis it here, early on, I wasn't impressed with the service. Via the card I used to have, then my Motorola P280 phone and a usb connection, I was lucky if I could see 2,000 bytes a second, and packets-loss was massive. I was very close to dropping the service at one point.

As recently as this summer, I was still seeing speeds below 3,000 bytes a second at the best of times, and massive packet losses and delays. But lately, in Vancouver and Richmond (BC, Canada) at least, I've seen some serious improvements. Packet losses and ping times can be low still, but most of the time, I see 4,500 up to 6,500 bytes/sec speeds (what's that in comparison to a 56kbit modem?), and I had the amazing phenomenon recently of being able to access the internet from rest stops on the highways, in the middle of no where.

The ability to connect to the net wirelessly while on business travel was amazing. I had my iBook connected to the net for almost two days straight while in Boston recently, and it's relatively painless - like a 56k modem, it seemed. IM'ing and email was easy, web page loading was a bit more difficult.

Still, GPRS isn't where I'd like to see it, just yet. I remember the lofty claims of ISDN speeds when it was first introduced or announced. It's no where near that speed, and at times, it crawls. Performance with my pocket pc or sony clie is similarly excrutiatingly slow - I accessed AvantGo once through my irda ports on the Motorola and clie, and it took over 30 minutes to get my 1.5mb worth of data. Since the ir should be 115kbit a second, where's the holdup?

Once GPRS does match it's initial claims of ISDN speeds, I'll be very happy with the service... maybe for a few years. Then I'll want more. :)

Mark

[Cipr0]

Why is this site so anti CDMA?

Sprint just went to a 10.00 a moth unlimited data plan on thier high end plans, but you guys dont even act like cdma is an option.


For those of us in the US that can use sprint, its the best option out there....

Real world speeds well over 60k

Cip

[marlof]

I'm far from anti-CDMA. I just don't write about it, as I have no possible way to have personal experience with it. In Europe, we don't have CDMA, and I write on GPRS for the Europe audience of this site, and for the US audience that is using GPRS. :)

[Janak Parekh]

Sprint just went to a 10.00 a moth unlimited data plan on thier high end plans, but you guys dont even act like cdma is an option.
I'm a big fan of 1xRTT CDMA in theory, and had both Verizon and Sprint phones, but until they decide either (a) to put Bluetooth in their phones or (b) to offer PPC Phone Edition / Smartphones I'm staying away. After having been exposed to Bluetooth, I'm never going back to cables between my PPC and my cell phone if I can help it.

It is worth mentioning, though, that if you have a 5-bar signal you can get more than 64kbps sustained on 1xRTT...

--bdj

[Jorj Bauer]

People who have used GSM phones to wirelessly connect to the Internet and have switched to GPRS will feel a relief when it comes to GPRS speed.

I disagree with that blanket statement. Everyone that makes this statement is overlooking a salient point: the latency of the connection. Yes, GPRS has a high throughput, but the latency on an individual packet is atrocious.

I've used CDPD, CDMA, legacy 9600 baud GSM, and GPRS wireless connections. Most of the work that I do is interactive -- terminal sessions and the likes. When I switched from CDMA (not 1x) to GPRS (as an early adopter), it became immediately apparent that GPRS was unusable for interactive sessions. I made some measurements to determine why.

These numbers are from memory, so the best I can do is ballpark. I'll have to see if I can find the original numbers. Or better yet, maybe this will inspire someone to do a more formal test to gather better data.

CDPD: 600 to 900 mS latency. Very slow, often difficult to use.

9600 baud GSM legacy data: 400 mS latency. Slow, but not unusable.

CDMA 14.4 data: 350 mS latency. Not noticeably different than 9600 baud GSM.

GPRS data: 1000 mS latency.

Yes, a whole second per packet for GPRS. It's so slow as to be nearly unusable for interactive sessions. That means that even CDPD (when it's not dropping packets) can outperform GPRS in an interactive session!

[Arne Hess]

CDPD: 600 to 900 mS latency. Very slow, often difficult to use.
9600 baud GSM legacy data: 400 mS latency. Slow, but not unusable.
CDMA 14.4 data: 350 mS latency. Not noticeably different than 9600 baud GSM.
GPRS data: 1000 mS latency.

Yes, a whole second per packet for GPRS. It's so slow as to be nearly unusable for interactive sessions. That means that even CDPD (when it's not dropping packets) can outperform GPRS in an interactive session!
I full agree with your statement that latency is even as important as the theoretic speed! That's why I wrote long time ago (Tuesday May 21 2002 @ 06:32PM MEST) also this column:

WIRELESS KNOWLEDGE: Latency or how fast is your packet sent (http://www.ppcw.net/stories.php?story=02/05/21/8730448)
While the whole wireless industry and the consumers are talking about bandwidth in wireless networks (GSM is able to provide up to 14.4 KBps, GPRS in theory 171.000 KBps and 3G in theory 2 MBps) one not less important fact is forgotten - the round trip delay or better known as latency.

It gives some hints for the non technical people here why it is important and how the latency is comparable between fixed line networks and wireless networks!

However, you have to keep in mind that CDPD is a US standard, no where else used in the world and we should focus more on today's and tomorrows (2.5/3G) barriers than on yesterday's.

[Janak Parekh]

However, you have to keep in mind that CDPD is a US standard, no where else used in the world and we should focus more on today's and tomorrows (2.5/3G) barriers than on yesterday's.
Moreso, it's being decommissioned slowly. AT&T will kill their CDPD service in a few years.

Does anyone have stats on latency on 1xRTT? I'd be curious to know, as I've also witnessed 1sec ping times with my GPRS connection. I thought it was the provider though. :(

--bdj

[Jorj Bauer]

However, you have to keep in mind that CDPD is a US standard, no where else used in the world and we should focus more on today's and tomorrows (2.5/3G) barriers than on yesterday's.

I never meant to imply that CDPD was a good technology; my intent was to emphasize the biggest flaw in GPRS. I'm hoping that successor 3g technologies will bring down the latency while improving the overall throughput... I just want it all!

[Arne Hess]

I never meant to imply that CDPD was a good technology; my intent was to emphasize the biggest flaw in GPRS. I'm hoping that successor 3g technologies will bring down the latency while improving the overall throughput... I just want it all!
You can hope for but you will not get it... 3G will not improve latency... 8O

[Jason Dunn]

We all were certainly fed a lot of marketing BS about the "high speed" nature of GPRS...truthfully, compared to the 9600 speed of GSM, it is "faster", but the best speed I've seen from Fido/Microcell is 51 Kbps, which translates to a real-world speed of 6.3 KB/s...rougly the same as a 56K modem. I haven't used a 56K modem since 1995. 8O

Still, I'm adapting to the "slow" speed - I love the freedom that wireless gives me.

[Arne Hess]

Still, I'm adapting to the "slow" speed - I love the freedom that wireless gives me.
True words! :D At least, GPRS gives me the best freedom available "everywhere" in Germany and Europe and hey - like Marlof said - it's better than 9.6 which I used to use since 1994... So for me GPRS is a real improvement, even if it is not the best which we could think about... :lol:

[disconnected]

Are there any articles that could explain to a complete non-techie (i.e. me) the difference or relationship between latency and bandwidth? I still don't get it.

Also, has anyone published real world examples and timings of network availability, connection speed (starting up the connection), and download speeds for US carriers -- T-Mobile and A T & T (GSM/CSD and GPRS) and Sprint and Verizon (old-style and 1xrtt) on PPC -- cables-to-phones versus the various aircards versus PPC phone edition, etc.? I've read various bits of anecdotal evidence, and have some of my own, but I'm picturing guys in white coats in labs actually testing all this, like they test other computer stuff.

When I got my first Pocket PC (iPAQ 3630 in June, 2000), with internet access being the most important feature for the purchase, I went through several combinations of cables and aircard before settling on a Supplynet cable to Motorola phone using Bluekite.The cable was kind of a pain, but the speed really was comparable to a 56K modem. I'm still using the same cable and phone, but PIE 2k2 is slower than the original, and there is no more Bluekite, so I feel like I've gone rapidly backwards.

It's over two years later, and I'm REALLY ready for the next generation, but nothing I've read so far is giving me a really warm fuzzy feeling about where to go next.

[Jorj Bauer]

Are there any articles that could explain to a complete non-techie (i.e. me) the difference or relationship between latency and bandwidth? I still don't get it.


I think I can summarize fairly quickly.

Bandwidth is all about volume of data, and latency is about the speed.

Imagine a fleet of motorcycles on a highway. The bigger the highway, the more bandwidth (number of motorcycles) can pass at once. But the motorcycles can only go as fast as their engines allow -- those with slower motors would have a higher latency (they would take longer to get from the start to the finish).

If you put a horde of slow motorcycles on a really wide highway, you can move a lot of people from the start to the finish quickly (on average). This is like GPRS.

If you put the same sized horde of motorcycles on a single lane road, but give the motorcycles much faster engines, then any single motorcycle can travel the distance faster... but the entire pack of motorcycles will take longer, because fewer of them fit across the width of the road at the same time. This would be more like GSM 9600.

Bandwidth is important when you're doing things like POP or IMAP mail, FTP or SCP transfer of files, and so on. Anything where a large amount of data travels together.

Latency is important when you're moving around lots of little pieces of information, like using SSH or Telnet to a remote server (where every keystroke travels from you to the server and back individually).

[disconnected]

jorj,

Thanks for the explanation. I'm still somewhat confused, but that's not an uncommon condition for me. :)

Does all this mean that an old (not 2.5 or 3g) network with a few users is the same speed as the new networks with many users? And, if so, what would have to happen to actually increase speed/latency?

[Jorj Bauer]

Does all this mean that an old (not 2.5 or 3g) network with a few users is the same speed as the new networks with many users? And, if so, what would have to happen to actually increase speed/latency?

No. My analogy is comparing motorcycles to ethernet packets, not to actual users.

If you want to get in to user load... well, that's another kettle of fish.