This is great news. I've had multiple customers request DC locations in South Africa for my monitoring service https://checklyhq.com. Also in general a nice boost for the underserved ZA Saas market.
Yeah, well, it's not as if they couldn't see the total and utter consolidation of cloud and hosting infra coming a mile away. The question was never "if", just "when". Should not be a shock to the local players.
Hopefully they can compete on price. It's like having Google develop a product that competes with you, in that it adds legitimacy to your market. And AWS is known as a feature leader, not a value leader.
There are plenty of organizations that prefer not to host with the PaaS providers, and instead go with traditional hosting providers. Sometimes it co-lo, sometimes it’s for better price/performance for dedicated hosts, and sometimes it’s competitive. Like the others said, this will probably hurt in the short, but help in the long run for these businesses by lending legitimacy and growing the overall space through a significant capital investment.
The 100 Mbps up and down mentioned is USD 85 per month at the current R 14.45 to the dollar in Johannesburg itself.
However, outside the urban areas internet is slower and there is not fibre yet. Cellphone coverage is generally good. For some "inexplicable" reason, however, 1GB cellphone data has been stuck at R 149 for over 5 years now...
You can get cellphone data for much, much cheaper (try Afrihost mobile).
Having recently moved to US, I'm amazed at how backward fixed internet access is here and how first-class it is in SA. Fibre is rare even in urban areas, I specifically asked my estate agent to find someplace with Comcast competition at best and Comcast fibre at worst - she couldn't manage either.
The issue with SEA is almost entirely to do with shitty telcos who run barely adequate gear, are staffed with semi-competent monkeys and who have no participation in peering fabrics. Yes, the infrastructure is in rough shape as well, just like in Africa.
SA might not be the most stable place to setup shop with the white property theft laws just around the corner. I wonder if they will also not hire white people based on the reverse affirmative action laws of the communist government.
Probably not at the top of everyones mind, but this will be great for gamers in SA. Lots of games-as-a-service (i.e. no dedicated server software you can run yourself) are hosted on AWS. Overwatch is a particular example that might now receive African servers off of this.
It doesn't necessarily mean the GaaSs (and equally PaaSs) will utilize the new region, there's a couple reasons, sometimes dependent services are rolled out slowly elsewhere - e.g. bigquery, cloud/pub, even their end user Workmail has limited regions, in addition sometimes the egress network and compute charges could be a multiple of us-east/west making the offering prohibitively expensive, and then it depends on the local connectivity itself, if major eyeball / gamer ISPs won't peer initially it might just be as well hosting in nearest Europe region to avoid the double RTL - you see this effect in markets like Singapore with Singtel, Germany with DTAC, and to a degree US with Comcast when server providers pick up local transit from HE/cogent even NTT in some parts - when AWS opened Singapore it was universally terrible routing until they got aggressive about peering and remote peering in the region.
> if major eyeball / gamer ISPs won't peer initially
Almost all the major retail ISPs in South Africa openly peer at NAPAfrica (https://www.napafrica.net/), mainly to pick up Akamai, Google, and more recently CloudFront, but also to bypass expensive local transit costs from the local 'carriers' (Telkom/Openserve, MTN, Vodacom, Internet Solutions etc.).
As far as I know, only the mobile networks (some of whom are building out FTTH networks, but with relatively limited market share) don't openly peer there:
- Vodacom
- MTN
Sorry, it’s unlikely that Overwatch latency would improve specifically from this as I am very doubtful that it is hosted on AWS, or any cloud provider for that matter.
I worked customer support for Blizzard Europe for several years until about 2012. At the time they ran their own infrastructure co-located in ISP data centres. It was very important to Blizzard to have as much control of the quality of the user gaming experience as possible, with control of their infrastructure being a large part of this, and I would be very surprised to hear this philosophy of theirs has changed.
Then again, I don’t know for sure what they’re doing these days.
Not blizzard specifically, more Activision-blizzard, but the latest call of duty open beta was reportedly running on vultr for at least some of the game servers.
I think they moved it back to in-house servers when it went live, though, and dropped the server rates from 60hz to 20hz. Kind of a shame, the difference was very noticeable.
Anecdotal, but afaik Blizzard uses AWS for Overwatch in the US. A friend who works as a systems engineer told me they hit some ridiculous cap on an AWS region a few years back and it caused issues for a bit.
Most of EC2 and S3 product development happened in South Africa. The original technical lead for AWS was South African, and was about to move back from Seattle after the original AWS pitch meeting. Bezos liked the idea so much that he let him develop the product in South Africa.
> Since then, most of AWS transitioned back to Seattle and the CT team focuses on some of the tooling around AWS.
This is not true.
Most of AWS was never in Cape Town, only most of EC2.
And lots of EC2 development still occurs in Cape Town. EC2 has just grown immensely, so even though EC2 developers in Cape Town are now at least 25 times more than at launch, there are teams working on EC2-specific stuff in a number of locations (but yes, many in Seattle).
However, all customer EC2 API calls (e.g. the AWS cli's 'aws ec2 run-instances' etc.) are handled by software that is currently developed and maintained by the Cape Town office, and I believe the EC2 parts of the AWS console are also mostly developed in Cape Town.
There are however teams besides EC2 that have developers in Cape Town (e.g. I think the Personal Health Dashboard, which is not EC2-specific, is also developed in Cape Town).
Cape Town also has a large team of support engineers in Premium Support, and handles all Premium Support calls for about 6 hours a day (IIRC).
This video talks a little bit about the EC2 team in Cape Town (a team that I’ve always enjoyed working with). It was made at the 10 yr anniversary of EC2.
In terms of maximising the metric of "getting the servers closest to the maximum number of end-users with the longest current ping-times" then a region in Johannesburg seems to make more sense - it's a bigger metropolitan area than Cape Town, and is closer to points north of South Africa.
However
1) Johannesburg is inland, the internet backbone cables land at the coast e.g. at Cape Town. ( https://www.submarinecablemap.com/ - Cape Town is at the South-West-most point where all those cables come in )
3) Johannesburg is one of the largest murder/crime zones in the country, whereas Cape Town is one of the last areas remaining in the country that has a low murder/crime rate.
My friend, there is a murder epidemic right nearby Cape Town and there is plenty of crime there, I would say more even than Gauteng (Pretoria and Johannesburg)
I realize that Africa is an enormous continent; I'm just curious if this doesn't also help out developers and startups in places like Nigeria...? I don't have a sense of the topology of the Internet throughout Africa vs particular African nations and the rest of the world.
The backbone cables loop around the coast of Africa. Cape Town is on that coast. So is Lagos, the most populous city in Nigeria, the most populous country. But the most populous area and economic hub of South Africa is around Johannesburg, which is far from the coast.
We assume that AWS took the decision that the South African economy today was more important than the Nigerian economy; this data centre won't be much closer to Nigeria than European data centres are. Or maybe the double distance, double lag time at South Africa is a bigger reduction to eliminate.
This press release https://www.allthingsdistributed.com/2018/10/an-aws-region-i... says "The new AWS Africa (Cape Town) Region will ... provide lower latency to end users across Sub-Saharan Africa" We can assume that this statement is truer the further south of the Sahara you go ;)
The backbone cables loop around the coast of Africa.
I believe you that this is the case, but the map you linked is a submarine cable map. If there were backbone ables inside of Africa, they wouldn't show up.
> Of the total activated/ sold bandwidth of 3.319 Tbps in Sub-Saharan Africa by December 2016, 3.065 Tbps (92.3%) was supplied directly by submarine cable.
> Of the total bandwidth of 3.319 Tbps in Sub-Saharan Africa by December 2016, 248 Gbps (7.5%) was supplied by terrestrial cross-border networks connected to submarine cables. The completion of new cross-border links, and the expansion of capacity on others, has seen the volume of intra-regional traffic backhauled to submarine cable landing points increase by 65% in the last year
> I believe at present most traffic between Nigeria and South Africa transits London
Simply, the network is at present not optimised for traffic between e.g. Johannesburg and Lagos. Why would it be? There's not a lot of it. An internet user in Johannesburg is mostly either going to look at domestic websites in SA, or the "top 1000" sites, hosted in EU and USA.
I hope that this will change over time, and better data centres in Africa is part of that.
If internet traffic is anything like road traffic then the presence of a highway would generate more traffic. Having a crappy connection pretty much guarantees that there is little traffic!
There are no overland backbones to my knowledge. At least not in the continuous sense - I suppose you could patch together a web of adjacent country networks. Country internal there is a pretty good web of them, mostly to connect cellphone towers.
In reality everything goes via the submarine cables though. It's easier than fighting through the landrights of 6,000 miles of sometimes shaky countries.
> If there were backbone ables inside of Africa, they wouldn't show up.
There are, and the information in sibling comments is either wrong or outdated. Liquid has a fiber network that literally stretches from "Cape to Cairo"[1]. Liquid has a presence in Nigeria, but it's not connected to the overland fiber network
> I realize that Africa is an enormous continent; I'm just curious if this doesn't also help out developers and startups in places like Nigeria...? I don't have a sense of the topology of the Internet throughout Africa vs particular African nations and the rest of the world.
It probably doesn't help developers in Nigeria get lower latency or cheaper access to AWS, but it may help them get more customers more quickly, as some other obvious markets that might be a better fit for services they develop would benefit from the region in South Africa.
However, it is quite expensive to operate data centres in Nigeria (as the electrical supply is not very reliable; I know that at least quite recently many DCs there need to run on generator about 10% of the time, I don't know if that has changed since).
There are, however, some initiatives to get more international intra-continental peering going (e.g. I believe at present most traffic between Nigeria and South Africa transits London).
A few years ago, I was somewhat involved with some data centres in Nigeria, which ran on generators 2-3 times a day for at last an hour each time, every day, for more than a year, due to the electrical grid in the city (Lagos) going down. The locals indicated that this was normal (e.g. had been like this for years).
Ghana has much more reliable electricity than Nigeria.
Does AWS publish performance statistics for the different regions, e.g. what is the average ping time to servers in this region when pinged from San Francisco & Tokyo?
Any guesses on when we'll see the first AWS region in space? Would it make sense to have shared computing resources 'above the clouds' to facilitate science and engineering missions? Is there ever a point at which it is better to compute in space rather than blast the data back to earth(which incurs significant latency and potentially bandwidth)?
You could do it right now, just shoot some snowball edges into space. You'd have all of the usual space-y problems, but I think edges are being used on tankers and things right now, so the not-always-connected chunk of cloud problem is sovled(ish).
It's all economics. As soon as the breakeven point of a space-based datacenter goes below a certain threshold, you'd better believe AWS will start building them.
This is great news for companies operating in South Africa that often face unreasonable latencies or costs of maintaining data centers here. Protectionist laws wrt electronics and lack of competition have led to high prices for smaller companies. Excited to see what this does for the startup scene here.
70 comments
[ 4.3 ms ] story [ 141 ms ] threadI wish this was a DO announcement. I’m a one man band and am loathe to invest the time to learn the whole AWS stack.
Maybe lightsail will be included...
The latency down here is a killer, often 300ms+ to EU/US, and for chatty web apps, a real pain.
What about general telecom infrastructure in SA? Sure, you may have a local AWS DC, but, if everyone's on copper or mobile..
Of course, compared to touching anything across the atlantic, anything local is like lightning.
Compared with the rubbish ADSL we've had to deal with in the past it's day and night.
However, outside the urban areas internet is slower and there is not fibre yet. Cellphone coverage is generally good. For some "inexplicable" reason, however, 1GB cellphone data has been stuck at R 149 for over 5 years now...
Having recently moved to US, I'm amazed at how backward fixed internet access is here and how first-class it is in SA. Fibre is rare even in urban areas, I specifically asked my estate agent to find someplace with Comcast competition at best and Comcast fibre at worst - she couldn't manage either.
Isn't Azure supposed to be up this year?
That's one game.
The largest game, by a large margin.
Good news nonetheless, early days though...
Almost all the major retail ISPs in South Africa openly peer at NAPAfrica (https://www.napafrica.net/), mainly to pick up Akamai, Google, and more recently CloudFront, but also to bypass expensive local transit costs from the local 'carriers' (Telkom/Openserve, MTN, Vodacom, Internet Solutions etc.).
As far as I know, only the mobile networks (some of whom are building out FTTH networks, but with relatively limited market share) don't openly peer there: - Vodacom - MTN
https://www.napafrica.net/traffic/ NAPAfrica currently does ~500Gbps of traffic. Approx 4 years ago it was ~10Gbps.
I worked customer support for Blizzard Europe for several years until about 2012. At the time they ran their own infrastructure co-located in ISP data centres. It was very important to Blizzard to have as much control of the quality of the user gaming experience as possible, with control of their infrastructure being a large part of this, and I would be very surprised to hear this philosophy of theirs has changed.
Then again, I don’t know for sure what they’re doing these days.
I think they moved it back to in-house servers when it went live, though, and dropped the server rates from 60hz to 20hz. Kind of a shame, the difference was very noticeable.
The comment by Bill Warnecke seems to indicate that some of the Overwatch fleet is hosted on AWS.
I interviewed with Amazon in CT and I regret not accepting the job offer. Seems like that office will really be picking up.
This is not true.
Most of AWS was never in Cape Town, only most of EC2.
And lots of EC2 development still occurs in Cape Town. EC2 has just grown immensely, so even though EC2 developers in Cape Town are now at least 25 times more than at launch, there are teams working on EC2-specific stuff in a number of locations (but yes, many in Seattle).
However, all customer EC2 API calls (e.g. the AWS cli's 'aws ec2 run-instances' etc.) are handled by software that is currently developed and maintained by the Cape Town office, and I believe the EC2 parts of the AWS console are also mostly developed in Cape Town.
There are however teams besides EC2 that have developers in Cape Town (e.g. I think the Personal Health Dashboard, which is not EC2-specific, is also developed in Cape Town).
Cape Town also has a large team of support engineers in Premium Support, and handles all Premium Support calls for about 6 hours a day (IIRC).
Legacy of the empire--there's almost always an English-speaking population in normal business hours.
https://youtu.be/9Gk_I_0eMDA
However
1) Johannesburg is inland, the internet backbone cables land at the coast e.g. at Cape Town. ( https://www.submarinecablemap.com/ - Cape Town is at the South-West-most point where all those cables come in )
2) AWS already has an operation in Cape Town.
https://en.wikipedia.org/wiki/List_of_cities_by_murder_rate
https://www.enca.com/south-africa/crime-stats-reveal-western...
https://businesstech.co.za/news/government/270771/this-is-so...
I linked to it in this comment https://news.ycombinator.com/item?id=18300309
The backbone cables loop around the coast of Africa. Cape Town is on that coast. So is Lagos, the most populous city in Nigeria, the most populous country. But the most populous area and economic hub of South Africa is around Johannesburg, which is far from the coast.
We assume that AWS took the decision that the South African economy today was more important than the Nigerian economy; this data centre won't be much closer to Nigeria than European data centres are. Or maybe the double distance, double lag time at South Africa is a bigger reduction to eliminate.
This press release https://www.allthingsdistributed.com/2018/10/an-aws-region-i... says "The new AWS Africa (Cape Town) Region will ... provide lower latency to end users across Sub-Saharan Africa" We can assume that this statement is truer the further south of the Sahara you go ;)
I believe you that this is the case, but the map you linked is a submarine cable map. If there were backbone ables inside of Africa, they wouldn't show up.
Good point. I haven't found a lot of resource on this topic. The exception, which discusses the growth of terrestrial networks is this:
http://www.africabandwidthmaps.com/?p=5505 e.g.
> Of the total activated/ sold bandwidth of 3.319 Tbps in Sub-Saharan Africa by December 2016, 3.065 Tbps (92.3%) was supplied directly by submarine cable.
> Of the total bandwidth of 3.319 Tbps in Sub-Saharan Africa by December 2016, 248 Gbps (7.5%) was supplied by terrestrial cross-border networks connected to submarine cables. The completion of new cross-border links, and the expansion of capacity on others, has seen the volume of intra-regional traffic backhauled to submarine cable landing points increase by 65% in the last year
I note also this comment:
https://news.ycombinator.com/item?id=18301912
> I believe at present most traffic between Nigeria and South Africa transits London
Simply, the network is at present not optimised for traffic between e.g. Johannesburg and Lagos. Why would it be? There's not a lot of it. An internet user in Johannesburg is mostly either going to look at domestic websites in SA, or the "top 1000" sites, hosted in EU and USA.
I hope that this will change over time, and better data centres in Africa is part of that.
If internet traffic is anything like road traffic then the presence of a highway would generate more traffic. Having a crappy connection pretty much guarantees that there is little traffic!
In reality everything goes via the submarine cables though. It's easier than fighting through the landrights of 6,000 miles of sometimes shaky countries.
There are, and the information in sibling comments is either wrong or outdated. Liquid has a fiber network that literally stretches from "Cape to Cairo"[1]. Liquid has a presence in Nigeria, but it's not connected to the overland fiber network
[1] https://www.liquidtelecom.com/about-us/network-map.html
> covering Africa's fastest-growing economies, where no fixed network has ever existed before.
It probably doesn't help developers in Nigeria get lower latency or cheaper access to AWS, but it may help them get more customers more quickly, as some other obvious markets that might be a better fit for services they develop would benefit from the region in South Africa.
However, it is quite expensive to operate data centres in Nigeria (as the electrical supply is not very reliable; I know that at least quite recently many DCs there need to run on generator about 10% of the time, I don't know if that has changed since).
There are, however, some initiatives to get more international intra-continental peering going (e.g. I believe at present most traffic between Nigeria and South Africa transits London).
Electrical supply in Cape Town has had issues too: https://www.google.com/search?q=cape+town+load+shedding
I expect that data centers will use a lot of power in summer when the temperature outside gets around 40 C (over 100 F)
> Electrical supply in Cape Town has had issues too: https://www.google.com/search?q=cape+town+load+shedding
Oh wow, one day of residential load shedding in the past year.
Nigeria had a country-wide electrical outage this year: https://www.premiumtimesng.com/news/headlines/254308-updated...
A few years ago, I was somewhat involved with some data centres in Nigeria, which ran on generators 2-3 times a day for at last an hour each time, every day, for more than a year, due to the electrical grid in the city (Lagos) going down. The locals indicated that this was normal (e.g. had been like this for years).
Ghana has much more reliable electricity than Nigeria.
A few more articles:
http://dailypost.ng/2018/07/24/nigeria-may-not-stable-power-...
https://africacheck.org/factsheets/factsheet-challenge-keepi...
"To cope with the frequent blackouts and lack of access, a quarter of Nigerians relied on generators in 2016"
South Africa's electricity generation (Eskom) has problems, but nothing like Nigeria's (yet).
Does AWS publish performance statistics for the different regions, e.g. what is the average ping time to servers in this region when pinged from San Francisco & Tokyo?
https://aws.amazon.com/snowball-edge/