7 ms·

8K transactions / second is hardly extreme load. It's good they can support it, and I doubt they need more, but many, many sites handle more load than that.

by sulam 10y ago

notyourwork 10y ago
Agreed, it is good they can support it. However, is 8k TPS really a big deal? The team I work with handles 1M TPS.
- ProAm 10y ago
  it's the government, we're thrilled they can do anything.
  - kartan 10y ago
    It all depends how much politics are involved, or how much public workers can freely do their jobs: https://www.nasa.gov/ https://www.nasa.gov/ http://www.darpa.mil/ http://www.darpa.mil/ And is Healthcare.gov developed by the government or just bough to an external company? It just an observation, I don't know if Healthcare.gov is a good or bad technology.
    - foota 10y ago
      I believe it was one of the first projects of the digital service. (https://www.usds.gov/ https://www.usds.gov/)
      - vertex-four 10y ago
        Not quite. It was designed by a contractor, developed by several others, and once it was released to the public and shown to be dreadful, another contractor was put in charge of fixing it. It's still primarily developed by contract to this day. The first head of the USDS was involved in coordinating the fixing of Healthcare.gov, though.
        foota 10y ago
        That's probably what I was thinking about, thanks.
    - leakybit 10y ago
      Development is contracted out, it usually go along the lines: Budget $1 Billion 1. Large government contractors: keep $400 million, outsource all work to-> 2. Other government contractors: keep $200 million, outsource all work to-> 3. Third party companies: keep $100 million, outsource all work to--> 4. .... 5 ..... .. 10. Bangladesh code farm does all coding for $10 million.
  - sumoboy 10y ago
    Especially after spending billions $$ just to create the platform, it should have been able to do that day 1.
  - Certhas 10y ago
    I know it's just a throw away joke, but this persistent idea that government is incompetent, no matter what is really damaging. When it comes to health care systems there is no evidence whatsoever that less government means more efficiency, if anything the US is a huge example to the contrary.
    - ProAm 10y ago
      Im not really joking, their ability to do anything efficiently or correctly for a reasonable sum of money is embarrassing. I know it's a joke as well, but it's funny because it's true.
      - Retric 10y ago
        Social Security has vastly lower overhead than any equivalent private program. Now, you can see crazy waste on both sides EX: Drug Advertising. But, it's hard to look at this objectively.
      - mozumder 10y ago
        My experience is the opposite. Private companies seem completely inefficient compared to government, and I've worked as engineers in both. Really it is only a right-wing profit-driven talking point to say government is inefficient, mostly to protect against their profits that government takes away. Government is exceedingly efficient.
        ProAm 10y ago
        I've worked in both as well and vehemently disagree.
        mozumder 10y ago
        The numbers support my case. You can send a mail with the USPS for 50 cents anywhere in the country, but it costs $5-$10 to do the same with UPS or FedEx. Literally 10x-20x costlier. Government just doesn't have to waste money on things like profit or marketing that private companies need to do.
- swalsh 10y ago
  To be fair, a "transaction" here is probably a lot more complex than people give credit. Healthcare is a very sophisticated beast... it's a huge many to many problem, and though people keep trying to "standardize" details always prevail.
  - notyourwork 10y ago
    That is a really good point.
  - sulam 10y ago
    Most sites are more complicated than they seem from the outside. Complexity is fractal -- the closer you get to something, the more complex it is. One of my past employers was Salesforce.com. Seems pretty simple at first glance (and the early implementations were very simple), but then you learn more. Ignoring all of the platform-related stuff, just doing a multi-tenant database at scale turns out to be plenty hard. The fact remains, they're a couple orders of magnitude away from extreme load.
- Thaxll 10y ago
  You know 1M select(1)/sec is easy, now 1000req/sec with complex process is not easy. The number of requests per seconds doesn't matter, what matters is the complexity of the logic behind the request.
  - notyourwork 10y ago
    I like that you are assuming the complexity of 1M TPS without any context what so ever.
    - Thaxll 10y ago
      Yes 1M req/sec must but pretty simple, the only time I've seen those numbers were definitely not like "login calls", let me guess advertisement?
kartan 10y ago
I will say that if the system scales well horizontally the important metric is how many resources you need to achieve that amount of transactions. If they used 70 4-core m3.xlarge that means that they can attend 114 transactions/second per 4-core server. And that looks slow without more data about what are the requirements for the login process. But if you look at the peak of 150 requests per second, then probably it will cost more money to improve the performance than to actually just add servers. At work we have over 1 billion users. But as it scales horizontally the relevant number is how many memory/disk database instances do you have. The same applies to the application layer for the number of transactions. In fact the development servers are slower than the live ones even that they have just tens of thousands of users. For our amount of users/transactions tuning for performance saves lots of money. So it makes sense to improve the code on that measure.
honkhonkpants 10y ago
At that rate you could enroll every household in America in 3.5 hours.
user5994461 10y ago
For comparison, when I received the RFP for the Italian and the British (gov uk verify) national authentication systems. The numbers that were asked for were 12k/auth per second during peak time for the Italians. The British wasn't clear but it hinted at 0.1% of user authenticating per second, which is about the same range. (The numbers of users itself was not defined, it depends on who would join the program). SLA: 1500 ms for the account verification. 150 ms for most operations. (Some more time when requesting optional attributes like name/address/phone) If this interests you, I've published a few numbers I got from preliminary testing (to determine whether we could answer the RFP or not): https://thehftguy.com/2015/10/23/10-millions-users-accounts-with-ldap-yes-we-can/ https://thehftguy.com/2015/10/23/10-millions-users-accounts-...

8K transactions / second is hardly extreme load. It's good they can support it, and I doubt they need more, but many, many sites handle more load than that.

by sulam 10y ago

8K transactions / second is hardly extreme load. It's good they can support it, and I doubt they need more, but many, many sites handle more load than that.

notyourwork 10y ago

Agreed, it is good they can support it. However, is 8k TPS really a big deal? The team I work with handles 1M TPS.

ProAm 10y ago
it's the government, we're thrilled they can do anything.
- kartan 10y ago
  It all depends how much politics are involved, or how much public workers can freely do their jobs: https://www.nasa.gov/ https://www.nasa.gov/ http://www.darpa.mil/ http://www.darpa.mil/ And is Healthcare.gov developed by the government or just bough to an external company? It just an observation, I don't know if Healthcare.gov is a good or bad technology.
  - foota 10y ago
    I believe it was one of the first projects of the digital service. (https://www.usds.gov/ https://www.usds.gov/)
    - vertex-four 10y ago
      Not quite. It was designed by a contractor, developed by several others, and once it was released to the public and shown to be dreadful, another contractor was put in charge of fixing it. It's still primarily developed by contract to this day. The first head of the USDS was involved in coordinating the fixing of Healthcare.gov, though.
      - foota 10y ago
        That's probably what I was thinking about, thanks.
  - leakybit 10y ago
    Development is contracted out, it usually go along the lines: Budget $1 Billion 1. Large government contractors: keep $400 million, outsource all work to-> 2. Other government contractors: keep $200 million, outsource all work to-> 3. Third party companies: keep $100 million, outsource all work to--> 4. .... 5 ..... .. 10. Bangladesh code farm does all coding for $10 million.
- sumoboy 10y ago
  Especially after spending billions $$ just to create the platform, it should have been able to do that day 1.
- Certhas 10y ago
  I know it's just a throw away joke, but this persistent idea that government is incompetent, no matter what is really damaging. When it comes to health care systems there is no evidence whatsoever that less government means more efficiency, if anything the US is a huge example to the contrary.
  - ProAm 10y ago
    Im not really joking, their ability to do anything efficiently or correctly for a reasonable sum of money is embarrassing. I know it's a joke as well, but it's funny because it's true.
    - Retric 10y ago
      Social Security has vastly lower overhead than any equivalent private program. Now, you can see crazy waste on both sides EX: Drug Advertising. But, it's hard to look at this objectively.
    - mozumder 10y ago
      My experience is the opposite. Private companies seem completely inefficient compared to government, and I've worked as engineers in both. Really it is only a right-wing profit-driven talking point to say government is inefficient, mostly to protect against their profits that government takes away. Government is exceedingly efficient.
      - ProAm 10y ago
        I've worked in both as well and vehemently disagree.
        mozumder 10y ago
        The numbers support my case. You can send a mail with the USPS for 50 cents anywhere in the country, but it costs $5-$10 to do the same with UPS or FedEx. Literally 10x-20x costlier. Government just doesn't have to waste money on things like profit or marketing that private companies need to do.
swalsh 10y ago
To be fair, a "transaction" here is probably a lot more complex than people give credit. Healthcare is a very sophisticated beast... it's a huge many to many problem, and though people keep trying to "standardize" details always prevail.
- notyourwork 10y ago
  That is a really good point.
- sulam 10y ago
  Most sites are more complicated than they seem from the outside. Complexity is fractal -- the closer you get to something, the more complex it is. One of my past employers was Salesforce.com. Seems pretty simple at first glance (and the early implementations were very simple), but then you learn more. Ignoring all of the platform-related stuff, just doing a multi-tenant database at scale turns out to be plenty hard. The fact remains, they're a couple orders of magnitude away from extreme load.
Thaxll 10y ago
You know 1M select(1)/sec is easy, now 1000req/sec with complex process is not easy. The number of requests per seconds doesn't matter, what matters is the complexity of the logic behind the request.
- notyourwork 10y ago
  I like that you are assuming the complexity of 1M TPS without any context what so ever.
  - Thaxll 10y ago
    Yes 1M req/sec must but pretty simple, the only time I've seen those numbers were definitely not like "login calls", let me guess advertisement?

kartan 10y ago

I will say that if the system scales well horizontally the important metric is how many resources you need to achieve that amount of transactions. If they used 70 4-core m3.xlarge that means that they can attend 114 transactions/second per 4-core server. And that looks slow without more data about what are the requirements for the login process. But if you look at the peak of 150 requests per second, then probably it will cost more money to improve the performance than to actually just add servers. At work we have over 1 billion users. But as it scales horizontally the relevant number is how many memory/disk database instances do you have. The same applies to the application layer for the number of transactions. In fact the development servers are slower than the live ones even that they have just tens of thousands of users. For our amount of users/transactions tuning for performance saves lots of money. So it makes sense to improve the code on that measure.

honkhonkpants 10y ago

At that rate you could enroll every household in America in 3.5 hours.

user5994461 10y ago

For comparison, when I received the RFP for the Italian and the British (gov uk verify) national authentication systems. The numbers that were asked for were 12k/auth per second during peak time for the Italians. The British wasn't clear but it hinted at 0.1% of user authenticating per second, which is about the same range. (The numbers of users itself was not defined, it depends on who would join the program). SLA: 1500 ms for the account verification. 150 ms for most operations. (Some more time when requesting optional attributes like name/address/phone) If this interests you, I've published a few numbers I got from preliminary testing (to determine whether we could answer the RFP or not): https://thehftguy.com/2015/10/23/10-millions-users-accounts-with-ldap-yes-we-can/ https://thehftguy.com/2015/10/23/10-millions-users-accounts-...