It is very difficult to run a basic database/web server with decent response times within the 64MB allowance on your basic $20/month plan. Based on your published contention ratios, a box hosting basic accounts would have 2.5GB of RAM (40 X 64MB).

Since box memory capacity and memory prices are constantly improving, when do you foresee offering a more reasonable amount of memory on the basic package? It would be very possible (and I believe you would still quickly recover your costs) to put 6GB into your boxes and start your packages at 128MB instead of 64MB. That would make the basic package so much more practical of a proposition. As it is now, it really borders on unusable. In my opinion you should not offer a basic package if it can't meet users' basic expectations.

Just so that you can understand what my expectations are, I'm not talking about a high-traffic web site. I have a simple site and I am almost the exclusive user of that site, but the memory limitations cause apache and mysql to respond very slowly.

Thanks for listening.

64mb should get you reasonable response times for simple Apache & MySQL usage. You may need to tweak their configurations a bit as by default they're not optimized for systems with that little memory.

Look through the forums to see examples of other people's configurations for Apache & MySQL.

1 user should not cause performance problems on a Linode 64 unless the server is poorly configured or the script (query, whatever) is huge.

I don't have any problems running LAMP on a 64. It's when you get into mail, IMAP, POP3, that it starts to squeeze pretty tight. There's always ways of tuning a server whether it be config or using a different lightweight service if you can get away with it to manage with 64MB.

Also take a min to consider that a bump to 128MB even on 6GB system doesn't leave much room for memory upgrades if clients want them.

Actually, I am doing POP3 and IMAP on it too, and "tight" is exactly what I have been experiencing. I have set the apache and mysql footprints to be very tight for this reason, but still my site responds very slowly and checking my mail is quite sluggish too.

The thing is, this box is my personal "web presence." I want to use it for my email and my web site. This is my basic level of expectation from a $20/month account. To me, this doesn't seem too unreasonable.



If clients want memory upgrades they can be transferred to other boxes where the contention ratio is lower. Or maybe the boxes could have 7GB instead of 6.

I am in a similar situation - I use my Linode for mainly personal use and am the primary user. I run dns, apache, smtp (with some spam filtering), imap, and mysql (very minor use). Traffic is generally light.

And having said that, my performance is acceptable. I've tried to minimize the processes that are running, and my machine swaps very little. My imap performance is very usable, too.

For the money, the 64mb Linode is a wonderful deal. If all you want is email and a web site, perhaps $20 seems high, especially with all the aggressive price drops in the "shared hosting" market these days. But for a UML host, the ability to build and control my own server instance is well worth the cost.

My original plans included getting another (larger) Linode, or perhaps even a small dedicated server to be my "main" online presence, with the 64mb Linode as the main backup for dns and mail, but to be honest the performance of the 64mb has been so good I haven't felt particularly rush to followed through on that.

Not that I would complain if all the nodes got more memory.

I think we should call for a vote.
Would we pay 1-2$ more for these upgrades to the systems?
It would take allot of capital to upgrade the boxes with more memory.

I would be willing to pay a one-time $50 or so to upgrade the existing hardware. I don't think that the base price of $20 should increase though. Memory prices and box capacity are improving all the time. It is not nearly as difficult for Linode to afford 128MB per account as it was when the company first started, and I believe that this much memory would provide a minimum level of acceptable performance for $20/month. There are already many other chrooted systems (not the same, I know) that provide much better performance for the same price. Having moved from a different virtual server account that was working well for me, I'm seriously wondering if I didn't make a bad choice, considering how much slower my linode's response time is. The main reason I switched was because my other account was hosted in Germany and the latency was poor, but the delayed responses I'm getting from apache and imap are even worse.

I have read time and time again how bloated systems are out of the box. If you search and ask I am sure you can find how to bring the performace up with what you have got available.I had lots of trouble with courer eating all my memory.

I think I would pay $30 one time fee to upgrade all the boxes with more memory. If it became an option for anyone to participate in I Would jump at helping everyone become less memory starved.

Just remember though if you have it you will use it. Mem that is.

I have no idea on the profit margins for these things, but I do know this:

You guys can't demand more ram, better support, and still expect everything else. You want the world.

I think that a linode 64 is *great* if you can take the time to tweak it out. Everyone seems to complain -- I'll tell it to you straight, if you think Linode is bad, you've got a new thing coming if you move to somewhere else. This place is *excellent* compared to all others.

_________________
Jay Faulkner
http://oldos.org

Not true. I was with a place called escapebox.net that had excellent service and managed freebsd virtual servers for $25/month. It was truly top-notch and very personalized and skilled service. The sysadmin regularly was right on top of bugtraq and other security issues and even wrote custom patches when project authors were too slow to patch their own software in a timely manner. The main reason I left was because the servers were hosted in Germany so the latency was bad, and they used a traffic shaper to limit maximum throughput. Considering everything but these two issues, however, their package was a better deal than linode.

I think that Linode could still make plenty of money if it configured its boxes with 7GB of RAM and gave basic accounts 128MB. Memory is getting cheaper all the time, and the increased initial costs would be recovered fairly quickly. Consider the fact that the contention ratio is 40 per box for the basic account. This means $20 x 40 per month. That means that each box is bringing in $800/month. 1GB chips are going for less than $200 these days. Assuming the boxes already have 3GB or so in them, that would mean each box would need about 4 more GB. Even at average prices (you can find better) that would mean an expenditure of $800 per box. All this would really mean is a loss of one month of revenue. Ask customers to contribute a one-time fee of $20 to get their account upgraded to 128MB and they would easily go for it, which would mean linode wouldn't have to pay anything. Furthermore, linode can depreciate these expenditures on its tax return. In my opinion, this would really offset linode from the competition and bring in a lot more business too.

Same thing here, exactly.



Your forgetting that probably more than half that $800 is going towards bandwidth fees. Also, if the 64 plans were to be upgraded to 128, all the other plans would have to double their memory as well. If you can't manage to use 64mb efficiently on your own linode, either upgrade, or go with someone else. Given servers do a lot more than they used to 5 years ago, but if I can get away with an entire LAMP setup with Apache2, and PHP5, along with virtual mail for a number of domains on a Linode 64, 64mb ram is more than enough for $20/mo.

If it comes down to anything, caker will start selling new Linodes on new hosts with 128 for probably $30/mo, make the 96's the lowest package at $20/mo, and start lowering costs on everything else, give all the 64 users a price break as well, and offer an upgrade once there's a host available... an upgrade for 50 hosts is a big job. Don't forget your talking about taking down everyone's linode for a period of time even if they don't care about the upgrade and are/have been running fine with what they have.

Actually I was keeping that in mind too. What I said was that he would lose one month of revenue (income), not that there weren't other things that the $800 goes towards paying. I'm guessing that he has enough reserves to make it through this hiccup, but if not, like I said most customers would be willing to pony up $20 one time to get a permanent memory upgrade.



I reject your "love it or leave it" mentality. I've done a lot of optimization already, and I still think the performance is lacking. It's not terrible, but it's not that good either. The response times are especially slow for dynamic web pages that use perl or PHP or that access MySQL.

Implicit in my suggestion was the idea that all service plans would be upgraded; not just the basic plans. As I explained earlier, the contention ratios make it work out pretty well:

40:1 for the basic $20 plan
40 x 128MB = ~5.1GB

30:1 for the $30 plan
30 x 192MB = ~5.8GB

20:1 for the $40 plan
20 x 256MB = ~5.1GB

15:1 for the $60 plan
15 x 384MB = ~5.8GB

10:1 for the $80 plan
10 x 512MB = ~5.1GB

Each plan gets double the memory it had before and each box requires approximately 5 1/2 GB devoted to the UML nodes themselves and however much more it needs for the host OS. This is very doable and practical on today's hardware, and I believe it won't hurt Linode much financially (not at all if the 1-time-fee strategy were adopted) and place it head and shoulders above the competition.



Even 96MB would be a significant improvement. Thanks for your comments.

How certain are you that your performance problems are lack of memory related? It could be the host your on, as some of the hosts do not yet have a sysemu patched host kernel, which can increase mysql performance significantly. Are you watching "vmstat 1" and swap usage as you use your machine?

UML has more overhead than a FreeBSD jail by the nature of UML's design; that's not something we're going to be able to overcome even if you have gigabytes of memory in the guest system.

New host machines (starting with host39) do have 6GB of memory because I do anticipate increasing the default RAM included with all accounts, however at this time I can't give you any more details as to how much of an increase or when that will occur.

-Chris

Just a wild idea but it might be useful to give 1 MB extra RAM per monthly renewal until a set maximum is reached.

This would provide an incentive to sign up annually to get the 12 MB immediately rather than waiting.

And an incentive to continue renewing each month.

Lastly, if people need to reboot to get the extra RAM, it provides an opportunity for their latest-2.x-kernel config to kick-in.

I would gladly pay a one time fee to have my ram increased. I think it's a great idea.