For significant applications, single hard drives don't cut it. The seek times involved for ongoing reads and writes make them inefficient. This can be corrected somewhat using non-fragmenting file systems but even so, as I understand, if there's a bunch of little files continuously being modified (like journal files on a DB) the head's going to be jumping around too much and the app is going to be waiting on it.
If they're going to pack this much data on a drive, they'd better speed it up.
I just went through this decision-making process with the consultants who are going to build my company's OSS.
While OODBMS were an obvious choice to me for performance and ease of programming, my consultants told me that finding Oracle talent was so much easier than finding Versant talent (for example) that I would be wasting time and money using OODBMS. This is especially true of DBAs.
Once upon a time, dialup actually gave you a shot at paying your rent. A PRI or two, a MAX for ISDN, 10x dial oversubscription, Sendmail + QPOP, Apache with Virtual Hosting run on a dual Pentium Pro with a T1 to whatevernet and you had a business that could make you a few $k a month. If you could swing the occasional business T1, hell you could pay a couple of college guys $10 an hour to answer your phones for ya.
Then came broadband. DSL first, and it didn't look too awful. The local CLEC (Covad/Northpoint) would sell you a T1 to mux your customers on. You wouldn't make too terribly much margin but it was a little extra dough so whatever. But then the installs... the two guys you hired to answer the phone ended up spending half their day on the phone with your CLEC. You noticed your support starts to slip and your customers are bleating for broadband, the same service which is starting to drag you down.
Then the big trouble starts. The CLECs decide T1s aren't worth their weight in gold so they force you to buy a T3. That's big time $$$ monthly compared to what you're used to so you try to hold the fort with the T1 you've got. Fortunately you're under contract so they leave you alone. As for business... sure, some dial customers are happy where they're at and/or can't get DSL at their houses so you try to focus on them. But broadband is catching on and the Free ISPs are starting to pop up. You're down a $k or two in monthly revenue.
Now the LECs are in the DSL business. A T1 or T3 to them to resell DSL? Forget it... you can only make $10 margin on an ADSL customer because the LEC is price gouging the industry looking to kill the CLECs and, therefore, you. So you cling to the CLEC and the T1 you started with but the contract's done now and they're tired of your righteous yelling... you don't offer DSL anymore. Your customers are leaving dial in droves, the two guys who used to answer your phones have graduated and that ole dual Pentium Pro that runs your services needs upgrading. With all this beating on your mind, on the first sunny weekend in the spring, you spend your weekend piecing your webserver back from backups because a bunch of 13 year old IRC geeks have taken you down with the latest Trojan that you were a bit too slow to implement. You turn the lights off and decide to get a real job.
Covad will almost certainly follow Northpoint into oblivion (unless the resettling of NP's customers gives them an unexpected boost. They're in a terrible financial position. From what I've heard, that's the stock market's fault but maybe not in the way you'd expect.
Apparently, Covad's stock price and ability to secure more funding were based upon the number of businesses and residences it served. With that in mind, they built a network that spanned to the burbs and further. And it worked, they got a lot of funding, funding that wanted them to keep doing it. Now they're overstretched with an enormous debt. Recently, they shut down 150 COs. More will certainly follow and their $3 or so stock price won't draw a lot of investor interest. They're still in the hole and are working furiously to get their burn rate down before their operating capital dwindles down to nada.
The LECs really had their way with Covad. First, they played the traditional "F the CLEC" game. That was fun for a while (not to mention effective). But Covad kept selling circuits so the LECs had a sitdown and decided to kill them. And out they came with their ADSL product, at about $50... $100 or more less than Covad. And of course their product managed to get installed a tad more quickly than Covad's did.
I've been wondering about this for years ...
on
Preview Of Linux 2.5
·
· Score: 1
Linux IO has always seemed slow to me.
On large file ops, the update daemon starts pegging the CPU and the writes steadily slow down to a trickle. Example... I backup my MP3 collection (about 5 gigs) over 100Mb every so often onto my gateway box, running Linux with ext2 file systems. The transfer takes at least twice as long on Linux as it used to when FreeBSD was running on it, not to mention the Linux box becomes pretty much unusable during this period.
Do the journaling FSs, e.g. ReiserFS, make the update daemon, and so this problem, go away? Will tweaking update help?
Slashdot Mirror
For significant applications, single hard drives don't cut it. The seek times involved for ongoing reads and writes make them inefficient. This can be corrected somewhat using non-fragmenting file systems but even so, as I understand, if there's a bunch of little files continuously being modified (like journal files on a DB) the head's going to be jumping around too much and the app is going to be waiting on it. If they're going to pack this much data on a drive, they'd better speed it up.
I just went through this decision-making process with the consultants who are going to build my company's OSS. While OODBMS were an obvious choice to me for performance and ease of programming, my consultants told me that finding Oracle talent was so much easier than finding Versant talent (for example) that I would be wasting time and money using OODBMS. This is especially true of DBAs.
Then came broadband. DSL first, and it didn't look too awful. The local CLEC (Covad/Northpoint) would sell you a T1 to mux your customers on. You wouldn't make too terribly much margin but it was a little extra dough so whatever. But then the installs ... the two guys you hired to answer the phone ended up spending half their day on the phone with your CLEC. You noticed your support starts to slip and your customers are bleating for broadband, the same service which is starting to drag you down.
Then the big trouble starts. The CLECs decide T1s aren't worth their weight in gold so they force you to buy a T3. That's big time $$$ monthly compared to what you're used to so you try to hold the fort with the T1 you've got. Fortunately you're under contract so they leave you alone. As for business ... sure, some dial customers are happy where they're at and/or can't get DSL at their houses so you try to focus on them. But broadband is catching on and the Free ISPs are starting to pop up. You're down a $k or two in monthly revenue.
Now the LECs are in the DSL business. A T1 or T3 to them to resell DSL? Forget it ... you can only make $10 margin on an ADSL customer because the LEC is price gouging the industry looking to kill the CLECs and, therefore, you. So you cling to the CLEC and the T1 you started with but the contract's done now and they're tired of your righteous yelling ... you don't offer DSL anymore. Your customers are leaving dial in droves, the two guys who used to answer your phones have graduated and that ole dual Pentium Pro that runs your services needs upgrading. With all this beating on your mind, on the first sunny weekend in the spring, you spend your weekend piecing your webserver back from backups because a bunch of 13 year old IRC geeks have taken you down with the latest Trojan that you were a bit too slow to implement. You turn the lights off and decide to get a real job.
Apparently, Covad's stock price and ability to secure more funding were based upon the number of businesses and residences it served. With that in mind, they built a network that spanned to the burbs and further. And it worked, they got a lot of funding, funding that wanted them to keep doing it. Now they're overstretched with an enormous debt. Recently, they shut down 150 COs. More will certainly follow and their $3 or so stock price won't draw a lot of investor interest. They're still in the hole and are working furiously to get their burn rate down before their operating capital dwindles down to nada.
The LECs really had their way with Covad. First, they played the traditional "F the CLEC" game. That was fun for a while (not to mention effective). But Covad kept selling circuits so the LECs had a sitdown and decided to kill them. And out they came with their ADSL product, at about $50 ... $100 or more less than Covad. And of course their product managed to get installed a tad more quickly than Covad's did.
Linux IO has always seemed slow to me.
On large file ops, the update daemon starts pegging the CPU and the writes steadily slow down to a trickle. Example ... I backup my MP3 collection (about 5 gigs) over 100Mb every so often onto my gateway box, running Linux with ext2 file systems. The transfer takes at least twice as long on Linux as it used to when FreeBSD was running on it, not to mention the Linux box becomes pretty much unusable during this period.
Do the journaling FSs, e.g. ReiserFS, make the update daemon, and so this problem, go away? Will tweaking update help?