Hi Nadav! On Monday 07 May 2007, Nadav Har'El wrote: > On Sun, May 06, 2007, Shlomi Fish wrote about "RFC: Plea for a New Hamakor Server": > > At the moment, we have two problems with Eskimo: > > 1. Our low-end IDE disk that was bought to be used by the mirrors, is > > giving us many hardware problems. > > Can you eleborate? Did the disks fail?
Sure. The hard disk causes many problems, and often becomes read-only, even to root, which can only be reverted using a reboot or an fsck operation. > Are "low-end IDE disks" more prone > to hardware failure than some other types of disks (are there high-end IDE > disks? SCSI disks? or what?) I don't know, but I think so. > > > 2. The machine (a Compaq ProLiant 500 machine) tends to be very slow for > > today's demanding load. Plus, the board has a maximal RAM of 1 GB which > > is becoming hardly enough. While it is a good machine and can be re-used > > as a server in a different context, we would like to upgrade to a machine > > with better specs. > > I'm not saying that a new server isn't needed, but I wonder - for what kind > of "today's demanding load" is 1 GB ram and the slow CPU not enough? If I > remember correctly, eskimo hardly serves thousands of concurrent users... Well, 320 MB, which we have now is not enough for qmail, Apache 1 with many PHPs, a Perl CGI application, ssh access, FTP, etc. And I really would like to upgrade to postfix because qmail suffers from this - http://www.shlomifish.org/open-source/anti/qmail/ and is not FOSS. Postfix is a very fine SMTP server, but it is somewhat more memory hungry than qmail is. And like I said, we want: 1. mod_php. 2. A mod_php "accelerator", that will cache the PHP code. (Recommended for MW) 3. mod_perl 4. Other services. All of this will probably be more than 1 GB. And the more memory we have the more we can scale and increase the speed. > > > On the other hand beak, which is a new Pentium 4 server with more RAM has > > a huge hard-disk space problem: > > > > {{{{{{{{{{{{{{{ > > Filesystem 1K-blocks Used Available Use% Mounted on > > /dev/md0 3421584 2463168 958416 72% / > > tmpfs 257956 0 257956 0% /dev/shm > > /dev/md1 133527952 130207836 3320116 98% /srv > > }}}}}}}}}}}}}}} > > > > However, upgrading its hard-disks (which must be SCSI) would be costy > > until we could use a new server. > > This lesson is extremely important to remember when we buy a new server: > the first component of most servers to "age" (i.e., look not enough) is the > disk space. Servers must be bought with the assumption that in just 2 or 3 > years, the disks will be replaced even if the machine itself isn't. > > Then again, SCSI disks can be replaced as well. Just how "costy" would that > be? > > Also, since you seem not to recommend IDE disks (see above) and SCSI disks > (see now), what kind of disks do you propose? > I don't mind getting a SCSI disks upgrade for beak. This is definitely an option. But I was told they cost quite a bit. Of course, assuming we are moving all of our mirrors to ISOC, we can certainly do well with less space on beak, as the mirrors are the hungry resource of hard-disk. > > We would like to merge both servers into a new one that will be the > > host. We are already moving our mirrors to ISOC, but need a server for > > providing the web-services, the email, DNS and so forth. > > Another issue to note is that beak and eskimo currently have diametrically > opposed system-administration philosophies, and this difference of > philosophies would need to be reconciled somehow before the two can be > merged. > Well, yes. I think that there should be less territorialism in the management of the servers. Heinlein once remarked that "Specialisation is for insects" and he was right. While each one of us has his strengthes and weaknesses, we should document how to modify things on the server, what we are doing, how others can do it, and allow other people to do it themselves. That's what we have a wiki for. > > 1. The board and the processor should have a maximum memory of at least > > 8-12 GB. > > A noble target, but I wonder if we'll really need such numbers for the > expected load we'll have in the next 5 years. It won't hurt, of course, > but if this demand doubles the price, I wonder if it's necessary. Hmmm... I'd rather pay some money now than risk another upgrade. While 4 GB are a lot, I believe we need more. Many servers nowadays can scale to more than that. (Xeons, X86-64, etc.) > > P.S. A question for the Linux gurus on this list: can Linux handle more > than 4 GB of memory? I'm not talking about process memory, but rather the > kernel's cache (which is very useful for web servers) - can Linux keep more > than 4 GB of files in its filesystem cache? > (of course, this question is only relevant to 32 bit architectures). > > > 5. Hard disk - I think we need at least 500 GB, in several disks. > > Again, the possibility of a future upgrade is even more important than > the current disks. If 500 GB seems enough for you now, in 2 years, you'll > feel choked without 1 TB... So plan ahead. > Sure. > > 7. The server needs to be a high-end server. Low-end computers that are > > sold for desktops, etc. tend to not last too long, and require a lot of > > mainteance. > > One important criterion you forgot is the machine's form-factor. Are you > looking of a 1U machine, a tower machine (like a normal desktop computer), > or what? I think Actcom can host almost anything that's not overly large. At the moment, Eskimo is a tower machine, and I don't remember what Beak is. Since we're deprecating Eskimo we can just get another tower machine. Regards, Shlomi Fish --------------------------------------------------------------------- Shlomi Fish [EMAIL PROTECTED] Homepage: http://www.shlomifish.org/ If it's not in my E-mail it doesn't happen. And if my E-mail is saying one thing, and everything else says something else - E-mail will conquer. -- An Israeli Linuxer
