Menu
▾
▴
[SM-USERS] Re: Clustering
|
From: David R. <d-...@ad...> - 2003-03-07 22:23:38
|
Dear Matt: Yes, the NAS device supports many different kinds of file sharing protocols, including NFS and SMB. The setup procedure should take no more than 10-15 minutes, which includes plugging the device in and setting basic configuration. Then, file sharing will be available for use by your favorite protocol, whether than be SMB or NFS. You could pull a "mount -t smbfs ..." which is the command that uses SMBMOUNT, if you want to use SMB to share your files. Or you could pull a standard "mount ..." if you want to use nfs. Alternatively, you could write either to the /etc/fstab file to automatically mount at boot time. One thing you might want to investigate is which protocol has the better (or any) locking mechanisms, if you are running multiple daemons on the same data files concurrently. Note, however, that some daemons, like databases (mysql, pgsql) won't be able to run on the same data files even with file locking. In general, if the daemon periodically writes to a file, it should be fine. In the case of mysql, the daemon opens the file once, stores its contents to memory, then systematically writes without reading the file again. As you can see, this is a problem. Also, NAS is a general term describing a set of devices, below the SAN devices, which use Fibre Channel and are much more expensive. The procedures described above apply to the POPNETSERVER, and may apply to others. You'll just have to check the documentation that comes with them. Hope this helps! David Matt wrote: >Question on this NAS... >I'm not familiar with using NAS with Linux. >What all is involved with making Linux use the NAS as a mounted >directory? Do you just basically SMBMOUNT it onto the system? > >On Mon, 2003-03-03 at 16:52, David R. wrote: > > >>Dear Matt: >> >>It's an NAS device. (Network Attached Storage). It's like an >>ultra-redundant very stable network appliance for storage, comes with an >>embedded OS for sharing files across many protocols and managing the >>device, and is connected to other machines through a LAN. Try >>http://www.google.com/search?q=POPNETSERVER+4500. You'll find a lot of >>information on the actual device there. >> >>Thanks! >> >>David >> >> >>Matt wrote: >> >> >> >>>Now what exactly is the popnetserver? I don't see a detailed >>>description on this page. >>>Is it just an external drive farm that you stick drives in and then >>>connect through like a fiber link or something to your main system? >>> >>>On Mon, 2003-03-03 at 10:18, David R. wrote: >>> >>> >>> >>> >>>>Dear Matt: >>>> >>>>You can find the FIA POPNETSERVER 4500 here: >>>>http://www.computergiants.com/items/cat_1.asp?category=87&cid=NAS%20STORAGE >>>> >>>>Notice also that a simple search and pricewatch.com should turn up >>>>several merchants. >>>> >>>>Hope this helps! >>>> >>>>David Rusenko >>>> >>>> >>>>Matt wrote: >>>> >>>> >>>> >>>> >>>> >>>>>David, >>>>>Do you have a source for the FIA POPNETSERVER4500 ? I couldn't find it. >>>>> >>>>>On Sat, 2003-03-01 at 16:09, David R. wrote: >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>>>>Dear Matt: >>>>>> >>>>>>There are a lot of different options for clustering, including without a "backend" as you call it. Many times you might use a utility such as "rsync" to keep the devices sync'ed near real-time. However, if you use qmail, you will run into some problems. As far as I know, qmail does not use the filesystem structure for it's queue, but rather addresses them directly by inode... This means you cannot simple copy files over... In that case, you might need a "backend" or shared storage solution... >>>>>> >>>>>>Also, you seem to want an Active-Active cluster configuration. I'm not sure if you are familiar with clustering terms or not, so I will list out the most common ones. Feel free to skip this next section if you already know this. >>>>>> >>>>>>- A "Director" is the server that receives packets, and does the load balancing. >>>>>>- A "Real Server" is a server that is a "node in the cluster". >>>>>>- "Failover" is what happens when something goes down. >>>>>>- "Director failover" is what happens when the load balancer goes down. >>>>>>- Usually "Real server failover" is handled by the load balancing software, but not always. >>>>>>- An "Active-passive" cluster configuration is a configuration where one server handles requests, and the other is passive, waiting to take over if the active server goes down. Active-passive does not usually involve load balancing. >>>>>>- An "Active-Active" configuration is where two servers are both active. If one of them goes down, the other takes over it's IP and/or services. >>>>>>- "High availability" refers to the ability of a cluster to "never go down". >>>>>>- "Load balanced" refers to the ability of a cluster to distribute load among several machines. HA and load balancing are not always combined. >>>>>> >>>>>>As I stated above, it seems as though you are interested in an Active-Active configuration, with two machines, no shared storage, HA and load balanced. There are many different software packages you may want to investigate. >>>>>> >>>>>>- "heartbeatd" is a package that handles high availability, specifically, failover between machines. It is typically used in an active-passive configuration, although could be used in an active-active configuration. It is not *typically* used in a load balanced environment. It might need to be combined with many other software packages to be useful. It is currently in a stable production release. >>>>>>- "keepalived" is a package that handles load balancing through the IPVS kernel patch (very well known and reliable), and director failover through the VRRP (Virtual Router Redundancy Protocol), which, among other things, is a very elegant election protocol. It also handles real server failover, which is not handled by LVS (Linux Virtual Server, the IPVS patch) It is currently in a stable production release. >>>>>>- You might also want to search for "UltraMonkey", a project that attempts to simplify implementation of existing clustering solutions, and the FreeBSD VRRP implementation (if you are using FreeBSD). >>>>>> >>>>>>One last note. Usually, the director and real servers are on different machines. There are more advanced possibilities, however. I am currently setting up a system for my company that will be a P2P clustering solution, of sorts. It works by integrating the Director and the Real Servers. When the director goes down, any of the real servers can take over this position, and continue to be real servers as well. It is a very interesting implementation, but is a bit more advanced than your standard implementations. I will be using "keepalived", and I would highly recommend it for your situation. >>>>>> >>>>>>I would also recommend investing in some sort of shared storage solution, such as a NAS device. These can be fairly cheap, some models (FIA POPNETSERVER4500 for example) full of features and starting at only ~$1500. >>>>>> >>>>>>Hope this helps! >>>>>> >>>>>>David Rusenko >>>>>>Aderes.net >>>>>> >>>>>> >>>>>>Matt Hoppes Wrote: >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>>>I could do that... except that that doesn't sync them. To have 2 >>>>>>>machines doing load balancing they need to be able to sync up their files >>>>>>>every now and then.. I don't know.. maybe it's not possible to do without >>>>>>>a backend. >>>>>>> >>>>>>> >>>>>>> >>>>>>>-- >>>>>>> >>>>>>> >>>>>>>Matt Hoppes >>>>>>>ChiliTech Internet Solutions >>>>>>>Network Operation Center >>>>>>>(570) 323-2166 x 126 >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>>>Matt, >>>>>>>>I have not done qmail per say, but could you not just copy the >>>>>>>>directories over? >>>>>>>> >>>>>>>>I mean that is the simple way >>>>>>>> >>>>>>>>ssh -llogin mac1 tar cvf /mail/dirs >>>>>>>>scp mac1:/mail/dirs.tar . >>>>>>>>tar xvf mail/dirs.tar >>>>>>>> >>>>>>>>I know this is not real time or fool proof, but you didn't say how you >>>>>>>>really wanted it? >>>>>>>> >>>>>>>>Ron >>>>>>>> >>>>>>>>On 28 Feb 2003, Matt wrote: >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>>>>>Date: 28 Feb 2003 14:54:33 -0500 >>>>>>>>>>From: Matt >>>>>>>>>>To: qm...@li... >>>>>>>>>>Subject: Clustering >>>>>>>>>> >>>>>>>>>>Has anyone ever done anything like this: >>>>>>>>>>Have 2 machines be mail servers. >>>>>>>>>>A basically mirror themselves. in otherwords... >>>>>>>>>>No central backend of RAID drives.... but rather 2 machines mirroring >>>>>>>>>>each other over some form of a network (cross-over cable or otherwise) >>>>>>>>>>-- >>>>>>>>>>Matt Hoppes >>>>>>>>>>ChiliTech Internet Solutions >>>>>>>>>>Network Operations Center >>>>>>>>>>(570) 323-2166 x 126 >>>>>>>>>>http://www.chilitech.net >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>------------------------------------------------------- >>>>>>This sf.net email is sponsored by:ThinkGeek >>>>>>Welcome to geek heaven. >>>>>>http://thinkgeek.com/sf >>>>>>-- >>>>>>squirrelmail-users mailing list >>>>>>List Address: squ...@li... >>>>>>List Archives: http://sourceforge.net/mailarchive/forum.php?forum_id=2995 >>>>>>List Info: https://lists.sourceforge.net/lists/listinfo/squirrelmail-users >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>> >>>>------------------------------------------------------- >>>>This sf.net email is sponsored by:ThinkGeek >>>>Welcome to geek heaven. >>>>http://thinkgeek.com/sf >>>>-- >>>>squirrelmail-users mailing list >>>>List Address: squ...@li... >>>>List Archives: http://sourceforge.net/mailarchive/forum.php?forum_id=2995 >>>>List Info: https://lists.sourceforge.net/lists/listinfo/squirrelmail-users >>>> >>>> >>>> >>>> >>>> >> >> >>------------------------------------------------------- >>This sf.net email is sponsored by:ThinkGeek >>Welcome to geek heaven. >>http://thinkgeek.com/sf >>-- >>squirrelmail-users mailing list >>List Address: squ...@li... >>List Archives: http://sourceforge.net/mailarchive/forum.php?forum_id=2995 >>List Info: https://lists.sourceforge.net/lists/listinfo/squirrelmail-users >> >> >> |