Re: [PERFORM] Putting tables or indexes in SSD or RAM: avoiding double caching?
Well for one thing on the IODrive. Be sure to use a FS that supports direct IO so you don't cache it on the FS level and thus take room an object not on SSD could use. We use vxfs with mincache=direct as our filesystem for just this reason. Also, there is an IO drive tuning manual that discusses the same. It's a good read if you don't already have it. I do not know of a way to partition the PG cache other than make it small and use the FS controls to force direct IO. 2.5 cents.. -kg -Original Message- From: pgsql-performance-ow...@postgresql.org on behalf of Shaul Dar Sent: Mon 5/25/2009 6:51 AM To: pgsql-performance@postgresql.org Subject: [PERFORM] Putting tables or indexes in SSD or RAM: avoiding double caching? Hi, I have sen many posts on using SSDs, and iodrive http://www.fusionio.comin particular, to accelerate the performance of Postgresql (or other DBMS) -- e.g. this discussionhttp://groups.google.co.il/group/pgsql.performance/browse_thread/thread/1d6d7434246afd97?pli=1. I have also seen the suggestion to use RAM for the same purpose by creating a tablespace on a RAM mount point.http://magazine.redhat.com/2007/12/12/tip-from-an-rhce-memory-storage-on-postgresql/Granted these make most sense when the whole database cannot fit into main memory, or if we want to avoid cold DB response times (i.e waiting for the DB to warm up as stuff gets cached in memory). My question is this: if we use either SSD or RAM tablespaces, I would imagine postgresql will be oblevient to this and would still cache the tablespace elemenst that are on SSD or RAM into memory - right? Is there a way to avoid that, i.e. to tell postgress NOT to cache tablespaces, or some other granularity of the DB? Thanks, -- Shaul *Dr. Shaul Dar* Email: i...@shauldar.com Web: www.shauldar.com
Re: [PERFORM] Problems with autovacuum
Tom Lane escribió: =?UTF-8?B?xYF1a2FzeiBKYWdpZcWCxYJv?= lukasz.jagie...@gforces.pl writes: That autovacuum working hole time, shoudn't be run only when db needs ? With 2000 databases to cycle through, autovac is going to be spending quite a lot of time just finding out whether it needs to do anything. I believe the interpretation of autovacuum_naptime is that it should examine each database that often, ie once a minute by default. So it's got more than 30 databases per second to look through. Note that this is correct in 8.1 and 8.2 but not 8.3 onwards. -- Alvaro Herrerahttp://www.CommandPrompt.com/ The PostgreSQL Company - Command Prompt, Inc. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
Alvaro Herrera alvhe...@commandprompt.com writes: Tom Lane escribió: I believe the interpretation of autovacuum_naptime is that it should examine each database that often, ie once a minute by default. So it's got more than 30 databases per second to look through. Note that this is correct in 8.1 and 8.2 but not 8.3 onwards. Oh? The current documentation still defines the variable thusly: Specifies the minimum delay between autovacuum runs on any given database. In each round the daemon examines the database and issues VACUUM and ANALYZE commands as needed for tables in that database. I suppose the use of minimum means that this is not technically incorrect, but it's sure not very helpful if there is some other rule involved that causes it to not behave as I said. (And if there is some other rule, what is that?) Please improve the docs. regards, tom lane -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
Tom Lane escribió: Alvaro Herrera alvhe...@commandprompt.com writes: Tom Lane escribi�: I believe the interpretation of autovacuum_naptime is that it should examine each database that often, ie once a minute by default. So it's got more than 30 databases per second to look through. Note that this is correct in 8.1 and 8.2 but not 8.3 onwards. Oh? The current documentation still defines the variable thusly: Specifies the minimum delay between autovacuum runs on any given database. In each round the daemon examines the database and issues VACUUM and ANALYZE commands as needed for tables in that database. Sorry, it's the other way around actually -- correct for 8.3 onwards, wrong for 8.1 and 8.2. In the earlier versions, it would do one run in a chosen database, sleep during naptime, then do another run. I suppose the use of minimum means that this is not technically incorrect, but it's sure not very helpful if there is some other rule involved that causes it to not behave as I said. (And if there is some other rule, what is that?) The word minimum is there because it's possible that all workers are busy with some other database(s). Please improve the docs. I'll see about that. -- Alvaro Herrerahttp://www.CommandPrompt.com/ PostgreSQL Replication, Consulting, Custom Development, 24x7 support -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
W dniu 26 maja 2009 20:28 użytkownik Tom Lane t...@sss.pgh.pa.us napisał: I believe the interpretation of autovacuum_naptime is that it should examine each database that often, ie once a minute by default. So it's got more than 30 databases per second to look through. Note that this is correct in 8.1 and 8.2 but not 8.3 onwards. Oh? The current documentation still defines the variable thusly: Specifies the minimum delay between autovacuum runs on any given database. In each round the daemon examines the database and issues VACUUM and ANALYZE commands as needed for tables in that database. I suppose the use of minimum means that this is not technically incorrect, but it's sure not very helpful if there is some other rule involved that causes it to not behave as I said. (And if there is some other rule, what is that?) Please improve the docs. After change autovacuum_naptime postgresql behave like you wrote before. -- Łukasz Jagiełło System Administrator G-Forces Web Management Polska sp. z o.o. (www.gforces.pl) Ul. Kruczkowskiego 12, 80-288 Gdańsk Spółka wpisana do KRS pod nr 246596 decyzją Sądu Rejonowego Gdańsk-Północ -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
Alvaro Herrera alvhe...@commandprompt.com writes: Sorry, it's the other way around actually -- correct for 8.3 onwards, wrong for 8.1 and 8.2. In the earlier versions, it would do one run in a chosen database, sleep during naptime, then do another run. Tom Lane escribió: I suppose the use of minimum means that this is not technically incorrect, but it's sure not very helpful if there is some other rule involved that causes it to not behave as I said. (And if there is some other rule, what is that?) The word minimum is there because it's possible that all workers are busy with some other database(s). Please improve the docs. I'll see about that. Hmm, maybe we need to improve the code too. This example suggests that there needs to be some limit on the worker launch rate, even if there are so many databases that that means we don't meet naptime exactly. regards, tom lane -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
[PERFORM] Hosted servers with good DB disk performance?
I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). Is anyone aware of a company that offers such a thing? -- * Greg Smith gsm...@gregsmith.com http://www.gregsmith.com Baltimore, MD -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
Greg Smith wrote: What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). We tried this with poor results. Most of the co-location and server-farm places are set up with generic systems that are optimized for small-to-medium-sized web sites. They use MySQL and are surprised to hear there's an alternative open-source DB. They claim to be able to create custom configurations, but it's a lie. The problem is that they run on thin profit margins, and their techs are mostly ignorant, they just follow scripts. If something goes wrong, or they make an error, you can't get anything through their thick heads. And you can't go down there and fix it yourself. For example, we told them EXACTLY how to set up our system, but they decided that automatic monthly RPM OS updates couldn't hurt. So the first of the month, we in the morning to find that Linux had been updated to libraries that were incompatible with our own software, the system automatically rebooted and our web site was dead. And many similar incidents. We finally bought some nice Dell servers and found a co-location site that provides us all the infrastructure (reliable power, internet, cooling, security...), and we're in charge of the computers. We've never looked back. Craig -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, May 26, 2009 at 11:51 PM, Greg Smith gsm...@gregsmith.com wrote: I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. Perhaps you'll be satisfied with http://www.ovh.co.uk/products/dedicated_list.xml ? Personally I have only one machine there (SuperPlan Mini) - I asked them to set up Proxmox (http://pve.proxmox.com/wiki/Main_Page ) for me and now I have four OpenVZ Linux containers with different setup and services. So far I can't be more happy. Regards, Marcin -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
Depends on the level of facility you are looking for. Peer1 (www.peer1.com) will sell you just about whatever you need contained in a single box and I believe their Atlanta facility and some others have a managed SAN option. Since you want a customized solution, make sure you talk with one of their solutions engineers. Another good option in this range up to mid-enterprise hosting solutions is Host My Site (www.hostmysite.com). On the very high end of the spectrum, gni (www.gni.com) seems to provide a good set of infrastructure as a service (IAAS) solutions including SAN storage and very high bandwidth - historically they have been very successful in the MPOG world. If you are interested, I can put you in touch with real people who can help you at all three organizations. Jerry Champlin|Absolute Performance Inc. -Original Message- From: pgsql-performance-ow...@postgresql.org [mailto:pgsql-performance-ow...@postgresql.org] On Behalf Of Greg Smith Sent: Tuesday, May 26, 2009 3:51 PM To: pgsql-performance@postgresql.org Subject: [PERFORM] Hosted servers with good DB disk performance? I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). Is anyone aware of a company that offers such a thing? -- * Greg Smith gsm...@gregsmith.com http://www.gregsmith.com Baltimore, MD -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, 2009-05-26 at 17:51 -0400, Greg Smith wrote: I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). Is anyone aware of a company that offers such a thing? Sure, CMD will do it, so will Rack Space and a host of others. If you are willing to go with a VPS SliceHost are decent folk. CMD doesn't rent hardware you would have to provide that, Rack Space does. Joshua D. Drake -- PostgreSQL - XMPP: jdr...@jabber.postgresql.org Consulting, Development, Support, Training 503-667-4564 - http://www.commandprompt.com/ The PostgreSQL Company, serving since 1997 -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
Tom Lane escribió: Hmm, maybe we need to improve the code too. This example suggests that there needs to be some limit on the worker launch rate, even if there are so many databases that that means we don't meet naptime exactly. We already have a 100ms lower bound on the sleep time (see launcher_determine_sleep()). Maybe that needs to be increased? -- Alvaro Herrerahttp://www.CommandPrompt.com/ The PostgreSQL Company - Command Prompt, Inc. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
Alvaro Herrera alvhe...@commandprompt.com writes:
Tom Lane escribió:
Hmm, maybe we need to improve the code too. This example suggests that
there needs to be some limit on the worker launch rate, even if there
are so many databases that that means we don't meet naptime exactly.
We already have a 100ms lower bound on the sleep time (see
launcher_determine_sleep()). Maybe that needs to be increased?
Maybe. I hesitate to suggest a GUC variable ;-)
One thought is that I don't trust the code implementing the minimum
too much:
/* 100ms is the smallest time we'll allow the launcher to sleep */
if (nap-tv_sec = 0 nap-tv_usec = 10)
{
nap-tv_sec = 0;
nap-tv_usec = 10; /* 100 ms */
}
What would happen if tv_sec is negative and tv_usec is say 50?
Maybe negative tv_sec is impossible here, but ...
regards, tom lane
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
Tom Lane escribió:
Alvaro Herrera alvhe...@commandprompt.com writes:
Tom Lane escribi�:
Hmm, maybe we need to improve the code too. This example suggests that
there needs to be some limit on the worker launch rate, even if there
are so many databases that that means we don't meet naptime exactly.
We already have a 100ms lower bound on the sleep time (see
launcher_determine_sleep()). Maybe that needs to be increased?
Maybe. I hesitate to suggest a GUC variable ;-)
Heh :-)
One thought is that I don't trust the code implementing the minimum
too much:
/* 100ms is the smallest time we'll allow the launcher to sleep */
if (nap-tv_sec = 0 nap-tv_usec = 10)
{
nap-tv_sec = 0;
nap-tv_usec = 10; /* 100 ms */
}
What would happen if tv_sec is negative and tv_usec is say 50?
Maybe negative tv_sec is impossible here, but ...
I don't think it's possible to get negative tv_sec here currently, but
perhaps you're right that we could make this code more future-proof.
However I think there's a bigger problem here, which is that if the user
has set naptime too low, i.e. to a value lower than
number-of-databases * 100ms, we'll be running the (expensive)
rebuild_database_list function on each iteration ... maybe we oughta put
a lower bound on naptime based on the number of databases to avoid this
problem.
--
Alvaro Herrerahttp://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On 5/26/09, Greg Smith gsm...@gregsmith.com wrote: I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). Is anyone aware of a company that offers such a thing? www.contegix.com offer just about the best support I've come across and are familiar with Postgres. They offer RHEL (and windows) managed servers on a variety of boxes. They're not a budget outfit though, but that's reflected in the service. -- Dave Page EnterpriseDB UK: http://www.enterprisedb.com -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, 26 May 2009, Joshua D. Drake wrote: CMD doesn't rent hardware you would have to provide that, Rack Space does. Part of the idea was to avoid buying a stack of servers, if this were just a where do I put the boxes at? problem I'd have just asked you about it already. I forgot to check Rack Space earlier, looks like they have Dell servers with up to 8 drives and a RAID controller in them available. Let's just hope it's not one of the completely useless PERC models there; can anyone confirm Dell's PowerEdge R900 has one of the decent performing PERC6 controllers I've heard rumors of in it? Craig, I share your concerns about outsourced hosting, but as the only custom application involved is one I build my own RPMs for I'm not really concerned about the system getting screwed up software-wise. The idea here is that I might rent an eval system to confirm performance is reasonable, and if it is then I'd be clear to get a bigger stack of them. Luckily there's a guy here who knows a bit about benchmarking for this sort of thing... -- * Greg Smith gsm...@gregsmith.com http://www.gregsmith.com Baltimore, MD -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
Greg, I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. Joyent will guarentee you a certain amount of disk bandwidth. As far as I know, they're the only hoster who does. -- Josh Berkus PostgreSQL Experts Inc. www.pgexperts.com -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On 5/26/09 6:17 PM, Greg Smith gsm...@gregsmith.com wrote: On Tue, 26 May 2009, Joshua D. Drake wrote: CMD doesn't rent hardware you would have to provide that, Rack Space does. Part of the idea was to avoid buying a stack of servers, if this were just a where do I put the boxes at? problem I'd have just asked you about it already. I forgot to check Rack Space earlier, looks like they have Dell servers with up to 8 drives and a RAID controller in them available. Let's just hope it's not one of the completely useless PERC models there; can anyone confirm Dell's PowerEdge R900 has one of the decent performing PERC6 controllers I've heard rumors of in it? Every managed hosting provider I've seen uses RAID controllers and support through the hardware provider. If its Dell its 99% likely a PERC (OEM'd LSI). HP, theirs (not sure who the OEM is), Sun theirs (OEM'd Adaptec). PERC6 in my testing was certainly better than PERC5, but its still sub-par in sequential transfer rate or scaling up past 6 or so drives in a volume. I did go through the process of using a managed hosting provider and getting custom RAID card and storage arrays -- but that takes a lot of hand-holding and time, and will most certainly cause setup delays and service issues when things go wrong and you've got the black-sheep server. Unless its absolutely business critical to get that last 10%-20% performance, I would go with whatever they have with no customization. Most likely if you ask for a database setup, they'll give you 6 or 8 drives in raid-5. Most of what these guys do is set up LAMP cookie-cutters... Craig, I share your concerns about outsourced hosting, but as the only custom application involved is one I build my own RPMs for I'm not really concerned about the system getting screwed up software-wise. The idea here is that I might rent an eval system to confirm performance is reasonable, and if it is then I'd be clear to get a bigger stack of them. Luckily there's a guy here who knows a bit about benchmarking for this sort of thing... -- * Greg Smith gsm...@gregsmith.com http://www.gregsmith.com Baltimore, MD -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, May 26, 2009 at 7:41 PM, Scott Carey sc...@richrelevance.com wrote: On 5/26/09 6:17 PM, Greg Smith gsm...@gregsmith.com wrote: On Tue, 26 May 2009, Joshua D. Drake wrote: CMD doesn't rent hardware you would have to provide that, Rack Space does. Part of the idea was to avoid buying a stack of servers, if this were just a where do I put the boxes at? problem I'd have just asked you about it already. I forgot to check Rack Space earlier, looks like they have Dell servers with up to 8 drives and a RAID controller in them available. Let's just hope it's not one of the completely useless PERC models there; can anyone confirm Dell's PowerEdge R900 has one of the decent performing PERC6 controllers I've heard rumors of in it? Every managed hosting provider I've seen uses RAID controllers and support through the hardware provider. If its Dell its 99% likely a PERC (OEM'd LSI). HP, theirs (not sure who the OEM is), Sun theirs (OEM'd Adaptec). PERC6 in my testing was certainly better than PERC5, but its still sub-par in sequential transfer rate or scaling up past 6 or so drives in a volume. I did go through the process of using a managed hosting provider and getting custom RAID card and storage arrays -- but that takes a lot of hand-holding and time, and will most certainly cause setup delays and service issues when things go wrong and you've got the black-sheep server. Unless its absolutely business critical to get that last 10%-20% performance, I would go with whatever they have with no customization. Most likely if you ask for a database setup, they'll give you 6 or 8 drives in raid-5. Most of what these guys do is set up LAMP cookie-cutters... Craig, I share your concerns about outsourced hosting, but as the only custom application involved is one I build my own RPMs for I'm not really concerned about the system getting screwed up software-wise. The idea here is that I might rent an eval system to confirm performance is reasonable, and if it is then I'd be clear to get a bigger stack of them. Luckily there's a guy here who knows a bit about benchmarking for this sort of thing... Yeah, the OP would be much better served ordering a server with an Areca or Escalade / 3ware controller setup and ready to go, shipped to the hosting center and sshing in and doing the rest than letting a hosted solution company try to compete. You can get a nice 16x15K SAS disk machine with an Areca controller, dual QC cpus, and 16 to 32 gig ram for $6000 to $8000 ready to go. We've since repurposed our Dell / PERC machines as file servers and left the real database server work to our aberdeen machines. Trying to wring reasonable performance out of most Dell servers is a testament to frustration. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Problems with autovacuum
2009/5/26 Tom Lane t...@sss.pgh.pa.us: Alvaro Herrera alvhe...@commandprompt.com writes: However I think there's a bigger problem here, which is that if the user has set naptime too low, i.e. to a value lower than number-of-databases * 100ms, we'll be running the (expensive) rebuild_database_list function on each iteration ... maybe we oughta put a lower bound on naptime based on the number of databases to avoid this problem. Bingo, that's surely exactly what was happening to the OP. He had 2000 databases and naptime at (I assume) the default; so he was rerunning rebuild_database_list every 100ms. So that recovery code path needs some more thought. Maybe a lower bound on how often to do rebuild_database_list? And/or don't set adl_next_worker to less than 100ms in the future to begin with? I'd be happy with logging telling me when things are getting pathological. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On 5/26/09 6:52 PM, Scott Marlowe scott.marl...@gmail.com wrote: On Tue, May 26, 2009 at 7:41 PM, Scott Carey sc...@richrelevance.com wrote: On 5/26/09 6:17 PM, Greg Smith gsm...@gregsmith.com wrote: On Tue, 26 May 2009, Joshua D. Drake wrote: CMD doesn't rent hardware you would have to provide that, Rack Space does. Part of the idea was to avoid buying a stack of servers, if this were just a where do I put the boxes at? problem I'd have just asked you about it already. I forgot to check Rack Space earlier, looks like they have Dell servers with up to 8 drives and a RAID controller in them available. Let's just hope it's not one of the completely useless PERC models there; can anyone confirm Dell's PowerEdge R900 has one of the decent performing PERC6 controllers I've heard rumors of in it? Every managed hosting provider I've seen uses RAID controllers and support through the hardware provider. If its Dell its 99% likely a PERC (OEM'd LSI). HP, theirs (not sure who the OEM is), Sun theirs (OEM'd Adaptec). PERC6 in my testing was certainly better than PERC5, but its still sub-par in sequential transfer rate or scaling up past 6 or so drives in a volume. I did go through the process of using a managed hosting provider and getting custom RAID card and storage arrays -- but that takes a lot of hand-holding and time, and will most certainly cause setup delays and service issues when things go wrong and you've got the black-sheep server. Unless its absolutely business critical to get that last 10%-20% performance, I would go with whatever they have with no customization. Most likely if you ask for a database setup, they'll give you 6 or 8 drives in raid-5. Most of what these guys do is set up LAMP cookie-cutters... Craig, I share your concerns about outsourced hosting, but as the only custom application involved is one I build my own RPMs for I'm not really concerned about the system getting screwed up software-wise. The idea here is that I might rent an eval system to confirm performance is reasonable, and if it is then I'd be clear to get a bigger stack of them. Luckily there's a guy here who knows a bit about benchmarking for this sort of thing... Yeah, the OP would be much better served ordering a server with an Areca or Escalade / 3ware controller setup and ready to go, shipped to the hosting center and sshing in and doing the rest than letting a hosted solution company try to compete. You can get a nice 16x15K SAS disk machine with an Areca controller, dual QC cpus, and 16 to 32 gig ram for $6000 to $8000 ready to go. We've since repurposed our Dell / PERC machines as file servers and left the real database server work to our aberdeen machines. Trying to wring reasonable performance out of most Dell servers is a testament to frustration. For a permanent server, yes. But for a sort lease? You have to go with what is easily available for lease, or work out something with a provider where they buy the HW from you and manage/lease it back (some do this, but all I've ever heard of involved 12+ servers to do so and sign on for 1 or 2 years). Expecting full I/O performance out of a DELL with a PERC is not really possible, but maybe that's not as important as a certain pricing model or the flexibility? That is really an independent business decision. I'll also but a caveat to the '3ware' above -- the last few I've used were slower than the PERC (9650 series versus PERC6, 9550 versus PERC5 -- all tests with 12 SATA drives raid 10). I have no experience with the 3ware 9690 series (SAS) though -- those might be just fine. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On 5/26/09 7:27 PM, Scott Carey sc...@richrelevance.com wrote: For a permanent server, yes. But for a sort lease? You have to go with Ahem ... 'short' not 'sort'. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, May 26, 2009 at 8:27 PM, Scott Carey sc...@richrelevance.com wrote: Yeah, the OP would be much better served ordering a server with an Areca or Escalade / 3ware controller setup and ready to go, shipped to the hosting center and sshing in and doing the rest than letting a hosted solution company try to compete. You can get a nice 16x15K SAS disk machine with an Areca controller, dual QC cpus, and 16 to 32 gig ram for $6000 to $8000 ready to go. We've since repurposed our Dell / PERC machines as file servers and left the real database server work to our aberdeen machines. Trying to wring reasonable performance out of most Dell servers is a testament to frustration. For a permanent server, yes. But for a sort lease? You have to go with what is easily available for lease, or work out something with a provider where they buy the HW from you and manage/lease it back (some do this, but all I've ever heard of involved 12+ servers to do so and sign on for 1 or 2 years). True, but given the low cost of a high drive count machine with spares etc you can come away spending a lot less than by leasing. Expecting full I/O performance out of a DELL with a PERC is not really possible, but maybe that's not as important as a certain pricing model or the flexibility? That is really an independent business decision. True. Plus if you only need 4 drives or something, you can do pretty well with a Dell with the RAID controller turned to JBOD and letting the linux kernel do the RAID work. I'll also but a caveat to the '3ware' above -- the last few I've used were slower than the PERC (9650 series versus PERC6, 9550 versus PERC5 -- all tests with 12 SATA drives raid 10). I have no experience with the 3ware 9690 series (SAS) though -- those might be just fine. My experience is primarily with Areca 1100, 1200, and 1600 series controllers, but others on the list have done well with 3ware controllers. We have an 8 port 11xx series areca card at work running RAID-6 as a multipurpose server, and it's really quite fast and well behaved for sequential throughput. But the 16xx series cards stomp the 11xx series in the ground for random IOPS. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, 26 May 2009, Scott Marlowe wrote: Plus if you only need 4 drives or something, you can do pretty well with a Dell with the RAID controller turned to JBOD and letting the linux kernel do the RAID work. I think most of the apps I'm considering would be OK with 4 drives and a useful write cache. The usual hosted configurations are only 1 or 2 and no usable cache, which really limits what you can do with the server before you run into a disk bottleneck. My rule of thumb is that any single core will be satisfied as long as you've got at least 4 disks to feed it, since it's hard for one process to use more than a couple of hundred MB/s for doing mostly sequential work. Obviously random access is much easier to get disk-bound, where you have to throw a lot more disks at it. It wouldn't surprise me to find it's impossible to get an optimal setup of 8+ disks from any hosting provider. Wasn't asking for great DB performance though, just good. -- * Greg Smith gsm...@gregsmith.com http://www.gregsmith.com Baltimore, MD -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
Greg Smith wrote: I keep falling into situations where it would be nice to host a server somewhere else. Virtual host solutions and the mysterious cloud are no good for the ones I run into though, as disk performance is important for all the applications I have to deal with. It's worth noting that some clouds are foggier than others. On Amazon's you can improve your disk performance by setting up software RAID over multiple of their virtual drives. And since they charge by GB, it doesn't cost you any more to do this than to set up a smaller number of larger drives. Here's a blog showing Bonnie++ comparing various RAID levels on Amazon's cloud - with a 4 disk RAID0 giving a nice performance increase over a single virtual drive. http://af-design.com/blog/2009/02/27/amazon-ec2-disk-performance/ Here's a guy who set up a 40TB RAID0 with 40 1TB virtual disks on Amazon. http://groups.google.com/group/ec2ubuntu/web/raid-0-on-ec2-ebs-volumes-elastic-block-store-using-mdadm http://groups.google.com/group/ec2ubuntu/browse_thread/thread/d520ae145edf746 I might get around to trying some pgbench runs on amazon in a week or so. Any suggestions what would be most interesting? What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). Is anyone aware of a company that offers such a thing? -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
On Tue, May 26, 2009 at 6:28 PM, Craig James craig_ja...@emolecules.com wrote: Greg Smith wrote: What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). We finally bought some nice Dell servers and found a co-location site that provides us all the infrastructure (reliable power, internet, cooling, security...), and we're in charge of the computers. We've never looked back. I ran this way on a Quad-processor Dell for many years, and then, after selling the business and starting a new one, decided to keep my DB on a remote-hosted machine. I have a dual-core2 with hardware RAID 5 (I know, I know) and a private network interface to the other servers (web, email, web-cache) Just today when the DB server went down (after 2 years of reliable service and 380 days of uptime) they gave me remote KVM access to the machine. Turns out I had messed up the fstab while fiddling with the server because I really don't know FreeBSD as well as Linux, I think remote leased-hosting works fine as long as you have a competent team on the other end and KVM over IP access. Many providers don't have that... and without it you can get stuck as you describe. I have used MANY providers over they years, at the peak with over 30 leased servers at 12 providers, and with many colocation situations as well. The only advantage with colocation I have seen is the reduced expense if you keep it going for a few years on the same box. which is a big advantage if it lets you buy a much more powerful box to begin with. Providers I prefer for high-end machines allow me to upgrade the hardware with no monthly fees (marked-up cost of upgrade + time/labor only) that keeps the cost down. -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
Re: [PERFORM] Hosted servers with good DB disk performance?
Greg Smith wrote: What I'd love to have is a way to rent a fairly serious piece of dedicated hardware, ideally with multiple (at least 4) hard drives in a RAID configuration and a battery-backed write cache. The cache is negotiable. Linux would be preferred, FreeBSD or Solaris would also work; not Windows though (see good DB performance). Is anyone aware of a company that offers such a thing? I've used http://softlayer.com/ in the past and highly recommend them. They sell a wide range of dedicated servers, including ones that handle up to 12 HDDs/SSDs, and servers with battery-backed RAID controllers (I've been told they use mostly Adaptec cards as well as some 3ware cards). In addition, all their servers are connected to a private network you can VPN into, and all include IPMI for remote management when you can't SSH into your server. They have a host of other features; click on the Services tab on their site to find out more. Alex -- Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
