Hi, I've been thinking about moving our outbound mail relay queue and deploying a NetApp. However I came across solid state disks.
I found that in terms of SCSI based solid state drives Curtis makes the best and most affordable stuff in the 1G-8GB size. It has SCA interfaces and dimmesions of a SCSI disk that will fit into a hot swap bay. These start at around US$2000 and seem to max out at 10,000 transactions per second.
Given that our mail queue back log does not grow above 100MB I am thinking of getting Cenetek's Rocket Drive with 1GB, this is a PCI board that has its own external power and hooks to your UPS.
A magazine benchmarked them to 30,000 transactions per second at US$1000 it seems like a bargain to me. Centaket rate them at 100K transactions per second. In its price range feature performance it seems to have a very clearly defined niche. It also beats ram drives as it does not require the machines CPU to do the IO.
Anyway I just want to know if any of you guys are using these to boost your servers and have any opinions or advice.
Regards, Maren.
-------------------------------------------------------------- HKdotCOM Ltd AIM: MarenHKdotCOM Tel: 852 2865-4865 ext 888 leizaola@hk.com ICQ: 39905706 Fax: 852 2865-4100 -------------------------------------------------------------- Join our liquidations list : http://www.lists.hk.com/lists/listinfo/liquidations
Maren S. Leizaola wrote:
It also beats ram drives as it does not require the machines CPU to do the IO.
How does the data get on and off the board then?
SSDs are most frequently used for accelerating databases, especially OLTP, which has some similarities to what MTAs do, so it's worth a shot. My concern would be those rare occasions where your queue was inflated beyond the capacity of such a device - would your MTA software cope gracefully?
Given the unpredictable torrents of spam and viruses which can clog and swell mail queues, making sure that the queue ccould grow without bursting, and, perhaps, be cleaned in situ, would seem to me to be as important as making sure it can be processed as quickly as possible in the steady-state.
On Thu, 6 May 2004, Mark Simmons wrote:
How does the data get on and off the board then?
PCI using its own chipset.
SSDs are most frequently used for accelerating databases, especially OLTP, which has some similarities to what MTAs do, so it's worth a shot. My concern would be those rare occasions where your queue was inflated beyond the capacity of such a device - would your MTA software cope gracefully?
If you are just thinking of the MTA it would give the other mail relays a temporary error and they would try. The ones that are not likely to retry (one hopes) are the spammers.
If you have backlog, so what you have a blazing fast disk I/O power. Clearing the queue would be fast as long as the destinations that you are relaying to are up and advailable.
Your concerns are good and initially I made many assumptions as to what could cause the bog down, however the cases of having a 1GB mail backlog is not common at least for us, normally our back log is 40MB.
Given the unpredictable torrents of spam and viruses which can clog and swell mail queues, making sure that the queue ccould grow without bursting, and, perhaps, be cleaned in situ, would seem to me to be as important as making sure it can be processed as quickly as possible in the steady-state.
You could have script that runs every few minutes to check if it is full if it is then it moves the queue to a side and clears the whole drive and then notifies you and firesup the MTA again. Once you deal with what ever attack you move the archived queues back in and blast them out..
Anyway if you are growing to 1GB backlogs then going to 4GB is an option. I don't think many people get themselves in the situation of having such queue sizes often...
I run the www.TPC.INT Fax cell for HK and that gets abused to a pretty serious level. This machine truns smtp traffic into faxes which does use out a bit of CPU/IO power. I've had queues of about 40,000 messages (faxes, TIFF, txt PDF) in a queue directory and that queue did not reach 1GB. Every day someone spams out with an address fred@8522111145.iddd.tpc.int and about 40,000-60,000 bounces per day come back each one has to be read written and parsed a few times using perl. Then it also gets normal email, spam, fax broadcasts attempts etc. IMHO Even a spamming company blasting out using your server as the return is not capable of clogging up 1GB on an highly overloaded machine. 4GB sounds like a solid plan for the paranoid. ;-)
Regards, Maren.
-------------------------------------------------------------- HKdotCOM Ltd AIM: MarenHKdotCOM Tel: 852 2865-4865 ext 888 leizaola@hk.com ICQ: 39905706 Fax: 852 2865-4100 -------------------------------------------------------------- Join our liquidations list : http://www.lists.hk.com/lists/listinfo/liquidations
Coincidentally this appeared today:
http://slashdot.org/comments.pl?sid=106683&cid=9078175
Maren S. Leizaola wrote:
On Thu, 6 May 2004, Mark Simmons wrote:
<snip>
-
Maren S. Leizaola -
Mark Simmons